+2014-10-29 Alexander Ivchenko <alexander.ivchenko@intel.com>
+ Maxim Kuznetsov <maxim.kuznetsov@intel.com>
+ Anna Tikhonova <anna.tikhonova@intel.com>
+ Ilya Tocar <ilya.tocar@intel.com>
+ Andrey Turetskiy <andrey.turetskiy@intel.com>
+ Ilya Verbin <ilya.verbin@intel.com>
+ Kirill Yukhin <kirill.yukhin@intel.com>
+ Michael Zolotukhin <michael.v.zolotukhin@intel.com>
+
+ * gcc.target/i386/avx512bw-check.h: New.
+ * gcc.target/i386/avx512bw-kunpckdq-1.c: Ditto.
+ * gcc.target/i386/avx512bw-kunpckwd-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vdbpsadbw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vdbpsadbw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vmovdqu16-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vmovdqu16-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vmovdqu8-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vmovdqu8-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpabsb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpabsb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpabsw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpabsw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpackssdw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpackssdw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpacksswb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpacksswb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpackusdw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpackusdw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpackuswb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpackuswb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddsb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddsb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddsw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddsw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddusb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddusb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddusw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddusw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpaddw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpalignr-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpalignr-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpavgb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpavgb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpavgw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpavgw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpblendmb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpblendmb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpblendmw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpblendmw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpbroadcastb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpbroadcastb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpbroadcastw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpbroadcastw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpeqb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpeqb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpeqw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpeqw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpgtb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpgtb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpgtw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpgtw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpub-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpub-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpuw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpuw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpcmpw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpermi2w-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpermi2w-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpermt2w-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpermt2w-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpermw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpermw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaddubsw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaddubsw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaddwd-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaddwd-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaxsb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaxsb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaxsw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaxsw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaxub-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaxub-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaxuw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmaxuw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpminsb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpminsb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpminsw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpminsw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpminub-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpminub-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpminuw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpminuw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovb2m-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovb2m-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovm2b-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovm2b-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovm2w-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovm2w-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovswb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovswb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovsxbw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovsxbw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovuswb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovuswb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovw2m-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovw2m-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovwb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovwb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovzxbw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmovzxbw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmulhrsw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmulhrsw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmulhuw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmulhuw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmulhw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmulhw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmullw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpmullw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpshufb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpshufb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpshufhw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpshufhw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpshuflw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpshuflw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpslldq-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsllvw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsllvw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsllw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsllw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsllwi-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsllwi-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsravw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsravw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsraw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsraw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrawi-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrawi-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrldq-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrlvw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrlvw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrlw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrlw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrlwi-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsrlwi-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubsb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubsb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubsw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubsw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubusb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubusb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubusw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubusw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpsubw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vptestmb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vptestmb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vptestmw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vptestmw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vptestnmb-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vptestnmb-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vptestnmw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vptestnmw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpunpckhbw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpunpckhbw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpunpckhwd-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpunpckhwd-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpunpcklbw-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpunpcklbw-2.c: Ditto.
+ * gcc.target/i386/avx512bw-vpunpcklwd-1.c: Ditto.
+ * gcc.target/i386/avx512bw-vpunpcklwd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-check.h: Ditto.
+ * gcc.target/i386/avx512dq-vandnpd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vandnpd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vandnps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vandnps-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vandpd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vandpd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vandps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vandps-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcastf32x2-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcastf32x2-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcastf32x8-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcastf32x8-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcastf64x2-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcastf64x2-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcasti32x2-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcasti32x2-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcasti32x8-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcasti32x8-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcasti64x2-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vbroadcasti64x2-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtpd2qq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtpd2qq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtpd2uqq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtpd2uqq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtps2qq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtps2qq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtps2uqq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtps2uqq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtqq2pd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtqq2pd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtqq2ps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtqq2ps-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvttpd2qq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvttpd2qq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvttpd2uqq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvttpd2uqq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvttps2qq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvttps2qq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvttps2uqq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvttps2uqq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtuqq2pd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtuqq2pd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtuqq2ps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vcvtuqq2ps-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vextractf32x8-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vextractf64x2-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vextractf64x2-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vextracti32x8-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vextracti64x2-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vextracti64x2-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vfpclasspd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vfpclasspd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vfpclassps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vfpclassps-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vfpclasssd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vfpclassss-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vinsertf32x8-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vinsertf64x2-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vinsertf64x2-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vinserti32x8-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vinserti64x2-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vinserti64x2-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vorpd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vorpd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vorps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vorps-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmovd2m-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmovd2m-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmovm2d-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmovm2d-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmovm2q-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmovm2q-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmovq2m-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmovq2m-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmullq-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vpmullq-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vrangepd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vrangepd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vrangeps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vrangeps-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vrangesd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vrangess-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vreducepd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vreducepd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vreduceps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vreduceps-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vreducesd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vreducess-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vxorpd-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vxorpd-2.c: Ditto.
+ * gcc.target/i386/avx512dq-vxorps-1.c: Ditto.
+ * gcc.target/i386/avx512dq-vxorps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-check.h: Ditto.
+ * gcc.target/i386/avx512vl-gather-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i32gatherd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i32gatherpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i32gatherps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i32gatherq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i32scatterd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i32scatterpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i32scatterps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i32scatterq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i64gatherd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i64gatherpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i64gatherps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i64gatherq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i64scatterd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i64scatterpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i64scatterps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-i64scatterq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vaddpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vaddpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vaddps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vaddps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-valignd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-valignd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-valignq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-valignq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vandnpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vandnps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vandpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vandps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vblendmpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vblendmpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vblendmps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vblendmps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcastf32x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcastf32x4-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcastf32x4-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcastf64x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcasti32x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcasti32x4-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcasti32x4-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcasti64x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcastsd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcastsd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcastss-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vbroadcastss-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcmppd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcmppd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcmpps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcmpps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcompresspd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcompresspd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcompressps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcompressps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtdq2pd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtdq2pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtdq2ps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtdq2ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtpd2dq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtpd2dq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtpd2ps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtpd2ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtpd2qq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtpd2udq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtpd2udq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtpd2uqq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtph2ps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtph2ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2dq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2dq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2pd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2ph-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2ph-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2qq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2udq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2udq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtps2uqq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtqq2pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtqq2ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttpd2dq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttpd2dq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttpd2qq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttpd2udq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttpd2udq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttpd2uqq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttps2dq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttps2dq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttps2qq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttps2udq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttps2udq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvttps2uqq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtudq2pd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtudq2pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtudq2ps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtudq2ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtuqq2pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vcvtuqq2ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vdbpsadbw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vdivpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vdivpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vdivps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vdivps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vexpandpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vexpandpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vexpandps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vexpandps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vextractf32x4-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vextractf32x4-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vextractf64x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vextracti32x4-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vextracti32x4-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vextracti64x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfixupimmpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfixupimmpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfixupimmps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfixupimmps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmaddXXXpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmaddXXXpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmaddXXXps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmaddXXXps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmaddsubXXXpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmaddsubXXXpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmaddsubXXXps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmaddsubXXXps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmsubXXXpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmsubXXXpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmsubXXXps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmsubXXXps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmsubaddXXXpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmsubaddXXXpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmsubaddXXXps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfmsubaddXXXps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfnmaddXXXpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfnmaddXXXpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfnmaddXXXps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfnmaddXXXps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfnmsubXXXpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfnmsubXXXpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfnmsubXXXps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vfnmsubXXXps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfpclasspd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vfpclassps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vgetexppd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vgetexppd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vgetexpps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vgetexpps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vgetmantpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vgetmantpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vgetmantps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vgetmantps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vinsertf32x4-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vinsertf32x4-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vinsertf64x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vinserti32x4-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vinserti32x4-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vinserti64x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmaxpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmaxpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmaxps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmaxps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vminpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vminpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vminps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vminps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovapd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovapd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovaps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovaps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovddup-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovddup-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqa32-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqa32-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqa64-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqa64-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqu16-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqu32-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqu32-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqu64-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqu64-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovdqu8-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovntdqa-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovshdup-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovshdup-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovsldup-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovsldup-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovupd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovupd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovups-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmovups-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmulpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmulpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vmulps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vmulps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vorpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vorps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpabsb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpabsd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpabsd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpabsq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpabsq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpabsw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpackssdw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpacksswb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpackusdw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpackuswb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddsb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddsw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddusb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddusw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpaddw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpalignr-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpandd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpandd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpandnd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpandnd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpandnq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpandnq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpandq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpandq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpavgb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpavgw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpblendmb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpblendmd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpblendmd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpblendmq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpblendmq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpblendmw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpbroadcastb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpbroadcastd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpbroadcastd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpbroadcastmb2q-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpbroadcastmw2d-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpbroadcastq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpbroadcastq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpbroadcastw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpeqb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpeqd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpeqd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpeqq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpeqq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpeqw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpged-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgeq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgeud-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgeuq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgtb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgtd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgtd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgtq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgtq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpgtw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpled-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpleq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpleud-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpleuq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpltd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpltq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpltud-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpltuq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpneqd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpneqq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpnequd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpnequq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpub-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpud-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpud-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpuq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpuq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpuw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcmpw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcompressd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcompressd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcompressq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpcompressq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpconflictd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpconflictq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2d-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2d-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2pd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2ps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2q-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2q-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermi2w-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermilpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermilpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermilpdi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermilpdi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermilps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermilps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermilpsi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermilpsi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermpdi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermpdi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermq-imm-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermq-imm-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermq-var-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermq-var-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2d-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2d-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2pd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2ps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2q-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2q-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermt2w-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpermw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpexpandd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpexpandd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpexpandq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpexpandq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vplzcntd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vplzcntq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaddubsw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaddwd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxsb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxsd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxsd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxsq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxsq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxsw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxub-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxud-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxud-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxuq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxuq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmaxuw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminsb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminsd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminsd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminsq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminsq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminsw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminub-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminud-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminud-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminuq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminuq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpminuw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovb2m-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovd2m-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovdb-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovdb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovdw-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovdw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovm2b-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovm2d-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovm2q-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovm2w-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovq2m-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovqb-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovqb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovqd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovqd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovqw-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovqw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsdb-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsdb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsdw-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsdw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsqb-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsqb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsqd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsqd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsqw-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsqw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovswb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxbd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxbd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxbq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxbq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxbw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxdq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxdq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxwd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxwd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxwq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovsxwq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusdb-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusdb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusdw-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusdw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusqb-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusqb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusqd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusqd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusqw-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovusqw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovuswb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovw2m-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovwb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxbd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxbd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxbq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxbq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxbw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxdq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxdq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxwd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxwd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxwq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmovzxwq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmuldq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmuldq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmulhrsw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmulhuw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmulhw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmulld-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmulld-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmullq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmullw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmuludq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpmuludq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpord-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpord-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vporq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vporq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vprold-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vprold-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vprolq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vprolq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vprolvd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vprolvd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vprolvq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vprolvq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vprord-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vprord-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vprorq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vprorq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vprorvd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vprorvd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vprorvq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vprorvq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsadbw-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpshufb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpshufd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpshufd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpshufhw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpshuflw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpslld-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpslld-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpslldi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpslldi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllqi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllqi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllvd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllvd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllvq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllvq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllvw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsllwi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrad-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrad-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsradi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsradi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsraq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsraq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsraqi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsraqi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsravd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsravd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsravq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsravq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsravw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsraw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrawi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrld-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrld-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrldi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrldi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlqi-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlqi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlvd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlvd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlvq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlvq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlvw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsrlwi-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubsb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubsw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubusb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubusw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpsubw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpternlogd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpternlogd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpternlogq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpternlogq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestmb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestmd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestmd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestmq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestmq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestmw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestnmb-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestnmd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestnmd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestnmq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestnmq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vptestnmw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpckhbw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpckhdq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpckhdq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpckhqdq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpckhqdq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpckhwd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpcklbw-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpckldq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpckldq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpcklqdq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpcklqdq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpunpcklwd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpxord-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpxord-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vpxorq-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vpxorq-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vrangepd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vrangeps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vrcp14pd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vrcp14pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vrcp14ps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vrcp14ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vreducepd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vreduceps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vrndscalepd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vrndscalepd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vrndscaleps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vrndscaleps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vrsqrt14pd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vrsqrt14pd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vrsqrt14ps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vrsqrt14ps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vscalefpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vscalefpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vscalefps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vscalefps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vshuff32x4-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vshuff32x4-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vshuff64x2-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vshuff64x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vshufi32x4-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vshufi32x4-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vshufi64x2-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vshufi64x2-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vshufpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vshufpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vshufps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vshufps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vsqrtpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vsqrtpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vsqrtps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vsqrtps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vsubpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vsubpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vsubps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vsubps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vunpckhpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vunpckhpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vunpckhps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vunpckhps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vunpcklpd-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vunpcklpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vunpcklps-1.c: Ditto.
+ * gcc.target/i386/avx512vl-vunpcklps-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vxorpd-2.c: Ditto.
+ * gcc.target/i386/avx512vl-vxorps-2.c: Ditto.
+ * gcc.target/i386/avx512f-helper.h: Make avx512vl friendly.
+ * gcc.target/i386/avx512f-mask-type.h: Ditto.
+ * gcc.target/i386/avx512f-vcmppd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vcmpps-2.c: Ditto.
+ * gcc.target/i386/avx512f-vcvtpd2ps-2.c: Ditto.
+ * gcc.target/i386/avx512f-vcvtpd2udq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vcvtph2ps-2.c: Ditto.
+ * gcc.target/i386/avx512f-vcvtps2ph-2.c: Ditto.
+ * gcc.target/i386/avx512f-vcvttpd2udq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpandd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpandnd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpandnq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpandq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpbroadcastq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpud-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpcmpuq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpermd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpermilpdi-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpermq-imm-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpermq-var-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovdb-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovdw-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovqb-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovqd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovqw-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovsdb-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovsdw-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovsqb-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovsqd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovsqw-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovusdb-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovusdw-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovusqb-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovusqd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpmovusqw-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpord-2.c: Ditto.
+ * gcc.target/i386/avx512f-vporq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpxord-2.c: Ditto.
+ * gcc.target/i386/avx512f-vpxorq-2.c: Ditto.
+ * gcc.target/i386/avx512f-vrndscalepd-2.c: Ditto.
+ * gcc.target/i386/avx512f-vrndscaleps-2.c: Ditto.
+ * gcc.target/i386/avx512f-vshuff32x4-2.c: Ditto.
+ * gcc.target/i386/avx512f-vshuff64x2-2.c: Ditto.
+ * gcc.target/i386/avx512f-vshufi32x4-2.c: Ditto.
+ * gcc.target/i386/avx512f-vshufi64x2-2.c: Ditto.
+ * gcc.target/i386/avx512f-vshufpd-2.c: Ditto.
+ * gcc.target/i386/i386.exp: Ditto.
+ * gcc.target/i386/m512-check.h: Ditto.
+
+2014-10-29 Alexander Ivchenko <alexander.ivchenko@intel.com>
+ Maxim Kuznetsov <maxim.kuznetsov@intel.com>
+ Anna Tikhonova <anna.tikhonova@intel.com>
+ Ilya Tocar <ilya.tocar@intel.com>
+ Andrey Turetskiy <andrey.turetskiy@intel.com>
+ Ilya Verbin <ilya.verbin@intel.com>
+ Kirill Yukhin <kirill.yukhin@intel.com>
+ Michael Zolotukhin <michael.v.zolotukhin@intel.com>
+
+ * g++.dg/other/i386-2.C: Add new options.
+ * g++.dg/other/i386-3.C: Ditto.
+ * gcc.target/i386/sse-12.c: Ditto.
+ * gcc.target/i386/sse-14.c: Ditto.
+ * gcc.target/i386/sse-22.c: Ditto.
+ * gcc.target/i386/sse-13.c: Add new options and builtins.
+ * gcc.target/i386/sse-23.c: Ditto.
+ * gcc.target/i386/avx-1.c: Add new builtins.
+
2014-10-28 Dominik Vogt <vogt@linux.vnet.ibm.com>
* gcc.misc-tests/godump.exp: New.
--- /dev/null
+#include <stdlib.h>
+#include "cpuid.h"
+#include "m512-check.h"
+#include "avx512f-os-support.h"
+
+static void avx512bw_test (void);
+
+static void __attribute__ ((noinline)) do_test (void)
+{
+ avx512bw_test ();
+}
+
+int
+main ()
+{
+ unsigned int eax, ebx, ecx, edx;
+
+ if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx))
+ return 0;
+
+ /* Run AVX512BW test only if host has AVX512BW support. */
+ if ((ecx & bit_OSXSAVE) == (bit_OSXSAVE))
+ {
+ if (__get_cpuid_max (0, NULL) < 7)
+ return 0;
+
+ __cpuid_count (7, 0, eax, ebx, ecx, edx);
+
+ if ((avx512f_os_support ()) && ((ebx & bit_AVX512BW) == bit_AVX512BW))
+ {
+ do_test ();
+#ifdef DEBUG
+ printf ("PASSED\n");
+#endif
+ return 0;
+ }
+#ifdef DEBUG
+ printf ("SKIPPED\n");
+#endif
+ }
+#ifdef DEBUG
+ else
+ printf ("SKIPPED\n");
+#endif
+
+ return 0;
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -O2" } */
+/* { dg-final { scan-assembler-times "kunpckdq\[ \\t\]+\[^\n\]*%k\[1-7\]" 1 } } */
+
+#include <immintrin.h>
+
+void
+avx512bw_test () {
+ __mmask64 k1, k2, k3;
+ volatile __m512i x;
+
+ __asm__( "kmovq %1, %0" : "=k" (k1) : "r" (1) );
+ __asm__( "kmovq %1, %0" : "=k" (k2) : "r" (2) );
+
+ k3 = _mm512_kunpackd (k1, k2);
+ x = _mm512_mask_avg_epu8 (x, k3, x, x);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -O2" } */
+/* { dg-final { scan-assembler-times "kunpckwd\[ \\t\]+\[^\n\]*%k\[1-7\]" 1 } } */
+
+#include <immintrin.h>
+
+void
+avx512bw_test () {
+ volatile __mmask32 k1, k2, k3;
+ volatile __m256i x;
+
+ __asm__( "kmovd %1, %0" : "=k" (k1) : "r" (1) );
+ __asm__( "kmovd %1, %0" : "=k" (k2) : "r" (2) );
+
+ k3 = _mm512_kunpackw (k1, k2);
+ //x = _mm256_mask_avg_epu8 (x, k3, x, x);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdbpsadbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x, a;
+volatile __m256i y, b;
+volatile __m128i z, c;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_dbsad_epu8 (a, a, 0xaa);
+ x = _mm512_mask_dbsad_epu8 (x, m1, a, a, 0xaa);
+ x = _mm512_maskz_dbsad_epu8 (m1, a, a, 0xaa);
+ y = _mm256_dbsad_epu8 (b, b, 0xbb);
+ y = _mm256_mask_dbsad_epu8 (y, m2, b, b, 0xbb);
+ y = _mm256_maskz_dbsad_epu8 (m2, b, b, 0xbb);
+ z = _mm_dbsad_epu8 (c, c, 0xcc);
+ z = _mm_mask_dbsad_epu8 (z, m3, c, c, 0xcc);
+ z = _mm_maskz_dbsad_epu8 (m3, c, c, 0xcc);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned short *dst, unsigned char *src1, unsigned char *src2,
+ int imm)
+{
+ int i, j, k, part, power;
+ unsigned char tmp[2 * SIZE];;
+
+ for (i = 0; i < 2 * SIZE; i += 16)
+ {
+ for (j = 0; j < 4; j++)
+ {
+ power = 1;
+ for (k = 0; k < j; k++)
+ power *= 4;
+ part = (imm & (3 * power)) >> (2 * j);
+ for (k = 0; k < 4; k++)
+ tmp[i + 4 * j + k] = src2[i + 4 * part + k];
+ }
+ }
+
+ for (i = 0; i < SIZE; i += 4)
+ {
+ dst[i] = dst[i + 1] = dst[i + 2] = dst[i + 3] = 0;
+ for (j = 0; j < 4; j++)
+ {
+ dst[i] += abs (src1[2 * i + j] - tmp[2 * i + j]);
+ dst[i + 1] += abs (src1[2 * i + j] - tmp[2 * i + j + 1]);
+ dst[i + 2] += abs (src1[2 * i + j + 4] - tmp[2 * i + j + 2]);
+ dst[i + 3] += abs (src1[2 * i + j + 4] - tmp[2 * i + j + 3]);
+ }
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE];
+ int imm = 0x22;
+
+ sign = -1;
+ for (i = 0; i < 2*SIZE; i++)
+ {
+ src1.a[i] = 1 + 34 * i * sign;
+ src1.a[i] = 179 - i;
+ sign = sign * -1;
+ }
+
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_dbsad_epu8) (src1.x, src2.x, imm);
+ res2.x = INTRINSIC (_mask_dbsad_epu8) (res2.x, mask, src1.x, src2.x, imm);
+ res3.x = INTRINSIC (_maskz_dbsad_epu8) (mask, src1.x, src2.x, imm);
+
+ CALC (res_ref, src1.a, src2.a, imm);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*\\)\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*\\)\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu16\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+short *p;
+volatile __m512i x1, yy;
+volatile __m256i x2, y2;
+volatile __m128i x3, y3;
+volatile __mmask32 m32;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512bw_test (void)
+{
+ x1 = _mm512_mask_mov_epi16 (x1, m32, yy);
+ x2 = _mm256_mask_mov_epi16 (x2, m16, y2);
+ x3 = _mm_mask_mov_epi16 (x3, m8, y3);
+
+ x1 = _mm512_maskz_mov_epi16 (m32, yy);
+ x2 = _mm256_maskz_mov_epi16 (m16, y2);
+ x3 = _mm_maskz_mov_epi16 (m8, y3);
+
+ x1 = _mm512_mask_loadu_epi16 (x1, m32, p);
+ x2 = _mm256_mask_loadu_epi16 (x2, m16, p);
+ x3 = _mm_mask_loadu_epi16 (x3, m8, p);
+
+ x1 = _mm512_maskz_loadu_epi16 (m32, p);
+ x2 = _mm256_maskz_loadu_epi16 (m16, p);
+ x3 = _mm_maskz_loadu_epi16 (m8, p);
+
+ _mm512_mask_storeu_epi16 (p, m32, x1);
+ _mm256_mask_storeu_epi16 (p, m16, x2);
+ _mm_mask_storeu_epi16 (p, m8, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE ((AVX512F_LEN) / 16)
+#include "avx512f-mask-type.h"
+
+typedef struct
+{
+ char c;
+ short a[SIZE];
+} __attribute__ ((packed)) EVAL(unaligned_array, AVX512F_LEN,);
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s3, res1, res2, res3, res4;
+ EVAL(unaligned_array, AVX512F_LEN,) s2, res5;
+ MASK_TYPE mask = MASK_VALUE;
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = 123 * i * sign;
+ s2.a[i] = 456 * i * sign;
+ s3.a[i] = 789 * i * sign;
+ res1.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ res5.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_mask_mov_epi16) (res1.x, mask, s1.x);
+ res2.x = INTRINSIC (_maskz_mov_epi16) (mask, s1.x);
+ res3.x = INTRINSIC (_mask_loadu_epi16) (res3.x, mask, s2.a);
+ res4.x = INTRINSIC (_maskz_loadu_epi16) (mask, s2.a);
+ INTRINSIC (_mask_storeu_epi16) (res5.a, mask, s3.x);
+
+ MASK_MERGE (i_w) (s1.a, mask, SIZE);
+ if (checkVs (res1.a, s1.a, SIZE))
+ abort ();
+
+ MASK_ZERO (i_w) (s1.a, mask, SIZE);
+ if (checkVs (res2.a, s1.a, SIZE))
+ abort ();
+
+ MASK_MERGE (i_w) (s2.a, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, s2.a))
+ abort ();
+
+ MASK_ZERO (i_w) (s2.a, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res4, s2.a))
+ abort ();
+
+ MASK_MERGE (i_w) (s3.a, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (s3, res5.a))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*\\)\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*\\)\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu8\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+char *p;
+volatile __m512i x1, yy;
+volatile __m256i x2, y2;
+volatile __m128i x3, y3;
+volatile __mmask64 m64;
+volatile __mmask32 m32;
+volatile __mmask16 m16;
+
+void extern
+avx512bw_test (void)
+{
+ x1 = _mm512_mask_mov_epi8 (x1, m64, yy);
+ x2 = _mm256_mask_mov_epi8 (x2, m32, y2);
+ x3 = _mm_mask_mov_epi8 (x3, m16, y3);
+
+ x1 = _mm512_maskz_mov_epi8 (m64, yy);
+ x2 = _mm256_maskz_mov_epi8 (m32, y2);
+ x3 = _mm_maskz_mov_epi8 (m16, y3);
+
+ x1 = _mm512_mask_loadu_epi8 (x1, m64, p);
+ x2 = _mm256_mask_loadu_epi8 (x2, m32, p);
+ x3 = _mm_mask_loadu_epi8 (x3, m16, p);
+
+ x1 = _mm512_maskz_loadu_epi8 (m64, p);
+ x2 = _mm256_maskz_loadu_epi8 (m32, p);
+ x3 = _mm_maskz_loadu_epi8 (m16, p);
+
+ _mm512_mask_storeu_epi8 (p, m64, x1);
+ _mm256_mask_storeu_epi8 (p, m32, x2);
+ _mm_mask_storeu_epi8 (p, m16, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE ((AVX512F_LEN) / 8)
+#include "avx512f-mask-type.h"
+
+typedef struct
+{
+ char c;
+ char a[SIZE];
+} __attribute__ ((packed)) EVAL(unaligned_array, AVX512F_LEN,);
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_b) s1, s3, res1, res2, res3, res4;
+ EVAL(unaligned_array, AVX512F_LEN,) s2, res5;
+ MASK_TYPE mask = MASK_VALUE;
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = (i + 1) * sign;
+ s2.a[i] = (i + 2) * sign;
+ s3.a[i] = (i * 2) * sign;
+ res1.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ res5.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_mask_mov_epi8) (res1.x, mask, s1.x);
+ res2.x = INTRINSIC (_maskz_mov_epi8) (mask, s1.x);
+ res3.x = INTRINSIC (_mask_loadu_epi8) (res3.x, mask, s2.a);
+ res4.x = INTRINSIC (_maskz_loadu_epi8) (mask, s2.a);
+ INTRINSIC (_mask_storeu_epi8) (res5.a, mask, s3.x);
+
+ MASK_MERGE (i_b) (s1.a, mask, SIZE);
+ if (checkVc (res1.a, s1.a, SIZE))
+ abort ();
+
+ MASK_ZERO (i_b) (s1.a, mask, SIZE);
+ if (checkVc (res2.a, s1.a, SIZE))
+ abort ();
+
+ MASK_MERGE (i_b) (s2.a, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, s2.a))
+ abort ();
+
+ MASK_ZERO (i_b) (s2.a, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res4, s2.a))
+ abort ();
+
+ MASK_MERGE (i_b) (s3.a, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (s3, res5.a))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask64 m1;
+volatile __mmask32 m2;
+volatile __mmask16 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_abs_epi8 (z);
+ z = _mm512_mask_abs_epi8 (z, m1, z);
+ z = _mm512_maskz_abs_epi8 (m1, z);
+ y = _mm256_mask_abs_epi8 (y, m2, y);
+ y = _mm256_maskz_abs_epi8 (m2, y);
+ x = _mm_mask_abs_epi8 (x, m3, x);
+ x = _mm_maskz_abs_epi8 (m3, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *s, char *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ if (s[i] < 0)
+ r[i] = -s[i];
+ else
+ r[i] = s[i];
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_b) s, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = i * 7 + (i << 15) + 356;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ CALC (s.a, res_ref);
+
+ res1.x = INTRINSIC (_abs_epi8) (s.x);
+ res2.x = INTRINSIC (_mask_abs_epi8) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_abs_epi8) (mask, s.x);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_abs_epi16 (z);
+ z = _mm512_mask_abs_epi16 (z, m1, z);
+ z = _mm512_maskz_abs_epi16 (m1, z);
+ y = _mm256_mask_abs_epi16 (y, m2, y);
+ y = _mm256_maskz_abs_epi16 (m2, y);
+ x = _mm_mask_abs_epi16 (x, m3, x);
+ x = _mm_maskz_abs_epi16 (m3, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *s, short *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ if (s[i] < 0)
+ r[i] = -s[i];
+ else
+ r[i] = s[i];
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = i * 7 + (i << 15) + 356;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ CALC (s.a, res_ref);
+
+ res1.x = INTRINSIC (_abs_epi16) (s.x);
+ res2.x = INTRINSIC (_mask_abs_epi16) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_abs_epi16) (mask, s.x);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpackssdw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpackssdw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackssdw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpackssdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackssdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackssdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpackssdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask64 mx;
+volatile __mmask32 my;
+volatile __mmask16 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_packs_epi32 (x, x);
+ x = _mm512_mask_packs_epi32 (x, mx, x, x);
+ x = _mm512_maskz_packs_epi32 (mx, x, x);
+ y = _mm256_mask_packs_epi32 (y, my, y, y);
+ y = _mm256_maskz_packs_epi32 (my, y, y);
+ z = _mm_mask_packs_epi32 (z, mz, z, z);
+ z = _mm_maskz_packs_epi32 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define DST_SIZE (AVX512F_LEN / 16)
+#define SRC_SIZE (AVX512F_LEN / 32)
+
+#include "limits.h"
+
+#include "avx512f-mask-type.h"
+
+static short
+EVAL(int_to_short, AVX512F_LEN,) (int iVal)
+{
+ short sVal;
+
+ if (iVal < SHRT_MIN)
+ sVal = SHRT_MIN;
+ else if (iVal > SHRT_MAX)
+ sVal = SHRT_MAX;
+ else
+ sVal = iVal;
+
+ return sVal;
+}
+
+void
+CALC (int *src1, int *src2, short *dst)
+{
+ int i;
+ int *ptr;
+
+ for (i = 0; i < DST_SIZE; i++)
+ {
+ ptr = (i / 4) % 2 ? src2 : src1;
+ dst[i] = EVAL(int_to_short, AVX512F_LEN,) (ptr[i % 4 + (i / 8) * 4]);
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_d) s1, s2;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short dst_ref[DST_SIZE];
+ int i;
+
+ for (i = 0; i < DST_SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ for (i = 0; i < SRC_SIZE; i++)
+ {
+ s1.a[i] = i + 10;
+ s2.a[i] = i + 15;
+ }
+
+ res1.x = INTRINSIC (_packs_epi32) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_packs_epi32) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_packs_epi32) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (dst_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (dst_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, dst_ref))
+ abort ();
+
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpacksswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpacksswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpacksswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpacksswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpacksswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpacksswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpacksswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask64 mx;
+volatile __mmask32 my;
+volatile __mmask16 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_packs_epi16 (x, x);
+ x = _mm512_mask_packs_epi16 (x, mx, x, x);
+ x = _mm512_maskz_packs_epi16 (mx, x, x);
+ y = _mm256_mask_packs_epi16 (y, my, y, y);
+ y = _mm256_maskz_packs_epi16 (my, y, y);
+ z = _mm_mask_packs_epi16 (z, mz, z, z);
+ z = _mm_maskz_packs_epi16 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define DST_SIZE (AVX512F_LEN / 8)
+#define SRC_SIZE (AVX512F_LEN / 16)
+
+#include "limits.h"
+
+#include "avx512f-mask-type.h"
+
+static char
+EVAL(short_to_char, AVX512F_LEN,) (short iVal)
+{
+ char sVal;
+
+ if (iVal < CHAR_MIN)
+ sVal = CHAR_MIN;
+ else if (iVal > CHAR_MAX)
+ sVal = CHAR_MAX;
+ else
+ sVal = iVal;
+
+ return sVal;
+}
+
+void
+CALC (short *src1, short *src2, char *dst)
+{
+ int i;
+ short *ptr;
+
+ for (i = 0; i < DST_SIZE; i++)
+ {
+ ptr = (i / 8) % 2 ? src2 : src1;
+ dst[i] = EVAL(short_to_char, AVX512F_LEN,) (ptr[i % 8 + (i / 16) * 8]);
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ char dst_ref[DST_SIZE];
+ int i;
+
+ for (i = 0; i < DST_SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ for (i = 0; i < SRC_SIZE; i++)
+ {
+ s1.a[i] = i + 10;
+ s2.a[i] = i + 15;
+ }
+
+ res1.x = INTRINSIC (_packs_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_packs_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_packs_epi16) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (dst_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (dst_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, dst_ref))
+ abort ();
+
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpackusdw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpackusdw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackusdw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpackusdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackusdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackusdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpackusdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask64 mx;
+volatile __mmask32 my;
+volatile __mmask16 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_packus_epi32 (x, x);
+ x = _mm512_mask_packus_epi32 (x, mx, x, x);
+ x = _mm512_maskz_packus_epi32 (mx, x, x);
+ y = _mm256_mask_packus_epi32 (y, my, y, y);
+ y = _mm256_maskz_packus_epi32 (my, y, y);
+ z = _mm_mask_packus_epi32 (z, mz, z, z);
+ z = _mm_maskz_packus_epi32 (mz, z, z);
+}
+
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define DST_SIZE (AVX512F_LEN / 16)
+#define SRC_SIZE (AVX512F_LEN / 32)
+
+#include "limits.h"
+
+#include "avx512f-mask-type.h"
+
+static unsigned short
+EVAL(int_to_ushort, AVX512F_LEN,) (int iVal)
+{
+ unsigned short sVal;
+
+ if (iVal < 0)
+ sVal = 0;
+ else if (iVal > USHRT_MAX)
+ sVal = USHRT_MAX;
+ else
+ sVal = iVal;
+
+ return sVal;
+}
+
+void
+CALC (int *src1, int *src2, unsigned short *dst)
+{
+ int i;
+ int *ptr;
+
+ for (i = 0; i < DST_SIZE; i++)
+ {
+ ptr = (i / 4) % 2 ? src2 : src1;
+ dst[i] = EVAL(int_to_ushort, AVX512F_LEN,) (ptr[i % 4 + (i / 8) * 4]);
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_d) s1, s2;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short dst_ref[DST_SIZE];
+ int i;
+
+ for (i = 0; i < DST_SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ for (i = 0; i < SRC_SIZE; i++)
+ {
+ s1.a[i] = i + 10;
+ s2.a[i] = i + 15;
+ }
+
+ res1.x = INTRINSIC (_packus_epi32) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_packus_epi32) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_packus_epi32) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (dst_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (dst_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, dst_ref))
+ abort ();
+
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpackuswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpackuswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackuswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpackuswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackuswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpackuswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpackuswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask64 mx;
+volatile __mmask32 my;
+volatile __mmask16 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_packus_epi16 (x, x);
+ x = _mm512_mask_packus_epi16 (x, mx, x, x);
+ x = _mm512_maskz_packus_epi16 (mx, x, x);
+ y = _mm256_mask_packus_epi16 (y, my, y, y);
+ y = _mm256_maskz_packus_epi16 (my, y, y);
+ z = _mm_mask_packus_epi16 (z, mz, z, z);
+ z = _mm_maskz_packus_epi16 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define DST_SIZE (AVX512F_LEN / 8)
+#define SRC_SIZE (AVX512F_LEN / 16)
+
+#include "limits.h"
+
+#include "avx512f-mask-type.h"
+
+static unsigned char
+EVAL(short_to_uchar, AVX512F_LEN,) (short iVal)
+{
+ unsigned char sVal;
+
+ if (iVal < 0)
+ sVal = 0;
+ else if (iVal > UCHAR_MAX)
+ sVal = UCHAR_MAX;
+ else
+ sVal = iVal;
+
+ return sVal;
+}
+
+void
+CALC (short *src1, short *src2, unsigned char *dst)
+{
+ int i;
+ short *ptr;
+
+ for (i = 0; i < DST_SIZE; i++)
+ {
+ ptr = (i / 8) % 2 ? src2 : src1;
+ dst[i] = EVAL(short_to_uchar, AVX512F_LEN,) (ptr[i % 8 + (i / 16) * 8]);
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned char dst_ref[DST_SIZE];
+ int i;
+
+ for (i = 0; i < DST_SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ for (i = 0; i < SRC_SIZE; i++)
+ {
+ s1.a[i] = i + 10;
+ s2.a[i] = i + 15;
+ }
+
+ res1.x = INTRINSIC (_packus_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_packus_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_packus_epi16) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (dst_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (dst_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, dst_ref))
+ abort ();
+
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask64 m512;
+volatile __mmask32 m256;
+volatile __mmask16 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_add_epi8 (x512, x512);
+ x512 = _mm512_mask_add_epi8 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_add_epi8 (m512, x512, x512);
+ x256 = _mm256_mask_add_epi8 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_add_epi8 (m256, x256, x256);
+ x128 = _mm_mask_add_epi8 (x128, m128, x128, x128);
+ x128 = _mm_maskz_add_epi8 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, char *s1, char *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s1[i] + s2[i];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ src2.a[i] = 3 + sign * 11 * (i % 377) * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_add_epi8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_add_epi8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_add_epi8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask64 m512;
+volatile __mmask32 m256;
+volatile __mmask16 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_adds_epi8 (x512, x512);
+ x512 = _mm512_mask_adds_epi8 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_adds_epi8 (m512, x512, x512);
+ x256 = _mm256_mask_adds_epi8 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_adds_epi8 (m256, x256, x256);
+ x128 = _mm_mask_adds_epi8 (x128, m128, x128, x128);
+ x128 = _mm_maskz_adds_epi8 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, char *s1, char *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (int)s1[i] + (int)s2[i];
+ if (tmp > 0x7F) tmp = 0x7F;
+ if (tmp < (char)0x80) tmp = (char)0x80;
+ r[i] = (char)tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ src2.a[i] = 3 + sign * 11 * (i % 377) * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_adds_epi8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_adds_epi8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_adds_epi8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask32 m512;
+volatile __mmask16 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_adds_epi16 (x512, x512);
+ x512 = _mm512_mask_adds_epi16 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_adds_epi16 (m512, x512, x512);
+ x256 = _mm256_mask_adds_epi16 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_adds_epi16 (m256, x256, x256);
+ x128 = _mm_mask_adds_epi16 (x128, m128, x128, x128);
+ x128 = _mm_maskz_adds_epi16 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, short *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (int)s1[i] + (int)s2[i];
+ if (tmp > 0x7FFF) tmp = 0x7FFF;
+ if (tmp < (short)0x8000) tmp = (short)0x8000;
+ r[i] = (short)tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ src2.a[i] = 3 + sign * 11 * (i % 377) * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_adds_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_adds_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_adds_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask64 m512;
+volatile __mmask32 m256;
+volatile __mmask16 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_adds_epu8 (x512, x512);
+ x512 = _mm512_mask_adds_epu8 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_adds_epu8 (m512, x512, x512);
+ x256 = _mm256_mask_adds_epu8 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_adds_epu8 (m256, x256, x256);
+ x128 = _mm_mask_adds_epu8 (x128, m128, x128, x128);
+ x128 = _mm_maskz_adds_epu8 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned char *r, unsigned char *s1, unsigned char *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (int)s1[i] + (int)s2[i];
+ if (tmp > 0xFF) tmp = 0xFF;
+ if (tmp < 0) tmp = 0;
+ r[i] = (unsigned char)tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned char res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + 7 * i % 291;
+ src2.a[i] = 3 + 11 * (i % 377) * i;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_adds_epu8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_adds_epu8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_adds_epu8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddusw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask32 m512;
+volatile __mmask16 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_adds_epu16 (x512, x512);
+ x512 = _mm512_mask_adds_epu16 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_adds_epu16 (m512, x512, x512);
+ x256 = _mm256_mask_adds_epu16 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_adds_epu16 (m256, x256, x256);
+ x128 = _mm_mask_adds_epu16 (x128, m128, x128, x128);
+ x128 = _mm_maskz_adds_epu16 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned short *r, unsigned short *s1, unsigned short *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (int)s1[i] + (int)s2[i];
+ if (tmp > 0xFFFF) tmp = 0xFFFF;
+ if (tmp < 0) tmp = 0;
+ r[i] = (unsigned short)tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + 7 * i % 291;
+ src2.a[i] = 3 + 11 * (i % 377) * i;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_adds_epu16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_adds_epu16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_adds_epu16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask32 m512;
+volatile __mmask16 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_add_epi16 (x512, x512);
+ x512 = _mm512_mask_add_epi16 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_add_epi16 (m512, x512, x512);
+ x256 = _mm256_mask_add_epi16 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_add_epi16 (m256, x256, x256);
+ x128 = _mm_mask_add_epi16 (x128, m128, x128, x128);
+ x128 = _mm_maskz_add_epi16 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, short *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s1[i] + s2[i];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ src2.a[i] = 3 + sign * 11 * (i % 377) * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_add_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_add_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_add_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpalignr\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask64 m1;
+volatile __mmask32 m2;
+volatile __mmask16 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_alignr_epi8 (z, z, 10);
+ z = _mm512_mask_alignr_epi8 (z, m1, z, z, 10);
+ z = _mm512_maskz_alignr_epi8 (m1, z, z, 10);
+ y = _mm256_mask_alignr_epi8 (y, m2, y, y, 10);
+ y = _mm256_maskz_alignr_epi8 (m2, y, y, 10);
+ x = _mm_mask_alignr_epi8 (x, m3, x, x, 10);
+ x = _mm_maskz_alignr_epi8 (m3, x, x, 10);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#include <string.h>
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+#define N 0x3
+
+void
+CALC (char *src1, char *src2, char * dst)
+{
+ /* result for EVEX.U1.512 version consists from 4 result block, each of them
+ * has length of 128 bits. */
+ unsigned block_len = 16;
+ unsigned double_block_len = 32;
+ unsigned shift = 0;
+ char buf[double_block_len];
+ char *bout = dst;
+ int bits, i;
+
+ for (bits = 0; bits < AVX512F_LEN; bits += 128)
+ {
+ memcpy (&buf[0], src2 + shift, block_len);
+ memcpy (&buf[block_len], src1 + shift, block_len);
+
+ for (i = 0; i < block_len; i++)
+ /* shift counts larger than 32 produces zero result. */
+ if (N >= 32 || N + i >= 32)
+ bout[i] = 0;
+ else
+ bout[i] = buf[N + i];
+
+ shift += block_len;
+ bout += block_len;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_b) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i;
+ s2.a[i] = i * 2;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_alignr_epi8) (s1.x, s2.x, N);
+ res2.x = INTRINSIC (_mask_alignr_epi8) (res2.x, mask, s1.x, s2.x, N);
+ res3.x = INTRINSIC (_maskz_alignr_epi8) (mask, s1.x, s2.x, N);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x;
+volatile __m256i y;
+volatile __m512i z;
+volatile __mmask64 m1;
+volatile __mmask32 m2;
+volatile __mmask16 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_avg_epu8 (z, z);
+ z = _mm512_mask_avg_epu8 (z, m1, z, z);
+ z = _mm512_maskz_avg_epu8 (m1, z, z);
+ y = _mm256_mask_avg_epu8 (y, m2, y, y);
+ y = _mm256_maskz_avg_epu8 (m2, y, y);
+ x = _mm_mask_avg_epu8 (x, m3, x, x);
+ x = _mm_maskz_avg_epu8 (m3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *s1, char *s2, char *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = ((unsigned char) s1[i] +
+ (unsigned char) s2[i] + 1) >> 1;
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_b) s1, s2, res1, res2 ,res3;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i + 15;
+ s2.a[i] = i + 14;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_avg_epu8) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_avg_epu8) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_avg_epu8) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpavgw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x;
+volatile __m256i y;
+volatile __m512i z;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_avg_epu16 (z, z);
+ z = _mm512_mask_avg_epu16 (z, m1, z, z);
+ z = _mm512_maskz_avg_epu16 (m1, z, z);
+ y = _mm256_mask_avg_epu16 (y, m2, y, y);
+ y = _mm256_maskz_avg_epu16 (m2, y, y);
+ x = _mm_mask_avg_epu16 (x, m3, x, x);
+ x = _mm_maskz_avg_epu16 (m3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *s1, short *s2, short *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = ((unsigned short) s1[i] +
+ (unsigned short) s2[i] + 1) >> 1;
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res1, res2 ,res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i + 15;
+ s2.a[i] = i + 14;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_avg_epu16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_avg_epu16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_avg_epu16) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "(vpblendmb|vmovdqu8)\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "(vpblendmb|vmovdqu8)\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "(vpblendmb|vmovdqu8)\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i xx;
+volatile __m512i xxx;
+volatile __mmask8 m;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm256_mask_blend_epi8 (m, x, x);
+ xx = _mm_mask_blend_epi8 (m, xx, xx);
+ xxx = _mm512_mask_blend_epi8 (m, xxx, xxx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, char *s1, char *s2, MASK_TYPE mask)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = (mask & (1LL << i)) ? s2[i] : s1[i];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 15 + 46 * i * sign;
+ src2.a[i] = -22 + i * sign;
+ sign = sign * -1;
+ }
+
+ res1.x = INTRINSIC (_mask_blend_epi8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a, mask);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "(vpblendmw|vmovdqu16)\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "(vpblendmw|vmovdqu16)\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "(vpblendmw|vmovdqu16)\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i xx;
+volatile __m512i xxx;
+volatile __mmask8 m;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm256_mask_blend_epi16 (m, x, x);
+ xx = _mm_mask_blend_epi16 (m, xx, xx);
+ xxx = _mm512_mask_blend_epi16 (m, xxx, xxx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, short *s2, MASK_TYPE mask)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = (mask & (1 << i)) ? s2[i] : s1[i];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 15 + 3467 * i * sign;
+ src2.a[i] = -2217 * i * sign;
+ sign = sign * -1;
+ }
+
+ res1.x = INTRINSIC (_mask_blend_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a, mask);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpbroadcastb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile char w;
+volatile __mmask64 mx;
+volatile __mmask32 my;
+volatile __mmask16 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_broadcastb_epi8 (z);
+ x = _mm512_mask_broadcastb_epi8 (x, mx, z);
+ x = _mm512_maskz_broadcastb_epi8 (mx, z);
+ y = _mm256_mask_broadcastb_epi8 (y, my, z);
+ y = _mm256_maskz_broadcastb_epi8 (my, z);
+ z = _mm_mask_broadcastb_epi8 (z, mz, z);
+ z = _mm_maskz_broadcastb_epi8 (mz, z);
+
+ x = _mm512_set1_epi8 (w);
+ x = _mm512_mask_set1_epi8 (x, mx, w);
+ x = _mm512_maskz_set1_epi8 (mx, w);
+ y = _mm256_mask_set1_epi8 (y, my, w);
+ y = _mm256_maskz_set1_epi8 (my, w);
+ z = _mm_mask_set1_epi8 (z, mz, w);
+ z = _mm_maskz_set1_epi8 (mz, w);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, char *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s[0];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3;
+ UNION_TYPE (128, i_b) src;
+ MASK_TYPE mask = SIZE | 123;
+ char res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < 16; i++)
+ {
+ src.a[i] = 1 + 3 * i * sign;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ CALC (res_ref, src.a);
+
+ if (AVX512F_LEN == 512)
+ {
+ res1.x = INTRINSIC (_broadcastb_epi8) (src.x);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+ }
+
+ res2.x = INTRINSIC (_mask_broadcastb_epi8) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_broadcastb_epi8) (mask, src.x);
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+
+ CALC (res_ref, src.a);
+
+ if (AVX512F_LEN == 512)
+ {
+ res1.x = INTRINSIC (_set1_epi8) (src.a[0]);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+ }
+
+ res2.x = INTRINSIC (_mask_set1_epi8) (res2.x, mask, src.a[0]);
+ res3.x = INTRINSIC (_maskz_set1_epi8) (mask, src.a[0]);
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpbroadcastw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile short w;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_broadcastw_epi16 (z);
+ x = _mm512_mask_broadcastw_epi16 (x, mx, z);
+ x = _mm512_maskz_broadcastw_epi16 (mx, z);
+ y = _mm256_mask_broadcastw_epi16 (y, my, z);
+ y = _mm256_maskz_broadcastw_epi16 (my, z);
+ z = _mm_mask_broadcastw_epi16 (z, mz, z);
+ z = _mm_maskz_broadcastw_epi16 (mz, z);
+
+ x = _mm512_set1_epi16 (w);
+ x = _mm512_mask_set1_epi16 (x, mx, w);
+ x = _mm512_maskz_set1_epi16 (mx, w);
+ y = _mm256_mask_set1_epi16 (y, my, w);
+ y = _mm256_maskz_set1_epi16 (my, w);
+ z = _mm_mask_set1_epi16 (z, mz, w);
+ z = _mm_maskz_set1_epi16 (mz, w);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s[0];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3;
+ UNION_TYPE (128, i_w) src;
+ MASK_TYPE mask = SIZE | 123;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < 8; i++)
+ {
+ src.a[i] = 1 + 3 * i * sign;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ CALC (res_ref, src.a);
+
+ if (AVX512F_LEN == 512)
+ {
+ res1.x = INTRINSIC (_broadcastw_epi16) (src.x);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+ }
+
+ res2.x = INTRINSIC (_mask_broadcastw_epi16) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_broadcastw_epi16) (mask, src.x);
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+
+ CALC (res_ref, src.a);
+
+ if (AVX512F_LEN == 512)
+ {
+ res1.x = INTRINSIC (_set1_epi16) (src.a[0]);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+ }
+
+ res2.x = INTRINSIC (_mask_set1_epi16) (res2.x, mask, src.a[0]);
+ res3.x = INTRINSIC (_maskz_set1_epi16) (mask, src.a[0]);
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler "vpcmpb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m512i xq;
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask16 m;
+volatile __mmask32 mm;
+volatile __mmask64 mmm;
+
+void extern
+avx512bw_test (void)
+{
+ mmm = _mm512_cmp_epi8_mask (xq, xq, _MM_CMPINT_GE);
+ mmm = _mm512_mask_cmp_epi8_mask (m, xq, xq, _MM_CMPINT_NLE);
+ mm = _mm256_cmp_epi8_mask (x, x, _MM_CMPINT_GT);
+ mm = _mm256_mask_cmp_epi8_mask (m, x, x, _MM_CMPINT_EQ);
+ m = _mm_cmp_epi8_mask (xx, xx, _MM_CMPINT_LT);
+ m = _mm_mask_cmp_epi8_mask (m, xx, xx, _MM_CMPINT_LE);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#include <math.h>
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+#if AVX512F_LEN == 512
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 64; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm512_loadu_si512 (s1); \
+ source2.x = _mm512_loadu_si512 (s2); \
+ dst1 = _mm512_cmp_epi8_mask (source1.x, source2.x, imm);\
+ dst2 = _mm512_mask_cmp_epi8_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 32; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_si256 ((__m256i*)s1); \
+ source2.x = _mm256_loadu_si256 ((__m256i*)s2); \
+ dst1 = _mm256_cmp_epi8_mask (source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_epi8_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 16; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_si128 ((__m128i*)s1); \
+ source2.x = _mm_loadu_si128 ((__m128i*)s2); \
+ dst1 = _mm_cmp_epi8_mask (source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_epi8_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+void
+TEST ()
+{
+ char s1[64] = {34, 78, 53, 64,
+ 1, 57, 11, 231,
+ 14, 45, 71, 75,
+ 55, 66, 21, 73,
+ 34, 68, 3, 56,
+ 1, 57, 111, 241,
+ 14, 15, 61, 75,
+ 55, 16, 52, 3,
+ 34, 78, 53, 64,
+ 1, 57, 11, 231,
+ 14, 45, 71, 75,
+ 45, 66, 21, 73,
+ 34, 68, 3, 56,
+ 1, 57, 111, 241,
+ 14, 15, 61, 75,
+ 55, 16, 52, 3};
+ char s2[64] = {4, 68, 86, 8,
+ 1, 46, 1, 1,
+ 45, 67, 36, 3,
+ 4, 39, 56, 56,
+ 124, 78, 53, 56,
+ 1, 46, 1, 12,
+ 45, 47, 36, 13,
+ 4, 35, 56, 67,
+ 4, 68, 86, 8,
+ 1, 46, 1, 1,
+ 45, 67, 36, 3,
+ 4, 39, 56, 56,
+ 124, 78, 53, 56,
+ 1, 46, 1, 12,
+ 45, 47, 36, 13,
+ 4, 35, 56, 67};
+ UNION_TYPE (AVX512F_LEN, i_b) source1, source2;
+ MASK_TYPE dst1, dst2, dst_ref;
+ MASK_TYPE mask = MASK_VALUE;
+ int i;
+
+ CMP(0x00, s1[i] == s2[i]);
+ CMP(0x01, s1[i] < s2[i]);
+ CMP(0x02, s1[i] <= s2[i]);
+ CMP(0x03, 0);
+ CMP(0x04, s1[i] != s2[i]);
+ CMP(0x05, s1[i] >= s2[i]);
+ CMP(0x06, s1[i] > s2[i]);
+ CMP(0x07, 1);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpcmpeqb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]\[\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpeqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\[\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpeqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\[\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpeqb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpeqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpeqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+volatile __mmask64 m64;
+
+void extern
+avx512bw_test (void)
+{
+ m16 = _mm_cmpeq_epi8_mask (x128, x128);
+ m32 = _mm256_cmpeq_epi8_mask (x256, x256);
+ m64 = _mm512_cmpeq_epi8_mask (x512, x512);
+ m16 = _mm_mask_cmpeq_epi8_mask (3, x128, x128);
+ m32 = _mm256_mask_cmpeq_epi8_mask (3, x256, x256);
+ m64 = _mm512_mask_cmpeq_epi8_mask (3, x512, x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *r, char *s1, char *s2)
+{
+ int i;
+ *r = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] == s2[i])
+ *r = *r | (one << i);
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+ res1 = 0;
+ res2 = 0;
+
+ for (i = 0; i < SIZE / 2; i++)
+ {
+ src1.a[i * 2] = i;
+ src1.a[i * 2 + 1] = i * i;
+ src2.a[i * 2] = 2 * i;
+ src2.a[i * 2 + 1] = i * i;
+ }
+
+ res1 = INTRINSIC (_cmpeq_epi8_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpeq_epi8_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res_ref != res1)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res_ref != res2)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpcmpeqw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpeqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpeqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpeqw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpeqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpeqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+
+void extern
+avx512bw_test (void)
+{
+ m8 = _mm_cmpeq_epi16_mask (x128, x128);
+ m8 = _mm_mask_cmpeq_epi16_mask (3, x128, x128);
+ m16 = _mm256_cmpeq_epi16_mask (x256, x256);
+ m16 = _mm256_mask_cmpeq_epi16_mask (3, x256, x256);
+ m32 = _mm512_mask_cmpeq_epi16_mask (3, x512, x512);
+ m32 = _mm512_cmpeq_epi16_mask (x512, x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *r, short *s1, short *s2)
+{
+ int i;
+ *r = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] == s2[i])
+ *r = *r | (one << i);
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE / 2; i++)
+ {
+ src1.a[i * 2] = i;
+ src1.a[i * 2 + 1] = i * i;
+ src2.a[i * 2] = 2 * i;
+ src2.a[i * 2 + 1] = i * i;
+ }
+
+ res1 = INTRINSIC (_cmpeq_epi16_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpeq_epi16_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res_ref != res1)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res_ref != res2)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpcmpgtb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpgtb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpgtb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpgtb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpgtb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpgtb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+volatile __mmask64 m64;
+
+void extern
+avx512bw_test (void)
+{
+ m16 = _mm_cmpgt_epi8_mask (x128, x128);
+ m32 = _mm256_cmpgt_epi8_mask (x256, x256);
+ m64 = _mm512_cmpgt_epi8_mask (x512, x512);
+ m16 = _mm_mask_cmpgt_epi8_mask (3, x128, x128);
+ m32 = _mm256_mask_cmpgt_epi8_mask (3, x256, x256);
+ m64 = _mm512_mask_cmpgt_epi8_mask (3, x512, x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *r, char *s1, char *s2)
+{
+ int i;
+ *r = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] > s2[i])
+ *r = *r | (one << i);
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE / 2; i++)
+ {
+ src1.a[i * 2] = i;
+ src1.a[i * 2 + 1] = i * i;
+ src2.a[i * 2] = 2 * i;
+ src2.a[i * 2 + 1] = i * i;
+ }
+
+ res1 = INTRINSIC (_cmpgt_epi8_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpgt_epi8_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res_ref != res1)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res_ref != res2)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpcmpgtw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpgtw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpgtw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpgtw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpgtw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpgtw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+
+void extern
+avx512bw_test (void)
+{
+ m8 = _mm_cmpgt_epi16_mask (x128, x128);
+ m16 = _mm256_cmpgt_epi16_mask (x256, x256);
+ m32 = _mm512_cmpgt_epi16_mask (x512, x512);
+ m8 = _mm_mask_cmpgt_epi16_mask (3, x128, x128);
+ m16 = _mm256_mask_cmpgt_epi16_mask (3, x256, x256);
+ m32 = _mm512_mask_cmpgt_epi16_mask (3, x512, x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *r, short *s1, short *s2)
+{
+ int i;
+ *r = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] > s2[i])
+ *r = *r | (one << i);
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE / 2; i++)
+ {
+ src1.a[i * 2] = i;
+ src1.a[i * 2 + 1] = i * i;
+ src2.a[i * 2] = 2 * i;
+ src2.a[i * 2 + 1] = i * i;
+ }
+
+ res1 = INTRINSIC (_cmpgt_epi16_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_cmpgt_epi16_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res_ref != res1)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res_ref != res2)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler "vpcmpub\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpub\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpub\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpub\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpub\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpub\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m512i xq;
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask16 m;
+volatile __mmask32 mm;
+volatile __mmask64 mmm;
+
+void extern
+avx512bw_test (void)
+{
+ mmm = _mm512_cmp_epu8_mask (xq, xq, _MM_CMPINT_EQ);
+ mmm = _mm512_mask_cmp_epu8_mask (m, xq, xq, _MM_CMPINT_LT);
+ mm = _mm256_cmp_epu8_mask (x, x, _MM_CMPINT_LE);
+ mm = _mm256_mask_cmp_epu8_mask (m, x, x, _MM_CMPINT_UNUSED);
+ m = _mm_cmp_epu8_mask (xx, xx, _MM_CMPINT_NE);
+ m = _mm_mask_cmp_epu8_mask (m, xx, xx, _MM_CMPINT_NLT);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#include <math.h>
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+#if AVX512F_LEN == 512
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 64; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm512_loadu_si512 (s1); \
+ source2.x = _mm512_loadu_si512 (s2); \
+ dst1 = _mm512_cmp_epu8_mask (source1.x, source2.x, imm);\
+ dst2 = _mm512_mask_cmp_epu8_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 32; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_si256 ((__m256i*)s1); \
+ source2.x = _mm256_loadu_si256 ((__m256i*)s2); \
+ dst1 = _mm256_cmp_epu8_mask (source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_epu8_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 16; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_si128 ((__m128i*)s1); \
+ source2.x = _mm_loadu_si128 ((__m128i*)s2); \
+ dst1 = _mm_cmp_epu8_mask (source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_epu8_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+void
+TEST ()
+{
+ unsigned char s1[64] = {34, 78, 53, 64,
+ 1, 57, 11, 231,
+ 14, 45, 71, 75,
+ 55, 66, 21, 73,
+ 34, 68, 3, 56,
+ 1, 57, 111, 241,
+ 14, 15, 61, 75,
+ 55, 16, 52, 3,
+ 34, 78, 53, 64,
+ 1, 57, 11, 231,
+ 14, 45, 71, 75,
+ 55, 66, 21, 73,
+ 34, 68, 3, 56,
+ 1, 57, 111, 241,
+ 14, 15, 61, 75,
+ 55, 16, 52, 3};
+ unsigned char s2[64] = {4, 68, 86, 8,
+ 1, 46, 1, 1,
+ 45, 67, 36, 3,
+ 4, 39, 56, 56,
+ 124, 78, 53, 56,
+ 1, 46, 1, 12,
+ 45, 47, 36, 13,
+ 4, 35, 56, 67,
+ 4, 68, 86, 8,
+ 1, 46, 1, 1,
+ 45, 67, 36, 3,
+ 4, 39, 56, 56,
+ 124, 78, 53, 56,
+ 1, 46, 1, 12,
+ 45, 47, 36, 13,
+ 4, 35, 56, 67};
+ UNION_TYPE (AVX512F_LEN, i_b) source1, source2;
+ MASK_TYPE dst1, dst2, dst_ref;
+ MASK_TYPE mask = MASK_VALUE;
+ int i;
+
+ CMP(0x00, s1[i] == s2[i]);
+ CMP(0x01, s1[i] < s2[i]);
+ CMP(0x02, s1[i] <= s2[i]);
+ CMP(0x03, 0);
+ CMP(0x04, s1[i] != s2[i]);
+ CMP(0x05, s1[i] >= s2[i]);
+ CMP(0x06, s1[i] > s2[i]);
+ CMP(0x07, 1);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler "vpcmpuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpuw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpuw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpuw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpuw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m512i xq;
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask8 m;
+volatile __mmask16 mm;
+volatile __mmask32 mmm;
+
+void extern
+avx512bw_test (void)
+{
+ mmm = _mm512_cmp_epu16_mask (xq, xq, _MM_CMPINT_NE);
+ mmm = _mm512_mask_cmp_epu16_mask (m, xq, xq, _MM_CMPINT_NLT);
+ mm = _mm256_cmp_epu16_mask (x, x, _MM_CMPINT_GE);
+ mm = _mm256_mask_cmp_epu16_mask (m, x, x, _MM_CMPINT_NLE);
+ m = _mm_cmp_epu16_mask (xx, xx, _MM_CMPINT_GT);
+ m = _mm_mask_cmp_epu16_mask (m, xx, xx, _MM_CMPINT_EQ);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#include <math.h>
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+#if AVX512F_LEN == 512
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 32; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm512_loadu_si512 (s1); \
+ source2.x = _mm512_loadu_si512 (s2); \
+ dst1 = _mm512_cmp_epu16_mask (source1.x, source2.x, imm);\
+ dst2 = _mm512_mask_cmp_epu16_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 16; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_si256 ((__m256i*)s1); \
+ source2.x = _mm256_loadu_si256 ((__m256i*)s2); \
+ dst1 = _mm256_cmp_epu16_mask (source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_epu16_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 8; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_si128 ((__m128i*)s1); \
+ source2.x = _mm_loadu_si128 ((__m128i*)s2); \
+ dst1 = _mm_cmp_epu16_mask (source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_epu16_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+void
+TEST ()
+{
+ unsigned short s1[32] = {2134, 6678, 453, 54646,
+ 231, 5674, 111, 23241,
+ 12314, 145, 671, 7575,
+ 23455, 166, 5321, 5673,
+ 2134, 6678, 453, 54646,
+ 231, 5674, 111, 23241,
+ 12314, 145, 671, 7575,
+ 23455, 166, 5321, 5673};
+ unsigned short s2[32] = {41124, 6678, 8653, 856,
+ 231, 4646, 111, 124,
+ 2745, 4567, 3676, 123,
+ 714, 3589, 5683, 5673,
+ 41124, 6678, 8653, 856,
+ 231, 4646, 111, 124,
+ 2745, 4567, 3676, 123,
+ 714, 3589, 5683, 5673};
+ UNION_TYPE (AVX512F_LEN, i_w) source1, source2;
+ MASK_TYPE dst1, dst2, dst_ref;
+ MASK_TYPE mask = MASK_VALUE;
+ int i;
+
+ CMP(0x00, s1[i] == s2[i]);
+ CMP(0x01, s1[i] < s2[i]);
+ CMP(0x02, s1[i] <= s2[i]);
+ CMP(0x03, 0);
+ CMP(0x04, s1[i] != s2[i]);
+ CMP(0x05, s1[i] >= s2[i]);
+ CMP(0x06, s1[i] > s2[i]);
+ CMP(0x07, 1);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler "vpcmpw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m512i xq;
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask8 m;
+volatile __mmask16 mm;
+volatile __mmask32 mmm;
+
+void extern
+avx512bw_test (void)
+{
+ mmm = _mm512_cmp_epi16_mask (xq, xq, _MM_CMPINT_GT);
+ mmm = _mm512_mask_cmp_epi16_mask (m, xq, xq, _MM_CMPINT_EQ);
+ mm = _mm256_cmp_epi16_mask (x, x, _MM_CMPINT_EQ);
+ mm = _mm256_mask_cmp_epi16_mask (m, x, x, _MM_CMPINT_LT);
+ m = _mm_cmp_epi16_mask (xx, xx, _MM_CMPINT_LE);
+ m = _mm_mask_cmp_epi16_mask (m, xx, xx, _MM_CMPINT_UNUSED);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#include <math.h>
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+#if AVX512F_LEN == 512
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 32; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm512_loadu_si512 (s1); \
+ source2.x = _mm512_loadu_si512 (s2); \
+ dst1 = _mm512_cmp_epi16_mask (source1.x, source2.x, imm);\
+ dst2 = _mm512_mask_cmp_epi16_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 16; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_si256 ((__m256i*)s1); \
+ source2.x = _mm256_loadu_si256 ((__m256i*)s2); \
+ dst1 = _mm256_cmp_epi16_mask (source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_epi16_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 8; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_si128 ((__m128i*)s1); \
+ source2.x = _mm_loadu_si128 ((__m128i*)s2); \
+ dst1 = _mm_cmp_epi16_mask (source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_epi16_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+void
+TEST ()
+{
+ short s1[32] = {2134, 6678, 453, 54646,
+ 231, 5674, 111, 23241,
+ 12314, 145, 671, 7575,
+ 23455, 166, 5321, 5673,
+ 2134, 6678, 453, 54646,
+ 231, 5674, 111, 23241,
+ 12314, 145, 671, 7575,
+ 23455, 166, 5321, 5673};
+ short s2[32] = {41124, 6678, 8653, 856,
+ 231, 4646, 111, 124,
+ 2745, 4567, 3676, 123,
+ 714, 3589, 5683, 5673,
+ 41124, 6678, 8653, 856,
+ 231, 4646, 111, 124,
+ 2745, 4567, 3676, 123,
+ 714, 3589, 5683, 5673};
+ UNION_TYPE (AVX512F_LEN, i_w) source1, source2;
+ MASK_TYPE dst1, dst2, dst_ref;
+ MASK_TYPE mask = MASK_VALUE;
+ int i;
+
+ CMP(0x00, s1[i] == s2[i]);
+ CMP(0x01, s1[i] < s2[i]);
+ CMP(0x02, s1[i] <= s2[i]);
+ CMP(0x03, 0);
+ CMP(0x04, s1[i] != s2[i]);
+ CMP(0x05, s1[i] >= s2[i]);
+ CMP(0x06, s1[i] > s2[i]);
+ CMP(0x07, 1);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermi2w\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermi2w\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermi2w\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x3;
+volatile __m256i x2;
+volatile __m128i x1;
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m3;
+volatile __mmask16 m2;
+volatile __mmask8 m1;
+
+void extern
+avx512bw_test (void)
+{
+ x3 = _mm512_mask2_permutex2var_epi16 (x3, z, m3, x3);
+ x2 = _mm256_mask2_permutex2var_epi16 (x2, y, m2, x2);
+ x1 = _mm_mask2_permutex2var_epi16 (x1, x, m1, x1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "math.h"
+#include "values.h"
+#include "avx512f-mask-type.h"
+
+#define NUM 32
+
+void
+CALC (short *dst, short *src1, short *ind, short *src2)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ unsigned long long offset = ind[i] & (SIZE - 1);
+ unsigned long long cond = ind[i] & SIZE;
+
+ dst[i] = cond ? src2[offset] : src1[offset];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, j;
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res, ind;
+ short res_ref[SIZE];
+
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < NUM; i++)
+ {
+ for (j = 0; j < SIZE; j++)
+ {
+ ind.a[j] = DEFAULT_VALUE;
+ s1.a[j] = i * 2 * j + 1;
+ s2.a[j] = i * 2 * j;
+
+ res.a[j] = DEFAULT_VALUE;
+ }
+
+ CALC (res_ref, s1.a, ind.a, s2.a);
+
+ res.x =
+ INTRINSIC (_mask2_permutex2var_epi16) (s1.x, ind.x, mask,
+ s2.x);
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res, res_ref))
+ abort ();
+ }
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } *
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2w\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x3;
+volatile __m256i x2;
+volatile __m128i x1;
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m3;
+volatile __mmask16 m2;
+volatile __mmask8 m1;
+
+void extern
+avx512bw_test (void)
+{
+ x3 = _mm512_permutex2var_epi16 (x3, z, x3);
+ x3 = _mm512_mask_permutex2var_epi16 (x3, m3, z, x3);
+ x3 = _mm512_maskz_permutex2var_epi16 (m3, x3, z, x3);
+ x2 = _mm256_permutex2var_epi16 (x2, y, x2);
+ x2 = _mm256_mask_permutex2var_epi16 (x2, m2, y, x2);
+ x2 = _mm256_maskz_permutex2var_epi16 (m2, x2, y, x2);
+ x1 = _mm_permutex2var_epi16 (x1, x, x1);
+ x1 = _mm_mask_permutex2var_epi16 (x1, m1, x, x1);
+ x1 = _mm_maskz_permutex2var_epi16 (m1, x1, x, x1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "math.h"
+#include "values.h"
+#include "avx512f-mask-type.h"
+
+#define NUM 32
+
+void
+CALC (short *dst, short *src1, short *ind, short *src2)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ unsigned long long offset = ind[i] & (SIZE - 1);
+ unsigned long long cond = ind[i] & SIZE;
+
+ dst[i] = cond ? src2[offset] : src1[offset];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, j;
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res1, res2, res3, ind;
+ short res_ref[SIZE];
+
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < NUM; i++)
+ {
+ for (j = 0; j < SIZE; j++)
+ {
+ ind.a[j] = i * (j << 1);
+ s1.a[j] = DEFAULT_VALUE;
+ s2.a[j] = 1.5 * i * 2 * j;
+
+ res1.a[j] = DEFAULT_VALUE;
+ res2.a[j] = DEFAULT_VALUE;
+ res3.a[j] = DEFAULT_VALUE;
+ }
+
+ CALC (res_ref, s1.a, ind.a, s2.a);
+
+ res1.x = INTRINSIC (_permutex2var_epi16) (s1.x, ind.x, s2.x);
+ res2.x =
+ INTRINSIC (_mask_permutex2var_epi16) (s1.x, mask, ind.x, s2.x);
+ res3.x =
+ INTRINSIC (_maskz_permutex2var_epi16) (mask, s1.x, ind.x,
+ s2.x);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+ }
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x1;
+volatile __m256i x2;
+volatile __m128i x3;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ x1 = _mm512_permutexvar_epi16 (x1, x1);
+ x1 = _mm512_maskz_permutexvar_epi16 (m1, x1, x1);
+ x1 = _mm512_mask_permutexvar_epi16 (x1, m1, x1, x1);
+ x2 = _mm256_permutexvar_epi16 (x2, x2);
+ x2 = _mm256_maskz_permutexvar_epi16 (m2, x2, x2);
+ x2 = _mm256_mask_permutexvar_epi16 (x2, m2, x2, x2);
+ x3 = _mm_permutexvar_epi16 (x3, x3);
+ x3 = _mm_maskz_permutexvar_epi16 (m3, x3, x3);
+ x3 = _mm_mask_permutexvar_epi16 (x3, m3, x3, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *ind, short *src, short *res)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ res[i] = src[ind[i] & (SIZE - 1)];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res1, res2, res3;
+ short res_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * i * i;
+ s2.a[i] = i + 20;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_permutexvar_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_permutexvar_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_permutexvar_epi16) (mask, s1.x, s2.x);
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w)(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w)(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmaddubsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaddubsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddubsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddubsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddubsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddubsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddubsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i xq;
+volatile __m128i xw;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_maddubs_epi16 (x, x);
+ x = _mm512_mask_maddubs_epi16 (x, 2, x, x);
+ x = _mm512_maskz_maddubs_epi16 (2, x, x);
+ xq = _mm256_mask_maddubs_epi16 (xq, 2, xq, xq);
+ xq = _mm256_maskz_maddubs_epi16 (2, xq, xq);
+ xw = _mm_mask_maddubs_epi16 (xw, 2, xw, xw);
+ xw = _mm_maskz_maddubs_epi16 (2, xw, xw);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#include <math.h>
+#include <values.h>
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *i1, short *i2, short *r)
+{
+ unsigned char *ub1 = (unsigned char *) i1;
+ char *sb2 = (char *) i2;
+ short *sout = (short *) r;
+ int t0;
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ t0 = ((int) ub1[2 * i] * (int) sb2[2 * i] +
+ (int) ub1[2 * i + 1] * (int) sb2[2 * i + 1]);
+ if (t0 > (int) 0x7fff)
+ sout[i] = 0x7fff;
+ else if (t0 < (int) 0xffff8000)
+ sout[i] = 0x8000;
+ else
+ sout[i] = (short) t0;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res1, res2, res3;
+ short res_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int i;
+ int fail = 0;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * 17 + i;
+ s2.a[i] = i * -17 + i * 2;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_maddubs_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_maddubs_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_maddubs_epi16) (mask, s1.x, s2.x);
+
+ CALC(s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmaddwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaddwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddwd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddwd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaddwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i xq;
+volatile __m128i xw;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_madd_epi16 (x, x);
+ x = _mm512_mask_madd_epi16 (x, 2, x, x);
+ x = _mm512_maskz_madd_epi16 (2, x, x);
+ xq = _mm256_mask_madd_epi16 (xq, 2, xq, xq);
+ xq = _mm256_maskz_madd_epi16 (2, xq, xq);
+ xw = _mm_mask_madd_epi16 (xw, 2, xw, xw);
+ xw = _mm_maskz_madd_epi16 (2, xw, xw);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#include <math.h>
+#include <values.h>
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *i1, short *i2, int *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = ((int) i1[2 * i] * (int) i2[2 * i] +
+ (int) i1[2 * i + 1] * (int) i2[2 * i + 1]);
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2;
+ UNION_TYPE (AVX512F_LEN, i_d) res1, res2, res3;
+ int res_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int i;
+
+ for (i = 0; i < SIZE * 2; i++)
+ {
+ s1.a[i] = i * 17 + i;
+ s2.a[i] = i * -17 + i * 2;
+ }
+
+ for (i = 0; i < SIZE; i++)
+ {
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_madd_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_madd_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_madd_epi16) (mask, s1.x, s2.x);
+
+ CALC(s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmaxsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaxsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_max_epi8 (x, x);
+ x = _mm512_mask_max_epi8 (x, mx, x, x);
+ x = _mm512_maskz_max_epi8 (mx, x, x);
+ y = _mm256_mask_max_epi8 (y, my, y, y);
+ y = _mm256_maskz_max_epi8 (my, y, y);
+ z = _mm_mask_max_epi8 (z, mz, z, z);
+ z = _mm_maskz_max_epi8 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *src1, char *src2, char *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] > src2[i] ? src1[i] : src2[i];
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * sign;
+ src2.a[i] = (i + 20) * sign;
+ sign = -sign;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_max_epi8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_max_epi8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_max_epi8) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmaxsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaxsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_max_epi16 (x, x);
+ x = _mm512_mask_max_epi16 (x, mx, x, x);
+ x = _mm512_maskz_max_epi16 (mx, x, x);
+ y = _mm256_mask_max_epi16 (y, my, y, y);
+ y = _mm256_maskz_max_epi16 (my, y, y);
+ z = _mm_mask_max_epi16 (z, mz, z, z);
+ z = _mm_maskz_max_epi16 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *src1, short *src2, short *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] > src2[i] ? src1[i] : src2[i];
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * sign;
+ src2.a[i] = (i + 20) * sign;
+ sign = -sign;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_max_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_max_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_max_epi16) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmaxub\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaxub\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxub\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxub\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxub\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxub\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxub\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_max_epu8 (x, x);
+ x = _mm512_mask_max_epu8 (x, mx, x, x);
+ x = _mm512_maskz_max_epu8 (mx, x, x);
+ y = _mm256_mask_max_epu8 (y, my, y, y);
+ y = _mm256_maskz_max_epu8 (my, y, y);
+ z = _mm_mask_max_epu8 (z, mz, z, z);
+ z = _mm_maskz_max_epu8 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned char *src1, unsigned char *src2,
+ unsigned char *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] > src2[i] ? src1[i] : src2[i];
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned char res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * i;
+ src2.a[i] = i + 20;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_max_epu8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_max_epu8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_max_epu8) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmaxuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaxuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxuw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxuw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxuw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxuw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_max_epu16 (x, x);
+ x = _mm512_mask_max_epu16 (x, mx, x, x);
+ x = _mm512_maskz_max_epu16 (mx, x, x);
+ y = _mm256_mask_max_epu16 (y, my, y, y);
+ y = _mm256_maskz_max_epu16 (my, y, y);
+ z = _mm_mask_max_epu16 (z, mz, z, z);
+ z = _mm_maskz_max_epu16 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned short *src1, unsigned short *src2,
+ unsigned short *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] > src2[i] ? src1[i] : src2[i];
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * i;
+ src2.a[i] = i + 20;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_max_epu16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_max_epu16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_max_epu16) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpminsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpminsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_min_epi8 (x, x);
+ x = _mm512_mask_min_epi8 (x, mx, x, x);
+ x = _mm512_maskz_min_epi8 (mx, x, x);
+ y = _mm256_mask_min_epi8 (y, my, y, y);
+ y = _mm256_maskz_min_epi8 (my, y, y);
+ z = _mm_mask_min_epi8 (z, mz, z, z);
+ z = _mm_maskz_min_epi8 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *src1, char *src2, char *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] < src2[i] ? src1[i] : src2[i];
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * sign;
+ src2.a[i] = (i + 20) * sign;
+ sign = -sign;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_min_epi8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_min_epi8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_min_epi8) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpminsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpminsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_min_epi16 (x, x);
+ x = _mm512_mask_min_epi16 (x, mx, x, x);
+ x = _mm512_maskz_min_epi16 (mx, x, x);
+ y = _mm256_mask_min_epi16 (y, my, y, y);
+ y = _mm256_maskz_min_epi16 (my, y, y);
+ z = _mm_mask_min_epi16 (z, mz, z, z);
+ z = _mm_maskz_min_epi16 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *src1, short *src2, short *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] < src2[i] ? src1[i] : src2[i];
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * sign;
+ src2.a[i] = (i + 20) * sign;
+ sign = -sign;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_min_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_min_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_min_epi16) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpminub\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpminub\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminub\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminub\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminub\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminub\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminub\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_min_epu8 (x, x);
+ x = _mm512_mask_min_epu8 (x, mx, x, x);
+ x = _mm512_maskz_min_epu8 (mx, x, x);
+ y = _mm256_mask_min_epu8 (y, my, y, y);
+ y = _mm256_maskz_min_epu8 (my, y, y);
+ z = _mm_mask_min_epu8 (z, mz, z, z);
+ z = _mm_maskz_min_epu8 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned char *src1, unsigned char *src2,
+ unsigned char *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] < src2[i] ? src1[i] : src2[i];
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned char res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * i;
+ src2.a[i] = i + 20;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_min_epu8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_min_epu8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_min_epu8) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpminuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpminuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminuw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminuw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminuw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminuw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask32 mx;
+volatile __mmask16 my;
+volatile __mmask8 mz;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_min_epu16 (x, x);
+ x = _mm512_mask_min_epu16 (x, mx, x, x);
+ x = _mm512_maskz_min_epu16 (mx, x, x);
+ y = _mm256_mask_min_epu16 (y, my, y, y);
+ y = _mm256_maskz_min_epu16 (my, y, y);
+ z = _mm_mask_min_epu16 (z, mz, z, z);
+ z = _mm_maskz_min_epu16 (mz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned short *src1, unsigned short *src2,
+ unsigned short *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] < src2[i] ? src1[i] : src2[i];
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * i;
+ src2.a[i] = i + 20;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_min_epu16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_min_epu16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_min_epu16) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpmovb2m\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+/* { dg-final { scan-assembler "vpmovb2m\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+/* { dg-final { scan-assembler "vpmovb2m\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask64 m64;
+volatile __mmask32 m32;
+volatile __mmask16 m16;
+
+void extern
+avx512bw_test (void)
+{
+ m16 = _mm_movepi8_mask (x128);
+ m32 = _mm256_movepi8_mask (x256);
+ m64 = _mm512_movepi8_mask (x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *r, char *s1)
+{
+ int i;
+ MASK_TYPE res = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] >> 7)
+ res = res | (one << i);
+
+ *r = res;
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) src;
+ MASK_TYPE res, res_ref = 0;
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src.a[i] = 2 * i * sign;
+ sign = sign * -1;
+ }
+
+ res = INTRINSIC (_movepi8_mask) (src.x);
+
+ CALC (&res_ref, src.a);
+
+ if (res_ref != res)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpmovm2b\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%xmm\[0-9\]" } } */
+/* { dg-final { scan-assembler "vpmovm2b\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%ymm\[0-9\]" } } */
+/* { dg-final { scan-assembler "vpmovm2b\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%zmm\[0-9\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask64 m64;
+volatile __mmask32 m32;
+volatile __mmask16 m16;
+
+void extern
+avx512bw_test (void)
+{
+ x128 = _mm_movm_epi8 (m16);
+ x256 = _mm256_movm_epi8 (m32);
+ x512 = _mm512_movm_epi8 (m64);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, MASK_TYPE s)
+{
+ int i;
+ char all_ones = 0xff;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = ((s >> i) & 1) ? all_ones : 0;
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_b) res, res_ref;
+ MASK_TYPE src = (MASK_TYPE) 0x1111abeffeec1234;
+
+ res.x = INTRINSIC (_movm_epi8) (src);
+
+ CALC (res_ref.a, src);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res, res_ref.a))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpmovm2w\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%xmm\[0-9\]" } } */
+/* { dg-final { scan-assembler "vpmovm2w\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%ymm\[0-9\]" } } */
+/* { dg-final { scan-assembler "vpmovm2w\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%zmm\[0-9\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask32 m32;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512bw_test (void)
+{
+ x128 = _mm_movm_epi16 (m8);
+ x256 = _mm256_movm_epi16 (m16);
+ x512 = _mm512_movm_epi16 (m32);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, MASK_TYPE s)
+{
+ int i;
+ short all_ones = 0xffff;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = ((s >> i) & 1) ? all_ones : 0;
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) res, res_ref;
+ MASK_TYPE src = (MASK_TYPE) 0x1111abc2;
+
+ res.x = INTRINSIC (_movm_epi16) (src);
+
+ CALC (res_ref.a, src);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res, res_ref.a))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __m512i u;
+volatile __mmask8 m1;
+volatile __mmask16 m2;
+volatile __mmask32 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm_cvtsepi16_epi8 (x);
+ z = _mm_mask_cvtsepi16_epi8 (z, m1, x);
+ z = _mm_maskz_cvtsepi16_epi8 (m1, x);
+ z = _mm256_cvtsepi16_epi8 (y);
+ z = _mm256_mask_cvtsepi16_epi8 (z, m2, y);
+ z = _mm256_maskz_cvtsepi16_epi8 (m2, y);
+ y = _mm512_cvtsepi16_epi8 (u);
+ y = _mm512_mask_cvtsepi16_epi8 (y, m3, u);
+ y = _mm512_maskz_cvtsepi16_epi8 (m3, u);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#define SIZE_HALF (AVX512F_LEN_HALF / 8)
+#include "avx512f-mask-type.h"
+#include <limits.h>
+
+void
+CALC (char *r, short *s)
+{
+ int i;
+ for (i = 0; i < SIZE_HALF; i++)
+ {
+ if (s[i] < CHAR_MIN)
+ r[i] = CHAR_MIN;
+ else if (s[i] > CHAR_MAX)
+ r[i] = CHAR_MAX;
+ else
+ r[i] = s[i];
+ r[i] = (i < SIZE) ? r[i] : 0;
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN_HALF, i_b) res1, res2, res3;
+ UNION_TYPE (AVX512F_LEN, i_w) src;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[32];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src.a[i] = 1 + 34 * i * sign;
+ sign = sign * -1;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_cvtsepi16_epi8) (src.x);
+ res2.x = INTRINSIC (_mask_cvtsepi16_epi8) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_cvtsepi16_epi8) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsxbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i res1;
+volatile __m256i s1, res2;
+volatile __m128i s2, res3;
+volatile __mmask32 m32;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512bw_test (void)
+{
+ res1 = _mm512_cvtepi8_epi16 (s1);
+
+ res1 = _mm512_mask_cvtepi8_epi16 (res1, m32, s1);
+ res2 = _mm256_mask_cvtepi8_epi16 (res2, m16, s2);
+ res3 = _mm_mask_cvtepi8_epi16 (res3, m8, s2);
+
+ res1 = _mm512_maskz_cvtepi8_epi16 (m32, s1);
+ res2 = _mm256_maskz_cvtepi8_epi16 (m16, s2);
+ res3 = _mm_maskz_cvtepi8_epi16 (m8, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SRC_SIZE (AVX512F_LEN_HALF / 8)
+#define DST_SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *s, short *r)
+{
+ int i;
+
+ for (i = 0; i < DST_SIZE; i++)
+ {
+ r[i] = (short) s[i];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN_HALF, i_b) s;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[DST_SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SRC_SIZE; i++)
+ {
+ s.a[i] = 8 * i * sign;
+ sign = -sign;
+ }
+
+ for (i = 0; i < DST_SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_cvtepi8_epi16) (s.x);
+ res2.x = INTRINSIC (_mask_cvtepi8_epi16) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvtepi8_epi16) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovuswb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __m512i u;
+volatile __mmask8 m1;
+volatile __mmask16 m2;
+volatile __mmask32 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm_cvtusepi16_epi8 (x);
+ z = _mm_mask_cvtusepi16_epi8 (z, m1, x);
+ z = _mm_maskz_cvtusepi16_epi8 (m1, x);
+ z = _mm256_cvtusepi16_epi8 (y);
+ z = _mm256_mask_cvtusepi16_epi8 (z, m2, y);
+ z = _mm256_maskz_cvtusepi16_epi8 (m2, y);
+ y = _mm512_cvtusepi16_epi8 (u);
+ y = _mm512_mask_cvtusepi16_epi8 (y, m3, u);
+ y = _mm512_maskz_cvtusepi16_epi8 (m3, u);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#define SIZE_HALF (AVX512F_LEN_HALF / 8)
+#include "avx512f-mask-type.h"
+#include <limits.h>
+
+void
+CALC (unsigned char *r, unsigned short *s)
+{
+ int i;
+ for (i = 0; i < SIZE_HALF; i++)
+ r[i] = (i < SIZE) ? ((s[i] > UCHAR_MAX) ? UCHAR_MAX : s[i]) : 0;
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN_HALF, i_b) res1, res2, res3;
+ UNION_TYPE (AVX512F_LEN, i_w) src;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned char res_ref[32];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src.a[i] = 1 + 34 * i;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_cvtusepi16_epi8) (src.x);
+ res2.x = INTRINSIC (_mask_cvtusepi16_epi8) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_cvtusepi16_epi8) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpmovw2m\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+/* { dg-final { scan-assembler "vpmovw2m\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+/* { dg-final { scan-assembler "vpmovw2m\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+
+void extern
+avx512bw_test (void)
+{
+ m8 = _mm_movepi16_mask (x128);
+ m16 = _mm256_movepi16_mask (x256);
+ m32 = _mm512_movepi16_mask (x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *r, short *s1)
+{
+ int i;
+ MASK_TYPE res = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] >> 15)
+ res = res | (one << i);
+
+ *r = res;
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) src;
+ MASK_TYPE res, res_ref = 0;
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src.a[i] = 2 * i * sign;
+ sign = sign * -1;
+ }
+
+ res = INTRINSIC (_movepi16_mask) (src.x);
+
+ CALC (&res_ref, src.a);
+
+ if (res_ref != res)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovwb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __m512i u;
+volatile __mmask8 m1;
+volatile __mmask16 m2;
+volatile __mmask32 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm_cvtepi16_epi8 (x);
+ z = _mm_mask_cvtepi16_epi8 (z, m1, x);
+ z = _mm_maskz_cvtepi16_epi8 (m1, x);
+ z = _mm256_cvtepi16_epi8 (y);
+ z = _mm256_mask_cvtepi16_epi8 (z, m2, y);
+ z = _mm256_maskz_cvtepi16_epi8 (m2, y);
+ y = _mm512_cvtepi16_epi8 (u);
+ y = _mm512_mask_cvtepi16_epi8 (y, m3, u);
+ y = _mm512_maskz_cvtepi16_epi8 (m3, u);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#define SIZE_HALF (AVX512F_LEN_HALF / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, short *s)
+{
+ int i;
+ for (i = 0; i < SIZE_HALF; i++)
+ {
+ r[i] = (i < SIZE) ? (char) s[i] : 0;
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN_HALF, i_b) res1, res2, res3;
+ UNION_TYPE (AVX512F_LEN, i_w) src;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[32];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src.a[i] = 1 + 34 * i * sign;
+ sign = sign * -1;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_cvtepi16_epi8) (src.x);
+ res2.x = INTRINSIC (_mask_cvtepi16_epi8) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_cvtepi16_epi8) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovzxbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i res1;
+volatile __m256i s1, res2;
+volatile __m128i s2, res3;
+volatile __mmask32 m32;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512bw_test (void)
+{
+ res1 = _mm512_cvtepu8_epi16 (s1);
+
+ res1 = _mm512_mask_cvtepu8_epi16 (res1, m32, s1);
+ res2 = _mm256_mask_cvtepu8_epi16 (res2, m16, s2);
+ res3 = _mm_mask_cvtepu8_epi16 (res3, m8, s2);
+
+ res1 = _mm512_maskz_cvtepu8_epi16 (m32, s1);
+ res2 = _mm256_maskz_cvtepu8_epi16 (m16, s2);
+ res3 = _mm_maskz_cvtepu8_epi16 (m8, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SRC_SIZE (AVX512F_LEN_HALF / 8)
+#define DST_SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned char *s, short *r)
+{
+ int i;
+
+ for (i = 0; i < DST_SIZE; i++)
+ {
+ r[i] = s[i];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN_HALF, i_b) s;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[DST_SIZE];
+ int i;
+
+ for (i = 0; i < SRC_SIZE; i++)
+ {
+ s.a[i] = 16 * i;
+ }
+
+ for (i = 0; i < DST_SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_cvtepu8_epi16) (s.x);
+ res2.x = INTRINSIC (_mask_cvtepu8_epi16) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvtepu8_epi16) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, DST_SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmulhrsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmulhrsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhrsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhrsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhrsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhrsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhrsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x, y, z;
+volatile __m256i xq, yq, zq;
+volatile __m128i xw, yw, zw;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_mulhrs_epi16 (y, z);
+ x = _mm512_mask_mulhrs_epi16 (x, 2, y, z);
+ x = _mm512_maskz_mulhrs_epi16 (2, y, z);
+ xq = _mm256_mask_mulhrs_epi16 (xq, 2, yq, zq);
+ xq = _mm256_maskz_mulhrs_epi16 (2, yq, zq);
+ xw = _mm_mask_mulhrs_epi16 (xw, 2, yw, zw);
+ xw = _mm_maskz_mulhrs_epi16 (2, yw, zw);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *src1, short *src2, short *dst)
+{
+ int i, t0;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ t0 = (((int) src1[i] * (int) src2[i]) >> 14) + 1;
+ dst[i] = (short) (t0 >> 1);
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2, dst1, dst2, dst3;
+ short dst_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int i, sign = -1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i % 2;
+ src2.a[i] = i * sign;
+ dst2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ dst1.x = INTRINSIC (_mulhrs_epi16) (src1.x, src2.x);
+ dst2.x =
+ INTRINSIC (_mask_mulhrs_epi16) (dst2.x, mask, src1.x, src2.x);
+ dst3.x = INTRINSIC (_maskz_mulhrs_epi16) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst3, dst_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmulhuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmulhuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhuw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhuw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhuw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhuw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhuw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x, y, z;
+volatile __m256i xq, yq, zq;
+volatile __m128i xw, yw, zw;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_mulhi_epu16 (y, z);
+ x = _mm512_mask_mulhi_epu16 (x, 2, y, z);
+ x = _mm512_maskz_mulhi_epu16 (2, y, z);
+ xq = _mm256_mask_mulhi_epu16 (xq, 2, yq, zq);
+ xq = _mm256_maskz_mulhi_epu16 (2, yq, zq);
+ xw = _mm_mask_mulhi_epu16 (xw, 2, yw, zw);
+ xw = _mm_maskz_mulhi_epu16 (2, yw, zw);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *src1, short *src2, short *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = (src1[i] * src2[i]) >> 16;
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2, dst1, dst2, dst3;
+ short dst_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int i, sign = -1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i % 2;
+ src2.a[i] = i * sign;
+ dst2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ dst1.x = INTRINSIC (_mulhi_epu16) (src1.x, src2.x);
+ dst2.x =
+ INTRINSIC (_mask_mulhi_epu16) (dst2.x, mask, src1.x, src2.x);
+ dst3.x = INTRINSIC (_maskz_mulhi_epu16) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst3, dst_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmulhw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmulhw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulhw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x, y, z;
+volatile __m256i xq, yq, zq;
+volatile __m128i xw, yw, zw;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_mulhi_epi16 (y, z);
+ x = _mm512_mask_mulhi_epi16 (x, 2, y, z);
+ x = _mm512_maskz_mulhi_epi16 (2, y, z);
+ xq = _mm256_mask_mulhi_epi16 (xq, 2, yq, zq);
+ xq = _mm256_maskz_mulhi_epi16 (2, yq, zq);
+ xw = _mm_mask_mulhi_epi16 (xw, 2, yw, zw);
+ xw = _mm_maskz_mulhi_epi16 (2, yw, zw);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *src1, short *src2, short *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = (src1[i] * src2[i]) >> 16;
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2, dst1, dst2, dst3;
+ short dst_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int i, sign = -1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i % 2;
+ src2.a[i] = i * sign;
+ dst2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ dst1.x = INTRINSIC (_mulhi_epi16) (src1.x, src2.x);
+ dst2.x =
+ INTRINSIC (_mask_mulhi_epi16) (dst2.x, mask, src1.x, src2.x);
+ dst3.x = INTRINSIC (_maskz_mulhi_epi16) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst3, dst_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmullw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmullw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x, y, z;
+volatile __m256i xq, yq, zq;
+volatile __m128i xw, yw, zw;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_mullo_epi16 (y, z);
+ x = _mm512_mask_mullo_epi16 (x, 3, y, z);
+ x = _mm512_maskz_mullo_epi16 (3, y, z);
+ xq = _mm256_mask_mullo_epi16 (xq, 3, yq, zq);
+ xq = _mm256_maskz_mullo_epi16 (3, yq, zq);
+ xw = _mm_mask_mullo_epi16 (xw, 3, yw, zw);
+ xw = _mm_maskz_mullo_epi16 (3, yw, zw);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *src1, short *src2, short *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = (short) ((int) src1[i] * (int) src2[i]);
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2, dst1, dst2, dst3;
+ short dst_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * sign;
+ src2.a[i] = i + 20 * sign;
+ sign = -sign;
+ dst2.a[i] = DEFAULT_VALUE;
+ }
+
+ dst1.x = INTRINSIC (_mullo_epi16) (src1.x, src2.x);
+ dst2.x = INTRINSIC (_mask_mullo_epi16) (dst2.x, mask, src1.x, src2.x);
+ dst3.x = INTRINSIC (_maskz_mullo_epi16) (mask, src1.x, src2.x);
+
+ CALC (src1.a, src2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (dst3, dst_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpshufb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask64 m1;
+volatile __mmask32 m2;
+volatile __mmask16 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_shuffle_epi8 (z, z);
+ z = _mm512_mask_shuffle_epi8 (z, m1, z, z);
+ z = _mm512_maskz_shuffle_epi8 (m1, z, z);
+ y = _mm256_mask_shuffle_epi8 (y, m2, y, y);
+ y = _mm256_maskz_shuffle_epi8 (m2, y, y);
+ x = _mm_mask_shuffle_epi8 (x, m3, x, x);
+ x = _mm_maskz_shuffle_epi8 (m3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *s1, char *s2, char *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ if (s2[i] < 0)
+ r[i] = 0;
+ else
+ r[i] = s1[(s2[i] & 0xf) + 16 * (i / 16)];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_b) s1, s2, res1, res2, res3;
+ char res_ref[SIZE];
+ int i, sign = 1;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * i * sign;
+ s2.a[i] = 179 - i;
+ res1.a[i] = DEFAULT_VALUE;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_shuffle_epi8) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_shuffle_epi8) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_shuffle_epi8) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpshufhw\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpshufhw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufhw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufhw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufhw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufhw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufhw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_shufflehi_epi16 (z, _MM_PERM_AADB);
+ z = _mm512_mask_shufflehi_epi16 (z, m1, z, _MM_PERM_AADB);
+ z = _mm512_maskz_shufflehi_epi16 (m1, z, _MM_PERM_AADB);
+ y = _mm256_mask_shufflehi_epi16 (y, m2, y, _MM_PERM_AADB);
+ y = _mm256_maskz_shufflehi_epi16 (m2, y, _MM_PERM_AADB);
+ x = _mm_mask_shufflehi_epi16 (x, m3, x, _MM_PERM_AADB);
+ x = _mm_maskz_shufflehi_epi16 (m3, x, _MM_PERM_AADB);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *s, unsigned char imm, short *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE / 8; i++)
+ {
+ r[8 * i] = s[8 * i];
+ r[8 * i + 1] = s[8 * i + 1];
+ r[8 * i + 2] = s[8 * i + 2];
+ r[8 * i + 3] = s[8 * i + 3];
+ r[8 * i + 4] = s[8 * i + (imm >> 0 & 3) + 4];
+ r[8 * i + 5] = s[8 * i + (imm >> 2 & 3) + 4];
+ r[8 * i + 6] = s[8 * i + (imm >> 4 & 3) + 4];
+ r[8 * i + 7] = s[8 * i + (imm >> 6 & 3) + 4];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, res1, res2, res3;
+ short res_ref[SIZE];
+ int i, sign = 1;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * i * sign;
+ res1.a[i] = DEFAULT_VALUE;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_shufflehi_epi16) (s1.x, 0xec);
+ res2.x =
+ INTRINSIC (_mask_shufflehi_epi16) (res2.x, mask, s1.x, 0xec);
+ res3.x = INTRINSIC (_maskz_shufflehi_epi16) (mask, s1.x, 0xec);
+
+ CALC (s1.a, 0xec, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpshuflw\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpshuflw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshuflw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpshuflw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshuflw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpshuflw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshuflw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_shufflelo_epi16 (z, _MM_PERM_AADB);
+ z = _mm512_mask_shufflelo_epi16 (z, m1, z, _MM_PERM_AADB);
+ z = _mm512_maskz_shufflelo_epi16 (m1, z, _MM_PERM_AADB);
+ y = _mm256_mask_shufflelo_epi16 (y, m2, y, _MM_PERM_AADB);
+ y = _mm256_maskz_shufflelo_epi16 (m2, y, _MM_PERM_AADB);
+ x = _mm_mask_shufflelo_epi16 (x, m3, x, _MM_PERM_AADB);
+ x = _mm_maskz_shufflelo_epi16 (m3, x, _MM_PERM_AADB);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *s, unsigned char imm, short *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE / 8; i++)
+ {
+ r[8 * i] = s[8 * i + (imm >> 0 & 3)];
+ r[8 * i + 1] = s[8 * i + (imm >> 2 & 3)];
+ r[8 * i + 2] = s[8 * i + (imm >> 4 & 3)];
+ r[8 * i + 3] = s[8 * i + (imm >> 6 & 3)];
+ r[8 * i + 4] = s[8 * i + 4];
+ r[8 * i + 5] = s[8 * i + 5];
+ r[8 * i + 6] = s[8 * i + 6];
+ r[8 * i + 7] = s[8 * i + 7];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, res1, res2, res3;
+ short res_ref[SIZE];
+ int i, sign = 1;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * i * sign;
+ res1.a[i] = DEFAULT_VALUE;
+ res2.a[i] = DEFAULT_VALUE;
+ res3.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_shufflelo_epi16) (s1.x, 0xec);
+ res2.x =
+ INTRINSIC (_mask_shufflelo_epi16) (res2.x, mask, s1.x, 0xec);
+ res3.x = INTRINSIC (_maskz_shufflelo_epi16) (mask, s1.x, 0xec);
+
+ CALC (s1.a, 0xec, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -O2" } */
+/* { dg-final { scan-assembler-times "vpslldq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+extern volatile __m512i x;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_bslli_epi128 (x, 13);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_sllv_epi16 (z, z);
+ z = _mm512_mask_sllv_epi16 (z, m1, z, z);
+ z = _mm512_maskz_sllv_epi16 (m1, z, z);
+ y = _mm256_sllv_epi16 (y, y);
+ y = _mm256_mask_sllv_epi16 (y, m2, y, y);
+ y = _mm256_maskz_sllv_epi16 (m2, y, y);
+ x = _mm_sllv_epi16 (x, x);
+ x = _mm_mask_sllv_epi16 (x, m3, x, x);
+ x = _mm_maskz_sllv_epi16 (m3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *s1, short *s2, short *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; ++i)
+ {
+ r[i] = ((unsigned short) s1[i]) << s2[i];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * sign;
+ s2.a[i] = i >> 2;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_sllv_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_sllv_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_sllv_epi16) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 7 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+volatile __m512i x;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __m128i y;
+volatile __mmask32 m;
+volatile __mmask16 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_sll_epi16 (x, y);
+ x = _mm512_mask_sll_epi16 (x, m, x, y);
+ x = _mm512_maskz_sll_epi16 (m, x, y);
+ x256 = _mm256_mask_sll_epi16 (x256, m256, x256, y);
+ x256 = _mm256_maskz_sll_epi16 (m256, x256, y);
+ x128 = _mm_mask_sll_epi16 (x128, m128, x128, y);
+ x128 = _mm_maskz_sll_epi16 (m128, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, long long *s2)
+{
+ int i;
+ long long count = s2[0];
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = count < 16 ? (s1[i] << count) : 0;
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1;
+ UNION_TYPE (128, i_q) src2;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ long long imm;
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ sign = sign * -1;
+ }
+ for (i = 0; i < 2; i++)
+ {
+ src2.a[i] = 0;
+ }
+
+ for (imm = 1; imm <= 17; imm++)
+ {
+ src2.a[0] = imm;
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_sll_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_sll_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_sll_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+ }
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+volatile __m512i x;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask32 m;
+volatile __mmask16 m256;
+volatile __mmask8 m128;
+#define y 7
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_slli_epi16 (x, y);
+ x = _mm512_mask_slli_epi16 (x, m, x, y);
+ x = _mm512_maskz_slli_epi16 (m, x, y);
+ x256 = _mm256_mask_slli_epi16 (x256, m256, x256, y);
+ x256 = _mm256_maskz_slli_epi16 (m256, x256, y);
+ x128 = _mm_mask_slli_epi16 (x128, m128, x128, y);
+ x128 = _mm_maskz_slli_epi16 (m128, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, short count)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = count < 16 ? (s1[i] << count) : 0;
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ sign = sign * -1;
+ }
+
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_slli_epi16) (src1.x, 5);
+ res2.x = INTRINSIC (_mask_slli_epi16) (res2.x, mask, src1.x, 5);
+ res3.x = INTRINSIC (_maskz_slli_epi16) (mask, src1.x, 5);
+
+ CALC (res_ref, src1.a, 5);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+
+
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_slli_epi16) (src1.x, 17);
+ res2.x = INTRINSIC (_mask_slli_epi16) (res2.x, mask, src1.x, 17);
+ res3.x = INTRINSIC (_maskz_slli_epi16) (mask, src1.x, 17);
+
+ CALC (res_ref, src1.a, 17);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_srav_epi16 (z, z);
+ z = _mm512_mask_srav_epi16 (z, m1, z, z);
+ z = _mm512_maskz_srav_epi16 (m1, z, z);
+ y = _mm256_srav_epi16 (y, y);
+ y = _mm256_mask_srav_epi16 (y, m2, y, y);
+ y = _mm256_maskz_srav_epi16 (m2, y, y);
+ x = _mm_srav_epi16 (x, x);
+ x = _mm_mask_srav_epi16 (x, m3, x, x);
+ x = _mm_maskz_srav_epi16 (m3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *s1, short *s2, short *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; ++i)
+ {
+ r[i] = s2[i] < 16 ? (s1[i] >> s2[i]) : (s1[i] > 0 ? 0 : 0xFFFF);
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * sign;
+ s2.a[i] = i >> 2;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_srav_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_srav_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_srav_epi16) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_sra_epi16 (z, x);
+ z = _mm512_mask_sra_epi16 (z, m1, z, x);
+ z = _mm512_maskz_sra_epi16 (m1, z, x);
+ y = _mm256_mask_sra_epi16 (y, m2, y, x);
+ y = _mm256_maskz_sra_epi16 (m2, y, x);
+ x = _mm_mask_sra_epi16 (x, m3, x, x);
+ x = _mm_maskz_sra_epi16 (m3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+#include <string.h>
+
+#define N 0x5
+
+void
+CALC (short *s1, long long int *s2, short *r)
+{
+ int i;
+ long long int count = s2[0];
+
+ memset (r, 0, SIZE);
+
+ if (count < 16)
+ for (i = 0; i < SIZE; ++i)
+ r[i] = s1[i] >> count;
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ union128i_q s2;
+ short res_ref[SIZE];
+ int i, sign;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * sign;
+ sign = -sign;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ s2.a[0] = N;
+
+ res1.x = INTRINSIC (_sra_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_sra_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_sra_epi16) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*13\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*13\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*13\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*13\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraw\[ \\t\]+\[^\n\]*13\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_srai_epi16 (z, 13);
+ z = _mm512_mask_srai_epi16 (z, m1, z, 13);
+ z = _mm512_maskz_srai_epi16 (m1, z, 13);
+ y = _mm256_mask_srai_epi16 (y, m2, y, 13);
+ y = _mm256_maskz_srai_epi16 (m2, y, 13);
+ x = _mm_mask_srai_epi16 (x, m3, x, 13);
+ x = _mm_maskz_srai_epi16 (m3, x, 13);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+#include <string.h>
+
+#define N 0x5
+
+void
+CALC (short *s1, short *r)
+{
+ int i;
+
+ memset (r, 0, SIZE);
+
+ if (N < 16)
+ for (i = 0; i < SIZE; ++i)
+ r[i] = s1[i] >> N;
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ union128i_q s2;
+ short res_ref[SIZE];
+ int i, sign;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * sign;
+ sign = -sign;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_srai_epi16) (s1.x, N);
+ res2.x = INTRINSIC (_mask_srai_epi16) (res2.x, mask, s1.x, N);
+ res3.x = INTRINSIC (_maskz_srai_epi16) (mask, s1.x, N);
+
+ CALC (s1.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -O2" } */
+/* { dg-final { scan-assembler-times "vpsrldq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+extern volatile __m512i x;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm512_bsrli_epi128 (x, 13);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_srlv_epi16 (z, z);
+ z = _mm512_mask_srlv_epi16 (z, m1, z, z);
+ z = _mm512_maskz_srlv_epi16 (m1, z, z);
+ y = _mm256_srlv_epi16 (y, y);
+ y = _mm256_mask_srlv_epi16 (y, m2, y, y);
+ y = _mm256_maskz_srlv_epi16 (m2, y, y);
+ x = _mm_srlv_epi16 (x, x);
+ x = _mm_mask_srlv_epi16 (x, m3, x, x);
+ x = _mm_maskz_srlv_epi16 (m3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *s1, short *s2, short *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; ++i)
+ {
+ r[i] = ((unsigned short) s1[i]) >> s2[i];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_w) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = i * sign;
+ s2.a[i] = i >> 2;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_srlv_epi16) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_srlv_epi16) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_srlv_epi16) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_srl_epi16 (z, x);
+ z = _mm512_mask_srl_epi16 (z, m1, z, x);
+ z = _mm512_maskz_srl_epi16 (m1, z, x);
+ y = _mm256_mask_srl_epi16 (y, m2, y, x);
+ y = _mm256_maskz_srl_epi16 (m2, y, x);
+ x = _mm_mask_srl_epi16 (x, m3, x, x);
+ x = _mm_maskz_srl_epi16 (m3, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned short *r, unsigned short *s1, unsigned short *s2)
+{
+ int i;
+ unsigned short count = s2[0];
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = count < 16 ? (s1[i] >> count) : 0;
+ }
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1;
+ UNION_TYPE (128, i_w) src2;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + 7 * i % 291;
+ }
+
+ for (i = 0; i < 128 / 16; i++)
+ {
+ src2.a[i] = 0;
+ }
+
+ src2.a[0] = 1;
+ src2.a[1] = 0;
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_srl_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_srl_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_srl_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+
+ src2.a[0] = 17;
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_srl_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_srl_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_srl_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*13\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*13\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*13\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*13\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlw\[ \\t\]+\[^\n\]*13\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ z = _mm512_srli_epi16 (z, 13);
+ z = _mm512_mask_srli_epi16 (z, m1, z, 13);
+ z = _mm512_maskz_srli_epi16 (m1, z, 13);
+ y = _mm256_mask_srli_epi16 (y, m2, y, 13);
+ y = _mm256_maskz_srli_epi16 (m2, y, 13);
+ x = _mm_mask_srli_epi16 (x, m3, x, 13);
+ x = _mm_maskz_srli_epi16 (m3, x, 13);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned short *r, unsigned short *s1, unsigned short count)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = count < 16 ? (s1[i] >> count) : 0;
+ }
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + 7 * i % 291;
+ }
+
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_srli_epi16) (src1.x, 5);
+ res2.x = INTRINSIC (_mask_srli_epi16) (res2.x, mask, src1.x, 5);
+ res3.x = INTRINSIC (_maskz_srli_epi16) (mask, src1.x, 5);
+
+ CALC (res_ref, src1.a, 5);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_srli_epi16) (src1.x, 17);
+ res2.x = INTRINSIC (_mask_srli_epi16) (res2.x, mask, src1.x, 17);
+ res3.x = INTRINSIC (_maskz_srli_epi16) (mask, src1.x, 17);
+
+ CALC (res_ref, src1.a, 17);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask64 m512;
+volatile __mmask32 m256;
+volatile __mmask16 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_sub_epi8 (x512, x512);
+ x512 = _mm512_mask_sub_epi8 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_sub_epi8 (m512, x512, x512);
+ x256 = _mm256_mask_sub_epi8 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_sub_epi8 (m256, x256, x256);
+ x128 = _mm_mask_sub_epi8 (x128, m128, x128, x128);
+ x128 = _mm_maskz_sub_epi8 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, char *s1, char *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s1[i] - s2[i];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ src2.a[i] = 3 + sign * 11 * (i % 377) * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_sub_epi8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_sub_epi8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_sub_epi8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask64 m512;
+volatile __mmask32 m256;
+volatile __mmask16 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_subs_epi8 (x512, x512);
+ x512 = _mm512_mask_subs_epi8 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_subs_epi8 (m512, x512, x512);
+ x256 = _mm256_mask_subs_epi8 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_subs_epi8 (m256, x256, x256);
+ x128 = _mm_mask_subs_epi8 (x128, m128, x128, x128);
+ x128 = _mm_maskz_subs_epi8 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, char *s1, char *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (int)s1[i] - (int)s2[i];
+ if (tmp > 0x7F) tmp = 0x7F;
+ if (tmp < (char)0x80) tmp = (char)0x80;
+ r[i] = (char)tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ src2.a[i] = 3 + sign * 11 * (i % 377) * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_subs_epi8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_subs_epi8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_subs_epi8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubsw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask32 m512;
+volatile __mmask16 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_subs_epi16 (x512, x512);
+ x512 = _mm512_mask_subs_epi16 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_subs_epi16 (m512, x512, x512);
+ x256 = _mm256_mask_subs_epi16 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_subs_epi16 (m256, x256, x256);
+ x128 = _mm_mask_subs_epi16 (x128, m128, x128, x128);
+ x128 = _mm_maskz_subs_epi16 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, short *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (int)s1[i] - (int)s2[i];
+ if (tmp > 0x7FFF) tmp = 0x7FFF;
+ if (tmp < (short)0x8000) tmp = (short)0x8000;
+ r[i] = (short)tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ src2.a[i] = 3 + sign * 11 * (i % 377) * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_subs_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_subs_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_subs_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask64 m512;
+volatile __mmask32 m256;
+volatile __mmask16 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_subs_epu8 (x512, x512);
+ x512 = _mm512_mask_subs_epu8 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_subs_epu8 (m512, x512, x512);
+ x256 = _mm256_mask_subs_epu8 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_subs_epu8 (m256, x256, x256);
+ x128 = _mm_mask_subs_epu8 (x128, m128, x128, x128);
+ x128 = _mm_maskz_subs_epu8 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned char *r, unsigned char *s1, unsigned char *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (int)s1[i] - (int)s2[i];
+ if (tmp > 0xFF) tmp = 0xFF;
+ if (tmp < 0) tmp = 0;
+ r[i] = (unsigned char)tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned char res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + 7 * i % 291;
+ src2.a[i] = 3 + 11 * (i % 377) * i;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_subs_epu8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_subs_epu8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_subs_epu8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubusw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask32 m512;
+volatile __mmask16 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_subs_epu16 (x512, x512);
+ x512 = _mm512_mask_subs_epu16 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_subs_epu16 (m512, x512, x512);
+ x256 = _mm256_mask_subs_epu16 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_subs_epu16 (m256, x256, x256);
+ x128 = _mm_mask_subs_epu16 (x128, m128, x128, x128);
+ x128 = _mm_maskz_subs_epu16 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned short *r, unsigned short *s1, unsigned short *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (int)s1[i] - (int)s2[i];
+ if (tmp > 0xFFFF) tmp = 0xFFFF;
+ if (tmp < 0) tmp = 0;
+ r[i] = (unsigned short)tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned short res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + 7 * i % 291;
+ src2.a[i] = 3 + 11 * (i % 377) * i;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_subs_epu16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_subs_epu16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_subs_epu16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask32 m512;
+volatile __mmask16 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512bw_test (void)
+{
+ x512 = _mm512_sub_epi16 (x512, x512);
+ x512 = _mm512_mask_sub_epi16 (x512, m512, x512, x512);
+ x512 = _mm512_maskz_sub_epi16 (m512, x512, x512);
+ x256 = _mm256_mask_sub_epi16 (x256, m256, x256, x256);
+ x256 = _mm256_maskz_sub_epi16 (m256, x256, x256);
+ x128 = _mm_mask_sub_epi16 (x128, m128, x128, x128);
+ x128 = _mm_maskz_sub_epi16 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, short *s2)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s1[i] - s2[i];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 2 + sign * 7 * i % 291;
+ src2.a[i] = 3 + sign * 11 * (i % 377) * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_sub_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_sub_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_sub_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vptestmb\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestmb\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestmb\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestmb\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestmb\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestmb\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __m512i x512;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+volatile __mmask64 m64;
+
+void extern
+avx512bw_test (void)
+{
+ m16 = _mm_test_epi8_mask (x128, x128);
+ m32 = _mm256_test_epi8_mask (x256, x256);
+ m64 = _mm512_test_epi8_mask (x512, x512);
+ m16 = _mm_mask_test_epi8_mask (3, x128, x128);
+ m32 = _mm256_mask_test_epi8_mask (3, x256, x256);
+ m64 = _mm512_mask_test_epi8_mask (3, x512, x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *res, char *src1, char *src2)
+{
+ int i;
+ *res = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (src1[i] & src2[i])
+ *res = *res | one << i;
+}
+
+void
+TEST (void)
+{
+ int i, sign = 1;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * i * sign;
+ src2.a[i] = i + 20;
+ sign = -sign;
+ }
+
+ res1 = INTRINSIC (_test_epi8_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_test_epi8_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res1 != res_ref)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res2 != res_ref)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vptestmw\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestmw\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestmw\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestmw\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestmw\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestmw\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __m512i x512;
+volatile __mmask8 m8;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+
+void extern
+avx512bw_test (void)
+{
+ m8 = _mm_test_epi16_mask (x128, x128);
+ m16 = _mm256_test_epi16_mask (x256, x256);
+ m32 = _mm512_test_epi16_mask (x512, x512);
+ m8 = _mm_mask_test_epi16_mask (3, x128, x128);
+ m16 = _mm256_mask_test_epi16_mask (3, x256, x256);
+ m32 = _mm512_mask_test_epi16_mask (3, x512, x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *res, short *src1, short *src2)
+{
+ int i;
+ *res = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (src1[i] & src2[i])
+ *res = *res | one << i;
+}
+
+void
+TEST (void)
+{
+ int i, sign = 1;
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * i * sign;
+ src2.a[i] = i + 20;
+ sign = -sign;
+ }
+
+ res1 = INTRINSIC (_test_epi16_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_test_epi16_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res1 != res_ref)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res2 != res_ref)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vptestnmb\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmb\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmb\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmb\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestnmb\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestnmb\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __m512i x512;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+volatile __mmask64 m64;
+
+void extern
+avx512bw_test (void)
+{
+ m16 = _mm_testn_epi8_mask (x128, x128);
+ m32 = _mm256_testn_epi8_mask (x256, x256);
+ m64 = _mm512_testn_epi8_mask (x512, x512);
+ m16 = _mm_mask_testn_epi8_mask (3, x128, x128);
+ m32 = _mm256_mask_testn_epi8_mask (3, x256, x256);
+ m64 = _mm512_mask_testn_epi8_mask (3, x512, x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *res, char *src1, char *src2)
+{
+ int i;
+ *res = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (!(src1[i] & src2[i]))
+ *res = *res | one << i;
+}
+
+void
+TEST (void)
+{
+ int i, sign = 1;
+ UNION_TYPE (AVX512F_LEN, i_b) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * i * sign;
+ src2.a[i] = i + 20;
+ sign = -sign;
+ }
+
+ res1 = INTRINSIC (_testn_epi8_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_testn_epi8_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res1 != res_ref)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res2 != res_ref)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vptestnmw\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmw\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmw\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmw\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestnmw\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestnmw\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __m512i x512;
+volatile __mmask8 m8;
+volatile __mmask16 m16;
+volatile __mmask32 m32;
+
+void extern
+avx512bw_test (void)
+{
+ m8 = _mm_testn_epi16_mask (x128, x128);
+ m16 = _mm256_testn_epi16_mask (x256, x256);
+ m32 = _mm512_testn_epi16_mask (x512, x512);
+ m8 = _mm_mask_testn_epi16_mask (3, x128, x128);
+ m16 = _mm256_mask_testn_epi16_mask (3, x256, x256);
+ m32 = _mm512_mask_testn_epi16_mask (3, x512, x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *res, short *src1, short *src2)
+{
+ int i;
+ *res = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (!(src1[i] & src2[i]))
+ *res = *res | one << i;
+}
+
+void
+TEST (void)
+{
+ int i, sign = 1;
+ UNION_TYPE (AVX512F_LEN, i_w) src1, src2;
+ MASK_TYPE res_ref, res1, res2;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i * i * sign;
+ src2.a[i] = i + 20;
+ sign = -sign;
+ }
+
+ res1 = INTRINSIC (_testn_epi16_mask) (src1.x, src2.x);
+ res2 = INTRINSIC (_mask_testn_epi16_mask) (mask, src1.x, src2.x);
+
+ CALC (&res_ref, src1.a, src2.a);
+
+ if (res1 != res_ref)
+ abort ();
+
+ res_ref &= mask;
+
+ if (res2 != res_ref)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpunpckhbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i d, e, f;
+volatile __m256i x, y, z;
+volatile __m128i a, b, c;
+volatile __mmask64 m1;
+volatile __mmask32 m2;
+volatile __mmask16 m3;
+
+void extern
+avx512bw_test (void)
+{
+ d = _mm512_unpackhi_epi8 (e, f);
+ d = _mm512_mask_unpackhi_epi8 (d, m1, e, f);
+ d = _mm512_maskz_unpackhi_epi8 (m1, e, f);
+ x = _mm256_mask_unpackhi_epi8 (x, m2, y, z);
+ x = _mm256_maskz_unpackhi_epi8 (m2, y, z);
+ a = _mm_mask_unpackhi_epi8 (a, m3, b, c);
+ a = _mm_maskz_unpackhi_epi8 (m3, b, c);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, char *s1, char *s2)
+{
+ int i;
+ for (i = 0; i < SIZE/16; i++)
+ {
+ r[16 * i] = s1[16 * i + 8];
+ r[16 * i + 1] = s2[16 * i + 8];
+ r[16 * i + 2] = s1[16 * i + 9];
+ r[16 * i + 3] = s2[16 * i + 9];
+ r[16 * i + 4] = s1[16 * i + 10];
+ r[16 * i + 5] = s2[16 * i + 10];
+ r[16 * i + 6] = s1[16 * i + 11];
+ r[16 * i + 7] = s2[16 * i + 11];
+ r[16 * i + 8] = s1[16 * i + 12];
+ r[16 * i + 9] = s2[16 * i + 12];
+ r[16 * i + 10] = s1[16 * i + 13];
+ r[16 * i + 11] = s2[16 * i + 13];
+ r[16 * i + 12] = s1[16 * i + 14];
+ r[16 * i + 13] = s2[16 * i + 14];
+ r[16 * i + 14] = s1[16 * i + 15];
+ r[16 * i + 15] = s2[16 * i + 15];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 34 * i * sign;
+ src1.a[i] = 179 * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_unpackhi_epi8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_unpackhi_epi8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_unpackhi_epi8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpunpckhwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhwd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhwd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i d, e, f;
+volatile __m256i x, y, z;
+volatile __m128i a, b, c;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ d = _mm512_unpackhi_epi16 (e, f);
+ d = _mm512_mask_unpackhi_epi16 (d, m1, e, f);
+ d = _mm512_maskz_unpackhi_epi16 (m1, e, f);
+ x = _mm256_mask_unpackhi_epi16 (x, m2, y, z);
+ x = _mm256_maskz_unpackhi_epi16 (m2, y, z);
+ a = _mm_mask_unpackhi_epi16 (a, m3, b, c);
+ a = _mm_maskz_unpackhi_epi16 (m3, b, c);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, short *s2)
+{
+ int i;
+ for (i = 0; i < SIZE/8; i++)
+ {
+ r[8 * i] = s1[8 * i + 4];
+ r[8 * i + 1] = s2[8 * i + 4];
+ r[8 * i + 2] = s1[8 * i + 5];
+ r[8 * i + 3] = s2[8 * i + 5];
+ r[8 * i + 4] = s1[8 * i + 6];
+ r[8 * i + 5] = s2[8 * i + 6];
+ r[8 * i + 6] = s1[8 * i + 7];
+ r[8 * i + 7] = s2[8 * i + 7];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 34 * i * sign;
+ src1.a[i] = 179 * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_unpackhi_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_unpackhi_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_unpackhi_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpunpcklbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklbw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklbw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i d, e, f;
+volatile __m256i x, y, z;
+volatile __m128i a, b, c;
+volatile __mmask64 m1;
+volatile __mmask32 m2;
+volatile __mmask16 m3;
+
+void extern
+avx512bw_test (void)
+{
+ d = _mm512_unpacklo_epi8 (e, f);
+ d = _mm512_mask_unpacklo_epi8 (d, m1, e, f);
+ d = _mm512_maskz_unpacklo_epi8 (m1, e, f);
+ x = _mm256_mask_unpacklo_epi8 (x, m2, y, z);
+ x = _mm256_maskz_unpacklo_epi8 (m2, y, z);
+ a = _mm_mask_unpacklo_epi8 (a, m3, b, c);
+ a = _mm_maskz_unpacklo_epi8 (m3, b, c);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 8)
+#include "avx512f-mask-type.h"
+
+void
+CALC (char *r, char *s1, char *s2)
+{
+ int i;
+ for (i = 0; i < SIZE/16; i++)
+ {
+ r[16 * i] = s1[16 * i];
+ r[16 * i + 1] = s2[16 * i];
+ r[16 * i + 2] = s1[16 * i + 1];
+ r[16 * i + 3] = s2[16 * i + 1];
+ r[16 * i + 4] = s1[16 * i + 2];
+ r[16 * i + 5] = s2[16 * i + 2];
+ r[16 * i + 6] = s1[16 * i + 3];
+ r[16 * i + 7] = s2[16 * i + 3];
+ r[16 * i + 8] = s1[16 * i + 4];
+ r[16 * i + 9] = s2[16 * i + 4];
+ r[16 * i + 10] = s1[16 * i + 5];
+ r[16 * i + 11] = s2[16 * i + 5];
+ r[16 * i + 12] = s1[16 * i + 6];
+ r[16 * i + 13] = s2[16 * i + 6];
+ r[16 * i + 14] = s1[16 * i + 7];
+ r[16 * i + 15] = s2[16 * i + 7];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_b) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ char res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 34 * i * sign;
+ src1.a[i] = 179 * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_unpacklo_epi8) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_unpacklo_epi8) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_unpacklo_epi8) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_b) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpunpcklwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklwd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklwd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklwd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i d, e, f;
+volatile __m256i x, y, z;
+volatile __m128i a, b, c;
+volatile __mmask32 m1;
+volatile __mmask16 m2;
+volatile __mmask8 m3;
+
+void extern
+avx512bw_test (void)
+{
+ d = _mm512_unpacklo_epi16 (e, f);
+ d = _mm512_mask_unpacklo_epi16 (d, m1, e, f);
+ d = _mm512_maskz_unpacklo_epi16 (m1, e, f);
+ x = _mm256_mask_unpacklo_epi16 (x, m2, y, z);
+ x = _mm256_maskz_unpacklo_epi16 (m2, y, z);
+ a = _mm_mask_unpacklo_epi16 (a, m3, b, c);
+ a = _mm_maskz_unpacklo_epi16 (m3, b, c);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -DAVX512BW" } */
+/* { dg-require-effective-target avx512bw } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 16)
+#include "avx512f-mask-type.h"
+
+void
+CALC (short *r, short *s1, short *s2)
+{
+ int i;
+ for (i = 0; i < SIZE/8; i++)
+ {
+ r[8 * i] = s1[8 * i];
+ r[8 * i + 1] = s2[8 * i];
+ r[8 * i + 2] = s1[8 * i + 1];
+ r[8 * i + 3] = s2[8 * i + 1];
+ r[8 * i + 4] = s1[8 * i + 2];
+ r[8 * i + 5] = s2[8 * i + 2];
+ r[8 * i + 6] = s1[8 * i + 3];
+ r[8 * i + 7] = s2[8 * i + 3];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_w) res1, res2, res3, src1, src2;
+ MASK_TYPE mask = MASK_VALUE;
+ short res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = 34 * i * sign;
+ src1.a[i] = 179 * i;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_unpacklo_epi16) (src1.x, src2.x);
+ res2.x = INTRINSIC (_mask_unpacklo_epi16) (res2.x, mask, src1.x, src2.x);
+ res3.x = INTRINSIC (_maskz_unpacklo_epi16) (mask, src1.x, src2.x);
+
+ CALC (res_ref, src1.a, src2.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_w) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_w) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+#include <stdlib.h>
+#include "cpuid.h"
+#include "m512-check.h"
+#include "avx512f-os-support.h"
+
+static void avx512dq_test (void);
+
+static void __attribute__ ((noinline)) do_test (void)
+{
+ avx512dq_test ();
+}
+
+int
+main ()
+{
+ unsigned int eax, ebx, ecx, edx;
+
+ if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx))
+ return 0;
+
+ /* Run AVX512DQ test only if host has AVX512DQ support. */
+ if ((ecx & bit_OSXSAVE) == (bit_OSXSAVE))
+ {
+ if (__get_cpuid_max (0, NULL) < 7)
+ return 0;
+
+ __cpuid_count (7, 0, eax, ebx, ecx, edx);
+
+ if ((avx512f_os_support ()) && ((ebx & bit_AVX512DQ) == bit_AVX512DQ))
+ {
+ do_test ();
+#ifdef DEBUG
+ printf ("PASSED\n");
+#endif
+ return 0;
+ }
+#ifdef DEBUG
+ printf ("SKIPPED\n");
+#endif
+ }
+#ifdef DEBUG
+ else
+ printf ("SKIPPED\n");
+#endif
+
+ return 0;
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandnpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d z;
+volatile __m256d y;
+volatile __m128d x;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_andnot_pd (z, z);
+ z = _mm512_mask_andnot_pd (z, m, z, z);
+ z = _mm512_maskz_andnot_pd (m, z, z);
+ y = _mm256_mask_andnot_pd (y, m, y, y);
+ y = _mm256_maskz_andnot_pd (m, y, y);
+ x = _mm_mask_andnot_pd (x, m, x, x);
+ x = _mm_maskz_andnot_pd (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *s1, double *s2, double *r)
+{
+ int i;
+ long long tmp;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ tmp = (~(*(long long *) &s1[i])) & (*(long long *) &s2[i]);
+ r[i] = *(double *) &tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ double res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = 13. * i;
+ s2.a[i] = 17. * i;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_andnot_pd) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_andnot_pd) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_andnot_pd) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandnps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 z;
+volatile __m256 y;
+volatile __m128 x;
+volatile __mmask16 m1;
+volatile __mmask8 m2;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_andnot_ps (z, z);
+ z = _mm512_mask_andnot_ps (z, m1, z, z);
+ z = _mm512_maskz_andnot_ps (m1, z, z);
+ y = _mm256_mask_andnot_ps (y, m2, y, y);
+ y = _mm256_maskz_andnot_ps (m2, y, y);
+ x = _mm_mask_andnot_ps (x, m2, x, x);
+ x = _mm_maskz_andnot_ps (m2, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (float *s1, float *s2, float *r)
+{
+ int i;
+ int tmp;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ tmp = (~(*(int *) &s1[i])) & (*(int *) &s2[i]);
+ r[i] = *(float *) &tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, ) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ float res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = 13. * i;
+ s2.a[i] = 17. * i;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_andnot_ps) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_andnot_ps) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_andnot_ps) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, ) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE () (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, ) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO () (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, ) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d z;
+volatile __m256d y;
+volatile __m128d x;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_and_pd (z, z);
+ z = _mm512_mask_and_pd (z, m, z, z);
+ z = _mm512_maskz_and_pd (m, z, z);
+ y = _mm256_mask_and_pd (y, m, y, y);
+ y = _mm256_maskz_and_pd (m, y, y);
+ x = _mm_mask_and_pd (x, m, x, x);
+ x = _mm_maskz_and_pd (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *s1, double *s2, double *r)
+{
+ int i;
+ long long tmp;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ tmp = (*(long long *) &s1[i]) & (*(long long *) &s2[i]);
+ r[i] = *(double *) &tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ double res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = 13. * i;
+ s2.a[i] = 17. * i;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_and_pd) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_and_pd) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_and_pd) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vandps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 z;
+volatile __m256 y;
+volatile __m128 x;
+volatile __mmask16 m1;
+volatile __mmask8 m2;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_and_ps (z, z);
+ z = _mm512_mask_and_ps (z, m1, z, z);
+ z = _mm512_maskz_and_ps (m1, z, z);
+ y = _mm256_mask_and_ps (y, m2, y, y);
+ y = _mm256_maskz_and_ps (m2, y, y);
+ x = _mm_mask_and_ps (x, m2, x, x);
+ x = _mm_maskz_and_ps (m2, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (float *s1, float *s2, float *r)
+{
+ int i;
+ int tmp;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ tmp = (*(int *) &s1[i]) & (*(int *) &s2[i]);
+ r[i] = *(float *) &tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, ) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ float res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = 13. * i;
+ s2.a[i] = 17. * i;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_and_ps) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_and_ps) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_and_ps) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, ) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE () (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, ) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO () (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, ) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 x;
+volatile __m256 y;
+volatile __m128 z;
+volatile __mmask16 mx;
+volatile __mmask8 my;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm512_broadcast_f32x2 (z);
+ x = _mm512_mask_broadcast_f32x2 (x, mx, z);
+ x = _mm512_maskz_broadcast_f32x2 (mx, z);
+ y = _mm256_broadcast_f32x2 (z);
+ y = _mm256_mask_broadcast_f32x2 (y, my, z);
+ y = _mm256_maskz_broadcast_f32x2 (my, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (float *r, float *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s[i % 2];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN,) res1, res2, res3;
+ UNION_TYPE (128,) src;
+ MASK_TYPE mask = SIZE | 123;
+ float res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < 4; i++)
+ {
+ src.a[i] = 34.67 * i * sign;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_broadcast_f32x2) (src.x);
+ res2.x = INTRINSIC (_mask_broadcast_f32x2) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_broadcast_f32x2) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN,) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE ()(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO ()(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]|vshuff32x4\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vshuff32x4\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}|vshuff32x4\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 x;
+volatile __m256 y;
+volatile __mmask16 m;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm512_broadcast_f32x8 (y);
+ x = _mm512_mask_broadcast_f32x8 (x, m, y);
+ x = _mm512_maskz_broadcast_f32x8 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (float *r, float *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s[i % 8];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN,) res1, res2, res3;
+ UNION_TYPE (256,) src;
+ MASK_TYPE mask = SIZE | 123;
+ float res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < 8; i++)
+ {
+ src.a[i] = 34.67 * i * sign;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_broadcast_f32x8) (src.x);
+ res2.x = INTRINSIC (_mask_broadcast_f32x8) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_broadcast_f32x8) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN,) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE ()(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO ()(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcastf64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]|vshuff64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vshuff64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}|vshuff64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]|vshuff64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vshuff64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}|vshuff64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d x;
+volatile __m256d y;
+volatile __m128d z;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm512_broadcast_f64x2 (z);
+ x = _mm512_mask_broadcast_f64x2 (x, m, z);
+ x = _mm512_maskz_broadcast_f64x2 (m, z);
+ y = _mm256_broadcast_f64x2 (z);
+ y = _mm256_mask_broadcast_f64x2 (y, m, z);
+ y = _mm256_maskz_broadcast_f64x2 (m, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *r, double *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s[i % 2];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, d) res1, res2, res3;
+ UNION_TYPE (128, d) src;
+ MASK_TYPE mask = SIZE | 123;
+ double res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < 2; i++)
+ {
+ src.a[i] = 34.67 * i * sign;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_broadcast_f64x2) (src.x);
+ res2.x = INTRINSIC (_mask_broadcast_f64x2) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_broadcast_f64x2) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x2\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask16 mx;
+volatile __mmask8 my;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm512_broadcast_i32x2 (z);
+ x = _mm512_mask_broadcast_i32x2 (x, mx, z);
+ x = _mm512_maskz_broadcast_i32x2 (mx, z);
+ y = _mm256_broadcast_i32x2 (z);
+ y = _mm256_mask_broadcast_i32x2 (y, my, z);
+ y = _mm256_maskz_broadcast_i32x2 (my, z);
+ z = _mm_broadcast_i32x2 (z);
+ z = _mm_mask_broadcast_i32x2 (z, my, z);
+ z = _mm_maskz_broadcast_i32x2 (my, z);
+
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (int *r, int *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s[i % 2];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_d) res1, res2, res3;
+ UNION_TYPE (128, i_d) src;
+ MASK_TYPE mask = SIZE | 123;
+ int res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < 4; i++)
+ {
+ src.a[i] = 34 * i * sign;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_broadcast_i32x2) (src.x);
+ res2.x = INTRINSIC (_mask_broadcast_i32x2) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_broadcast_i32x2) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]|vshufi32x4\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vshufi32x4\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x8\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}|vshufi32x4\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __mmask16 m;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm512_broadcast_i32x8 (y);
+ x = _mm512_mask_broadcast_i32x8 (x, m, y);
+ x = _mm512_maskz_broadcast_i32x8 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (int *r, int *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s[i % 8];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_d) res1, res2, res3;
+ UNION_TYPE (256, i_d) src;
+ MASK_TYPE mask = SIZE | 123;
+ int res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < 8; i++)
+ {
+ src.a[i] = 34 * i * sign;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_broadcast_i32x8) (src.x);
+ res2.x = INTRINSIC (_mask_broadcast_i32x8) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_broadcast_i32x8) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcasti64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]|vshufi64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vshufi64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}|vshufi64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]|vshufi64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vshufi64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}|vshufi64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm512_broadcast_i64x2 (z);
+ x = _mm512_mask_broadcast_i64x2 (x, m, z);
+ x = _mm512_maskz_broadcast_i64x2 (m, z);
+ y = _mm256_broadcast_i64x2 (z);
+ y = _mm256_mask_broadcast_i64x2 (y, m, z);
+ y = _mm256_maskz_broadcast_i64x2 (m, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (long long *r, long long *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = s[i % 2];
+ }
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ UNION_TYPE (128, i_q) src;
+ MASK_TYPE mask = SIZE | 123;
+ long long res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < 2; i++)
+ {
+ src.a[i] = 34 * i * sign;
+ sign = sign * -1;
+ }
+ for (i = 0; i < SIZE; i++)
+ res2.a[i] = DEFAULT_VALUE;
+
+ res1.x = INTRINSIC (_broadcast_i64x2) (src.x);
+ res2.x = INTRINSIC (_mask_broadcast_i64x2) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_broadcast_i64x2) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 6 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*\{rn-sae\}\[^\n\]*%zmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*\{ru-sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2qq\[ \\t\]+\[^\n\]*\{rz-sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d s1;
+volatile __m256d s2;
+volatile __m128d s3;
+volatile __m512i res1;
+volatile __m256i res2;
+volatile __m128i res3;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ res1 = _mm512_cvtpd_epi64 (s1);
+ res2 = _mm256_cvtpd_epi64 (s2);
+ res3 = _mm_cvtpd_epi64 (s3);
+
+ res1 = _mm512_mask_cvtpd_epi64 (res1, m, s1);
+ res2 = _mm256_mask_cvtpd_epi64 (res2, m, s2);
+ res3 = _mm_mask_cvtpd_epi64 (res3, m, s3);
+
+ res1 = _mm512_maskz_cvtpd_epi64 (m, s1);
+ res2 = _mm256_maskz_cvtpd_epi64 (m, s2);
+ res3 = _mm_maskz_cvtpd_epi64 (m, s3);
+
+ res1 = _mm512_cvt_roundpd_epi64 (s1, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+ res1 = _mm512_mask_cvt_roundpd_epi64 (res1, m, s1, _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC);
+ res1 = _mm512_maskz_cvt_roundpd_epi64 (m, s1, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *s, long long *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = (s[i] >= 0) ? (long long) (s[i] + 0.5)
+ : (long long) (s[i] - 0.5);
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ long long res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123.456 * (i + 2000) * sign;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_cvtpd_epi64) (s.x);
+ res2.x = INTRINSIC (_mask_cvtpd_epi64) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvtpd_epi64) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 6 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*\{rn-sae\}\[^\n\]*%zmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*\{rd-sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2uqq\[ \\t\]+\[^\n\]*\{rz-sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d s1;
+volatile __m256d s2;
+volatile __m128d s3;
+volatile __m512i res1;
+volatile __m256i res2;
+volatile __m128i res3;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ res1 = _mm512_cvtpd_epu64 (s1);
+ res2 = _mm256_cvtpd_epu64 (s2);
+ res3 = _mm_cvtpd_epu64 (s3);
+
+ res1 = _mm512_mask_cvtpd_epu64 (res1, m, s1);
+ res2 = _mm256_mask_cvtpd_epu64 (res2, m, s2);
+ res3 = _mm_mask_cvtpd_epu64 (res3, m, s3);
+
+ res1 = _mm512_maskz_cvtpd_epu64 (m, s1);
+ res2 = _mm256_maskz_cvtpd_epu64 (m, s2);
+ res3 = _mm_maskz_cvtpd_epu64 (m, s3);
+
+ res1 = _mm512_cvt_roundpd_epu64 (s1, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+ res1 = _mm512_mask_cvt_roundpd_epu64 (res1, m, s1, _MM_FROUND_TO_NEG_INF | _MM_FROUND_NO_EXC);
+ res1 = _mm512_maskz_cvt_roundpd_epu64 (m, s1, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *s, unsigned long long *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = (unsigned long long) (s[i] + 0.5);
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned long long res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123.456 * (i + 2000);
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_cvtpd_epu64) (s.x);
+ res2.x = INTRINSIC (_mask_cvtpd_epu64) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvtpd_epu64) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*\{rn-sae\}\[^\n\]*%zmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*\{ru-sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2qq\[ \\t\]+\[^\n\]*\{rz-sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x1;
+volatile __m256i x2;
+volatile __m128i x3;
+volatile __m256 z1;
+volatile __m128 z2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm512_cvtps_epi64 (z1);
+ x1 = _mm512_mask_cvtps_epi64 (x1, m, z1);
+ x1 = _mm512_maskz_cvtps_epi64 (m, z1);
+ x2 = _mm256_cvtps_epi64 (z2);
+ x2 = _mm256_mask_cvtps_epi64 (x2, m, z2);
+ x2 = _mm256_maskz_cvtps_epi64 (m, z2);
+ x3 = _mm_cvtps_epi64 (z2);
+ x3 = _mm_mask_cvtps_epi64 (x3, m, z2);
+ x3 = _mm_maskz_cvtps_epi64 (m, z2);
+ x1 = _mm512_cvt_roundps_epi64 (z1, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+ x1 = _mm512_mask_cvt_roundps_epi64 (x1, m, z1, _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC);
+ x1 = _mm512_maskz_cvt_roundps_epi64 (m, z1, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (long long *r, float *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ r[i] = (long long) (s[i] + ((s[i] >= 0) ? 0.5 : -0.5));
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ UNION_TYPE (AVX512F_LEN_HALF,) src;
+ MASK_TYPE mask = MASK_VALUE;
+ long long res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ res2.a[i] = DEFAULT_VALUE;
+ src.a[i] = 1.5 + 34.67 * i * sign;
+ sign = sign * -1;
+ }
+
+ res1.x = INTRINSIC (_cvtps_epi64) (src.x);
+ res2.x = INTRINSIC (_mask_cvtps_epi64) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_cvtps_epi64) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*\{rn-sae\}\[^\n\]*%zmm\[0-9\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*\{rd-sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2uqq\[ \\t\]+\[^\n\]*\{rz-sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x1;
+volatile __m256i x2;
+volatile __m128i x3;
+volatile __m256 z1;
+volatile __m128 z2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm512_cvtps_epu64 (z1);
+ x1 = _mm512_mask_cvtps_epu64 (x1, m, z1);
+ x1 = _mm512_maskz_cvtps_epu64 (m, z1);
+ x2 = _mm256_cvtps_epu64 (z2);
+ x2 = _mm256_mask_cvtps_epu64 (x2, m, z2);
+ x2 = _mm256_maskz_cvtps_epu64 (m, z2);
+ x3 = _mm_cvtps_epu64 (z2);
+ x3 = _mm_mask_cvtps_epu64 (x3, m, z2);
+ x3 = _mm_maskz_cvtps_epu64 (m, z2);
+ x1 = _mm512_cvt_roundps_epu64 (z1, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+ x1 = _mm512_mask_cvt_roundps_epu64 (x1, m, z1, _MM_FROUND_TO_NEG_INF | _MM_FROUND_NO_EXC);
+ x1 = _mm512_maskz_cvt_roundps_epu64 (m, z1, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned long long *r, float *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ r[i] = (unsigned long long) (s[i] + ((s[i] >= 0) ? 0.5 : -0.5));
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ UNION_TYPE (AVX512F_LEN_HALF,) src;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned long long res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ res2.a[i] = DEFAULT_VALUE;
+ src.a[i] = 1.5 + 34.67 * i;
+ }
+
+ res1.x = INTRINSIC (_cvtps_epu64) (src.x);
+ res2.x = INTRINSIC (_mask_cvtps_epu64) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_cvtps_epu64) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*\{rn-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*\{rd-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2pd\[ \\t\]+\[^\n\]*\{rz-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i s1;
+volatile __m256i s2;
+volatile __m128i s3;
+volatile __m512d res1;
+volatile __m256d res2;
+volatile __m128d res3;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ res1 = _mm512_cvtepi64_pd (s1);
+ res1 = _mm512_mask_cvtepi64_pd (res1, m, s1);
+ res1 = _mm512_maskz_cvtepi64_pd (m, s1);
+ res2 = _mm256_cvtepi64_pd (s2);
+ res2 = _mm256_mask_cvtepi64_pd (res2, m, s2);
+ res2 = _mm256_maskz_cvtepi64_pd (m, s2);
+ res3 = _mm_cvtepi64_pd (s3);
+ res3 = _mm_mask_cvtepi64_pd (res3, m, s3);
+ res3 = _mm_maskz_cvtepi64_pd (m, s3);
+ res1 = _mm512_cvt_roundepi64_pd (s1, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+ res1 = _mm512_mask_cvt_roundepi64_pd (res1, m, s1, _MM_FROUND_TO_NEG_INF | _MM_FROUND_NO_EXC);
+ res1 = _mm512_maskz_cvt_roundepi64_pd (m, s1, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (long long *s, double *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = (double) s[i];
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_q) s;
+ UNION_TYPE (AVX512F_LEN, d) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ double res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123 * (i + 2000) * sign;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_cvtepi64_pd) (s.x);
+ res2.x = INTRINSIC (_mask_cvtepi64_pd) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvtepi64_pd) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtqq2psx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2psx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2psx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2psy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2psy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2psy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2ps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2ps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2ps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2ps\[ \\t\]+\[^\n\]*\{rn-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2ps\[ \\t\]+\[^\n\]*\{rd-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtqq2ps\[ \\t\]+\[^\n\]*\{rz-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i s1;
+volatile __m256i s2;
+volatile __m128i s3;
+volatile __m256 res1;
+volatile __m128 res2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ res1 = _mm512_cvtepi64_ps (s1);
+ res1 = _mm512_mask_cvtepi64_ps (res1, m, s1);
+ res1 = _mm512_maskz_cvtepi64_ps (m, s1);
+ res2 = _mm256_cvtepi64_ps (s2);
+ res2 = _mm256_mask_cvtepi64_ps (res2, m, s2);
+ res2 = _mm256_maskz_cvtepi64_ps (m, s2);
+ res2 = _mm_cvtepi64_ps (s3);
+ res2 = _mm_mask_cvtepi64_ps (res2, m, s3);
+ res2 = _mm_maskz_cvtepi64_ps (m, s3);
+ res1 = _mm512_cvt_roundepi64_ps (s1, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+ res1 = _mm512_mask_cvt_roundepi64_ps (res1, m, s1, _MM_FROUND_TO_NEG_INF | _MM_FROUND_NO_EXC);
+ res1 = _mm512_maskz_cvt_roundepi64_ps (m, s1, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#define SIZE_HALF (AVX512F_LEN_HALF / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (long long *s, float *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE_HALF; i++)
+ r[i] = (i < SIZE) ? (float) s[i] : 0;
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_q) s;
+ UNION_TYPE (AVX512F_LEN_HALF,) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ float res_ref[SIZE_HALF];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123 * (i + 2000) * sign;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_cvtepi64_ps) (s.x);
+ res2.x = INTRINSIC (_mask_cvtepi64_ps) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvtepi64_ps) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN_HALF,) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE ()(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF,) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO ()(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF,) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 6 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2qq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d s1;
+volatile __m256d s2;
+volatile __m128d s3;
+volatile __m512i res1;
+volatile __m256i res2;
+volatile __m128i res3;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ res1 = _mm512_cvttpd_epi64 (s1);
+ res2 = _mm256_cvttpd_epi64 (s2);
+ res3 = _mm_cvttpd_epi64 (s3);
+
+ res1 = _mm512_mask_cvttpd_epi64 (res1, m, s1);
+ res2 = _mm256_mask_cvttpd_epi64 (res2, m, s2);
+ res3 = _mm_mask_cvttpd_epi64 (res3, m, s3);
+
+ res1 = _mm512_maskz_cvttpd_epi64 (m, s1);
+ res2 = _mm256_maskz_cvttpd_epi64 (m, s2);
+ res3 = _mm_maskz_cvttpd_epi64 (m, s3);
+
+ res1 = _mm512_cvtt_roundpd_epi64 (s1, _MM_FROUND_NO_EXC);
+ res1 = _mm512_mask_cvtt_roundpd_epi64 (res1, m, s1, _MM_FROUND_NO_EXC);
+ res1 = _mm512_maskz_cvtt_roundpd_epi64 (m, s1, _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *s, long long *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = (long long) s[i];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ long long res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123.456 * (i + 2000) * sign;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_cvttpd_epi64) (s.x);
+ res2.x = INTRINSIC (_mask_cvttpd_epi64) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvttpd_epi64) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 6 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2uqq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d s1;
+volatile __m256d s2;
+volatile __m128d s3;
+volatile __m512i res1;
+volatile __m256i res2;
+volatile __m128i res3;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ res1 = _mm512_cvttpd_epu64 (s1);
+ res2 = _mm256_cvttpd_epu64 (s2);
+ res3 = _mm_cvttpd_epu64 (s3);
+
+ res1 = _mm512_mask_cvttpd_epu64 (res1, m, s1);
+ res2 = _mm256_mask_cvttpd_epu64 (res2, m, s2);
+ res3 = _mm_mask_cvttpd_epu64 (res3, m, s3);
+
+ res1 = _mm512_maskz_cvttpd_epu64 (m, s1);
+ res2 = _mm256_maskz_cvttpd_epu64 (m, s2);
+ res3 = _mm_maskz_cvttpd_epu64 (m, s3);
+
+ res1 = _mm512_cvtt_roundpd_epu64 (s1, _MM_FROUND_NO_EXC);
+ res1 = _mm512_mask_cvtt_roundpd_epu64 (res1, m, s1, _MM_FROUND_NO_EXC);
+ res1 = _mm512_maskz_cvtt_roundpd_epu64 (m, s1, _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *s, unsigned long long *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ r[i] = (unsigned long long) s[i];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned long long res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123.456 * (i + 2000);
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_cvttpd_epu64) (s.x);
+ res2.x = INTRINSIC (_mask_cvttpd_epu64) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvttpd_epu64) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2qq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x1;
+volatile __m256i x2;
+volatile __m128i x3;
+volatile __m256 z1;
+volatile __m128 z2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm512_cvttps_epi64 (z1);
+ x1 = _mm512_mask_cvttps_epi64 (x1, m, z1);
+ x1 = _mm512_maskz_cvttps_epi64 (m, z1);
+ x2 = _mm256_cvttps_epi64 (z2);
+ x2 = _mm256_mask_cvttps_epi64 (x2, m, z2);
+ x2 = _mm256_maskz_cvttps_epi64 (m, z2);
+ x3 = _mm_cvttps_epi64 (z2);
+ x3 = _mm_mask_cvttps_epi64 (x3, m, z2);
+ x3 = _mm_maskz_cvttps_epi64 (m, z2);
+ x1 = _mm512_cvtt_roundps_epi64 (z1, _MM_FROUND_NO_EXC);
+ x1 = _mm512_mask_cvtt_roundps_epi64 (x1, m, z1, _MM_FROUND_NO_EXC);
+ x1 = _mm512_maskz_cvtt_roundps_epi64 (m, z1, _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (long long *r, float *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ r[i] = (long long) s[i];
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ UNION_TYPE (AVX512F_LEN_HALF,) src;
+ MASK_TYPE mask = MASK_VALUE;
+ long long res_ref[SIZE];
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ res2.a[i] = DEFAULT_VALUE;
+ src.a[i] = 1.5 + 34.67 * i * sign;
+ sign = sign * -1;
+ }
+
+ res1.x = INTRINSIC (_cvttps_epi64) (src.x);
+ res2.x = INTRINSIC (_mask_cvttps_epi64) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_cvttps_epi64) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2uqq\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x1;
+volatile __m256i x2;
+volatile __m128i x3;
+volatile __m256 z1;
+volatile __m128 z2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm512_cvttps_epu64 (z1);
+ x1 = _mm512_mask_cvttps_epu64 (x1, m, z1);
+ x1 = _mm512_maskz_cvttps_epu64 (m, z1);
+ x2 = _mm256_cvttps_epu64 (z2);
+ x2 = _mm256_mask_cvttps_epu64 (x2, m, z2);
+ x2 = _mm256_maskz_cvttps_epu64 (m, z2);
+ x3 = _mm_cvttps_epu64 (z2);
+ x3 = _mm_mask_cvttps_epu64 (x3, m, z2);
+ x3 = _mm_maskz_cvttps_epu64 (m, z2);
+ x1 = _mm512_cvtt_roundps_epu64 (z1, _MM_FROUND_NO_EXC);
+ x1 = _mm512_mask_cvtt_roundps_epu64 (x1, m, z1, _MM_FROUND_NO_EXC);
+ x1 = _mm512_maskz_cvtt_roundps_epu64 (m, z1, _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned long long *r, float *s)
+{
+ int i;
+ for (i = 0; i < SIZE; i++)
+ r[i] = (unsigned long long) s[i];
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_q) res1, res2, res3;
+ UNION_TYPE (AVX512F_LEN_HALF,) src;
+ MASK_TYPE mask = MASK_VALUE;
+ unsigned long long res_ref[SIZE];
+
+ for (i = 0; i < SIZE; i++)
+ {
+ res2.a[i] = DEFAULT_VALUE;
+ src.a[i] = 1.5 + 34.67 * i;
+ }
+
+ res1.x = INTRINSIC (_cvttps_epu64) (src.x);
+ res2.x = INTRINSIC (_mask_cvttps_epu64) (res2.x, mask, src.x);
+ res3.x = INTRINSIC (_maskz_cvttps_epu64) (mask, src.x);
+
+ CALC (res_ref, src.a);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*\{rn-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*\{ru-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2pd\[ \\t\]+\[^\n\]*\{rz-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i s1;
+volatile __m256i s2;
+volatile __m128i s3;
+volatile __m512d res1;
+volatile __m256d res2;
+volatile __m128d res3;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ res1 = _mm512_cvtepu64_pd (s1);
+ res1 = _mm512_mask_cvtepu64_pd (res1, m, s1);
+ res1 = _mm512_maskz_cvtepu64_pd (m, s1);
+ res2 = _mm256_cvtepu64_pd (s2);
+ res2 = _mm256_mask_cvtepu64_pd (res2, m, s2);
+ res2 = _mm256_maskz_cvtepu64_pd (m, s2);
+ res3 = _mm_cvtepu64_pd (s3);
+ res3 = _mm_mask_cvtepu64_pd (res3, m, s3);
+ res3 = _mm_maskz_cvtepu64_pd (m, s3);
+ res1 = _mm512_cvt_roundepu64_pd (s1, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+ res1 = _mm512_mask_cvt_roundepu64_pd (res1, m, s1, _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC);
+ res1 = _mm512_maskz_cvt_roundepu64_pd (m, s1, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned long long *s, double *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = (double) s[i];
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_q) s;
+ UNION_TYPE (AVX512F_LEN, d) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ double res_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123 * (i + 2000);
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_cvtepu64_pd) (s.x);
+ res2.x = INTRINSIC (_mask_cvtepu64_pd) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvtepu64_pd) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtuqq2psx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2psx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2psx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2psy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2psy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2psy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2ps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2ps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2ps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2ps\[ \\t\]+\[^\n\]*\{rn-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2ps\[ \\t\]+\[^\n\]*\{ru-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtuqq2ps\[ \\t\]+\[^\n\]*\{rz-sae\}\[^\n\]*%zmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i s1;
+volatile __m256i s2;
+volatile __m128i s3;
+volatile __m256 res1;
+volatile __m128 res2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ res1 = _mm512_cvtepu64_ps (s1);
+ res1 = _mm512_mask_cvtepu64_ps (res1, m, s1);
+ res1 = _mm512_maskz_cvtepu64_ps (m, s1);
+ res2 = _mm256_cvtepu64_ps (s2);
+ res2 = _mm256_mask_cvtepu64_ps (res2, m, s2);
+ res2 = _mm256_maskz_cvtepu64_ps (m, s2);
+ res2 = _mm_cvtepu64_ps (s3);
+ res2 = _mm_mask_cvtepu64_ps (res2, m, s3);
+ res2 = _mm_maskz_cvtepu64_ps (m, s3);
+ res1 = _mm512_cvt_roundepu64_ps (s1, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+ res1 = _mm512_mask_cvt_roundepu64_ps (res1, m, s1, _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC);
+ res1 = _mm512_maskz_cvt_roundepu64_ps (m, s1, _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#define SIZE_HALF (AVX512F_LEN_HALF / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (unsigned long long *s, float *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE_HALF; i++)
+ r[i] = (i < SIZE) ? (float) s[i] : 0;
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_q) s;
+ UNION_TYPE (AVX512F_LEN_HALF,) res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ float res_ref[SIZE_HALF];
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123 * (i + 2000);
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_cvtepu64_ps) (s.x);
+ res2.x = INTRINSIC (_mask_cvtepu64_ps) (res2.x, mask, s.x);
+ res3.x = INTRINSIC (_maskz_cvtepu64_ps) (mask, s.x);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN_HALF,) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE ()(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF,) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO ()(res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN_HALF,) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vextractf32x8\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vextractf32x8\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vextractf32x8\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 x;
+volatile __m256 y;
+
+void extern
+avx512dq_test (void)
+{
+ y = _mm512_extractf32x8_ps (x, 1);
+ y = _mm512_mask_extractf32x8_ps (y, 2, x, 1);
+ y = _mm512_maskz_extractf32x8_ps (2, x, 1);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vextractf64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vextractf64x2\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vextractf64x2\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vextractf64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+
+#include <immintrin.h>
+
+volatile __m512d x1;
+volatile __m256d x2;
+volatile __m128d y;
+
+void extern
+avx512dq_test (void)
+{
+ y = _mm512_extractf64x2_pd (x1, 3);
+ y = _mm512_mask_extractf64x2_pd (y, 2, x1, 3);
+ y = _mm512_maskz_extractf64x2_pd (2, x1, 3);
+ y = _mm256_extractf64x2_pd (x2, 1);
+ y = _mm256_mask_extractf64x2_pd (y, 2, x2, 1);
+ y = _mm256_maskz_extractf64x2_pd (2, x2, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+#include "string.h"
+
+void
+CALC (double *s1, double *res_ref, int mask)
+{
+ memset (res_ref, 0, 16);
+ memcpy (res_ref, s1 + mask * 2, 16);
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s1;
+ union128d res1, res2, res3;
+ double res_ref[2];
+ MASK_TYPE mask = MASK_VALUE;
+ int j;
+
+ for (j = 0; j < SIZE; j++)
+ {
+ s1.a[j] = j * j / 4.56;
+ }
+
+ for (j = 0; j < 2; j++)
+ {
+ res1.a[j] = DEFAULT_VALUE;
+ res2.a[j] = DEFAULT_VALUE;
+ res3.a[j] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_extractf64x2_pd) (s1.x, 1);
+ res2.x = INTRINSIC (_mask_extractf64x2_pd) (res2.x, mask, s1.x, 1);
+ res3.x = INTRINSIC (_maskz_extractf64x2_pd) (mask, s1.x, 1);
+ CALC (s1.a, res_ref, 1);
+
+ if (check_union128d (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, 2);
+ if (check_union128d (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, 2);
+ if (check_union128d (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vextracti32x8\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vextracti32x8\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vextracti32x8\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+
+void extern
+avx512dq_test (void)
+{
+ y = _mm512_extracti32x8_epi32 (x, 1);
+ y = _mm512_mask_extracti32x8_epi32 (y, 2, x, 1);
+ y = _mm512_maskz_extracti32x8_epi32 (2, x, 1);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vextracti64x2\[ \\t\]+\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vextracti64x2\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vextracti64x2\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vextracti64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x1;
+volatile __m256i x2;
+volatile __m128i y;
+
+void extern
+avx512dq_test (void)
+{
+ y = _mm512_extracti64x2_epi64 (x1, 3);
+ y = _mm512_mask_extracti64x2_epi64 (y, 2, x1, 3);
+ y = _mm512_maskz_extracti64x2_epi64 (2, x1, 3);
+ y = _mm256_extracti64x2_epi64 (x2, 1);
+ y = _mm256_mask_extracti64x2_epi64 (y, 2, x2, 1);
+ y = _mm256_maskz_extracti64x2_epi64 (2, x2, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+#include "string.h"
+
+void
+CALC (long long int *s1, long long int *res_ref, int mask)
+{
+ memset (res_ref, 0, 16);
+ memcpy (res_ref, s1 + mask * 2, 16);
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_q) s1;
+ union128i_q res1, res2, res3;
+ long long int res_ref[2];
+ MASK_TYPE mask = MASK_VALUE;
+ int j;
+
+ for (j = 0; j < SIZE; j++)
+ {
+ s1.a[j] = j * j + 37;
+ }
+
+ for (j = 0; j < 2; j++)
+ {
+ res1.a[j] = DEFAULT_VALUE;
+ res2.a[j] = DEFAULT_VALUE;
+ res3.a[j] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_extracti64x2_epi64) (s1.x, 1);
+ res2.x =
+ INTRINSIC (_mask_extracti64x2_epi64) (res2.x, mask, s1.x, 1);
+ res3.x = INTRINSIC (_maskz_extracti64x2_epi64) (mask, s1.x, 1);
+ CALC (s1.a, res_ref, 1);
+
+ if (check_union128i_q (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, 2);
+ if (check_union128i_q (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, 2);
+ if (check_union128i_q (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfpclasspd\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n^k\]*%k\[1-7\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspd\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n^k\]*%k\[1-7\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspd\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n^k\]*%k\[1-7\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspd\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n^k\]*%k\[1-7\]\{" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspd\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n^k\]*%k\[1-7\]\{" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclasspd\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n^k\]*%k\[1-7\]\{" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d x512;
+volatile __m256d x256;
+volatile __m128d x128;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ m = _mm512_fpclass_pd_mask (x512, 13);
+ m = _mm256_fpclass_pd_mask (x256, 13);
+ m = _mm_fpclass_pd_mask (x128, 13);
+ m = _mm512_mask_fpclass_pd_mask (2, x512, 13);
+ m = _mm256_mask_fpclass_pd_mask (2, x256, 13);
+ m = _mm_mask_fpclass_pd_mask (2, x128, 13);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#include <math.h>
+#include <values.h>
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+#ifndef __FPCLASSPD__
+#define __FPCLASSPD__
+int check_fp_class_dp (double src, int imm)
+{
+ int qNaN_res = isnan (src);
+ int sNaN_res = isnan (src);
+ int Pzero_res = (src == 0.0);
+ int Nzero_res = (src == -0.0);
+ int PInf_res = (isinf (src) == 1);
+ int NInf_res = (isinf (src) == -1);
+ int Denorm_res = (fpclassify (src) == FP_SUBNORMAL);
+ int FinNeg_res = finite (src) && (src < 0);
+
+ int result = (((imm & 1) && qNaN_res)
+ || (((imm >> 1) & 1) && Pzero_res)
+ || (((imm >> 2) & 1) && Nzero_res)
+ || (((imm >> 3) & 1) && PInf_res)
+ || (((imm >> 4) & 1) && NInf_res)
+ || (((imm >> 5) & 1) && Denorm_res)
+ || (((imm >> 6) & 1) && FinNeg_res)
+ || (((imm >> 7) & 1) && sNaN_res));
+ return result;
+}
+#endif
+
+__mmask8
+CALC (double *s1, int imm)
+{
+ int i;
+ __mmask8 res = 0;
+
+ for (i = 0; i < SIZE; i++)
+ if (check_fp_class_dp(s1[i], imm))
+ res = res | (1 << i);
+
+ return res;
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, d) src;
+ __mmask8 res1, res2, res_ref = 0;
+ MASK_TYPE mask = MASK_VALUE;
+
+ src.a[0] = NAN;
+ src.a[1] = 1.0 / 0.0;
+ for (i = 2; i < SIZE; i++)
+ {
+ src.a[i] = -24.43 + 0.6 * i;
+ }
+
+ res1 = INTRINSIC (_fpclass_pd_mask) (src.x, 0xFF);
+ res2 = INTRINSIC (_mask_fpclass_pd_mask) (mask, src.x, 0xFF);
+
+ res_ref = CALC (src.a, 0xFF);
+
+ if (res_ref != res1)
+ abort ();
+
+ if ((res_ref & mask) != res2)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfpclassps\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n^k\]*%k\[1-7\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclassps\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n^k\]*%k\[1-7\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclassps\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n^k\]*%k\[1-7\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclassps\[ \\t\]+\[^\n\]*%zmm\[0-7\]\[^\n^k\]*%k\[1-7\]\{" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclassps\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n^k\]*%k\[1-7\]\{" 1 } } */
+/* { dg-final { scan-assembler-times "vfpclassps\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n^k\]*%k\[1-7\]\{" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 x512;
+volatile __m256 x256;
+volatile __m128 x128;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512dq_test (void)
+{
+ m16 = _mm512_fpclass_ps_mask (x512, 13);
+ m8 = _mm256_fpclass_ps_mask (x256, 13);
+ m8 = _mm_fpclass_ps_mask (x128, 13);
+ m16 = _mm512_mask_fpclass_ps_mask (2, x512, 13);
+ m8 = _mm256_mask_fpclass_ps_mask (2, x256, 13);
+ m8 = _mm_mask_fpclass_ps_mask (2, x128, 13);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#include <math.h>
+#include <values.h>
+#include "avx512f-mask-type.h"
+#define SIZE (AVX512F_LEN / 32)
+
+#ifndef __FPCLASSPD__
+#define __FPCLASSPD__
+int check_fp_class_sp (float src, int imm)
+{
+ int qNaN_res = isnan (src);
+ int sNaN_res = isnan (src);
+ int Pzero_res = (src == 0.0);
+ int Nzero_res = (src == -0.0);
+ int PInf_res = (isinf (src) == 1);
+ int NInf_res = (isinf (src) == -1);
+ int Denorm_res = (fpclassify (src) == FP_SUBNORMAL);
+ int FinNeg_res = finite (src) && (src < 0);
+
+ int result = (((imm & 1) && qNaN_res)
+ || (((imm >> 1) & 1) && Pzero_res)
+ || (((imm >> 2) & 1) && Nzero_res)
+ || (((imm >> 3) & 1) && PInf_res)
+ || (((imm >> 4) & 1) && NInf_res)
+ || (((imm >> 5) & 1) && Denorm_res)
+ || (((imm >> 6) & 1) && FinNeg_res)
+ || (((imm >> 7) & 1) && sNaN_res));
+ return result;
+}
+#endif
+
+MASK_TYPE
+CALC (float *s1, int imm)
+{
+ int i;
+ MASK_TYPE res = 0;
+
+ for (i = 0; i < SIZE; i++)
+ if (check_fp_class_sp(s1[i], imm))
+ res = res | (1 << i);
+
+ return res;
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN,) src;
+ MASK_TYPE res1, res2, res_ref = 0;
+ MASK_TYPE mask = MASK_VALUE;
+
+ src.a[0] = NAN;
+ src.a[1] = 1.0 / 0.0;
+ for (i = 2; i < SIZE; i++)
+ {
+ src.a[i] = -24.43 + 0.6 * i;
+ }
+
+ res1 = INTRINSIC (_fpclass_ps_mask) (src.x, 0xFF);
+ res2 = INTRINSIC (_mask_fpclass_ps_mask) (mask, src.x, 0xFF);
+
+
+ res_ref = CALC (src.a, 0xFF);
+
+ if (res_ref != res1)
+ abort ();
+
+ if ((mask & res_ref) != res2)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler "vfpclasssd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m128d x128;
+volatile __mmask8 m8;
+
+void extern
+avx512dq_test (void)
+{
+ m8 = _mm_fpclass_sd_mask (x128, 13);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler "vfpclassss\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m128 x128;
+volatile __mmask8 m8;
+
+void extern
+avx512dq_test (void)
+{
+ m8 = _mm_fpclass_ss_mask (x128, 13);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vinsertf32x8\[ \\t\]+\[^\n\]*\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vinsertf32x8\[ \\t\]+\[^\n\]*\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vinsertf32x8\[ \\t\]+\[^\n\]*" 3 } } */
+
+#include <immintrin.h>
+
+volatile __m512 x;
+volatile __m256 y;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm512_insertf32x8 (x, y, 1);
+ x = _mm512_mask_insertf32x8 (x, 2, x, y, 1);
+ x = _mm512_maskz_insertf32x8 (2, x, y, 1);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vinsertf64x2\[^\n\]*ymm" 3 } } */
+/* { dg-final { scan-assembler-times "vinsertf64x2\[^\n\]*\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vinsertf64x2\[^\n\]*\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vinsertf64x2\[^\n\]*zmm" 3 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x;
+volatile __m128d y;
+volatile __m512d z;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm256_insertf64x2 (x, y, 1);
+ x = _mm256_mask_insertf64x2 (x, 2, x, y, 1);
+ x = _mm256_maskz_insertf64x2 (2, x, y, 1);
+ z = _mm512_insertf64x2 (z, y, 1);
+ z = _mm512_mask_insertf64x2 (z, 2, z, y, 1);
+ z = _mm512_maskz_insertf64x2 (2, z, y, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+#include "string.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (UNION_TYPE (AVX512F_LEN, d) s1, union128d s2,
+ double *res_ref, int mask)
+{
+ memcpy (res_ref, s1.a, SIZE * sizeof (double));
+ memcpy (res_ref + mask * 2, s2.a, 16);
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s1, res1, res2, res3;
+ union128d s2;
+ double res_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int j;
+
+ for (j = 0; j < SIZE; j++)
+ {
+ s1.a[j] = j * j;
+ res1.a[j] = DEFAULT_VALUE;
+ res2.a[j] = DEFAULT_VALUE;
+ res3.a[j] = DEFAULT_VALUE;
+ }
+
+ for (j = 0; j < 2; j++)
+ s2.a[j] = j * j * j;
+
+ res1.x = INTRINSIC (_insertf64x2) (s1.x, s2.x, 1);
+ res2.x = INTRINSIC (_mask_insertf64x2) (res2.x, mask, s1.x, s2.x, 1);
+ res3.x = INTRINSIC (_maskz_insertf64x2) (mask, s1.x, s2.x, 1);
+
+ CALC (s1, s2, res_ref, 1);
+
+ if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vinserti32x8\[ \\t\]+\[^\n\]*" 3 } } */
+/* { dg-final { scan-assembler-times "vinserti32x8\[ \\t\]+\[^\n\]*\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vinserti32x8\[ \\t\]+\[^\n\]*\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+volatile __m256i y;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm512_inserti32x8 (x, y, 1);
+ x = _mm512_mask_inserti32x8 (x, 2, x, y, 1);
+ x = _mm512_maskz_inserti32x8 (2, x, y, 1);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vinserti64x2\[^\n\]*ymm" 3 } } */
+/* { dg-final { scan-assembler-times "vinserti64x2\[^\n\]*\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vinserti64x2\[^\n\]*\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vinserti64x2\[^\n\]*zmm" 3 } } */
+
+#include <immintrin.h>
+
+volatile __m512i z;
+volatile __m256i x;
+volatile __m128i y;
+
+void extern
+avx512dq_test (void)
+{
+ x = _mm256_inserti64x2 (x, y, 1);
+ x = _mm256_mask_inserti64x2 (x, 2, x, y, 1);
+ x = _mm256_maskz_inserti64x2 (2, x, y, 1);
+ z = _mm512_inserti64x2 (z, y, 0);
+ z = _mm512_mask_inserti64x2 (z, 2, z, y, 0);
+ z = _mm512_maskz_inserti64x2 (2, z, y, 0);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+#include "string.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (UNION_TYPE (AVX512F_LEN, i_q) s1, union128i_q s2,
+ long long *res_ref, int mask)
+{
+ memcpy (res_ref, s1.a, SIZE * sizeof (long long));
+ memcpy (res_ref + mask * 2, s2.a, 16);
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_q) s1, res1, res2, res3;
+ union128i_q s2;
+ long long res_ref[SIZE];
+ MASK_TYPE mask = MASK_VALUE;
+ int j;
+
+ for (j = 0; j < SIZE; j++)
+ {
+ s1.a[j] = j * j;
+ res1.a[j] = DEFAULT_VALUE;
+ res2.a[j] = DEFAULT_VALUE;
+ res3.a[j] = DEFAULT_VALUE;
+ }
+
+ for (j = 0; j < 2; j++)
+ s2.a[j] = j * j * j;
+
+ res1.x = INTRINSIC (_inserti64x2) (s1.x, s2.x, 1);
+ res2.x = INTRINSIC (_mask_inserti64x2) (res2.x, mask, s1.x, s2.x, 1);
+ res3.x = INTRINSIC (_maskz_inserti64x2) (mask, s1.x, s2.x, 1);
+
+ CALC (s1, s2, res_ref, 1);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vorpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d z;
+volatile __m256d y;
+volatile __m128d x;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_or_pd (z, z);
+ z = _mm512_mask_or_pd (z, m, z, z);
+ z = _mm512_maskz_or_pd (m, z, z);
+
+ y = _mm256_mask_or_pd (y, m, y, y);
+ y = _mm256_maskz_or_pd (m, y, y);
+
+ x = _mm_mask_or_pd (x, m, x, x);
+ x = _mm_maskz_or_pd (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *src1, double *src2, double *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ long long tmp = (*(long long *) &src1[i]) | (*(long long *) &src2[i]);
+ dst[i] = *(double *) &tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN,d) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ double dst_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++) {
+ s1.a[i] = 132.45 * i;
+ s2.a[i] = 43.6 - i * 4.4;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_or_pd) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_or_pd) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_or_pd) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN,d) (res1, dst_ref))
+ abort ();
+
+ MASK_MERGE (d) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,d) (res2, dst_ref))
+ abort ();
+
+ MASK_ZERO (d) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,d) (res3, dst_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vorps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 z;
+volatile __m256 y;
+volatile __m128 x;
+volatile __mmask8 n;
+volatile __mmask16 m;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_or_ps (z, z);
+ z = _mm512_mask_or_ps (z, m, z, z);
+ z = _mm512_maskz_or_ps (m, z, z);
+
+ y = _mm256_mask_or_ps (y, n, y, y);
+ y = _mm256_maskz_or_ps (n, y, y);
+
+ x = _mm_mask_or_ps (x, n, x, x);
+ x = _mm_maskz_or_ps (n, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (float *src1, float *src2, float *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (*(int *) &src1[i]) | (*(int *) &src2[i]);
+ dst[i] = *(float *) &tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN,) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ float dst_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++) {
+ s1.a[i] = 132.45 * i;
+ s2.a[i] = 43.6 - i * 4.4;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_or_ps) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_or_ps) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_or_ps) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN,) (res1, dst_ref))
+ abort ();
+
+ MASK_MERGE () (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,) (res2, dst_ref))
+ abort ();
+
+ MASK_ZERO () (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,) (res3, dst_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpmovd2m\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+/* { dg-final { scan-assembler "vpmovd2m\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+/* { dg-final { scan-assembler "vpmovd2m\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+volatile __mmask16 m16;
+
+void extern
+avx512dq_test (void)
+{
+ m8 = _mm_movepi32_mask (x128);
+ m8 = _mm256_movepi32_mask (x256);
+ m16 = _mm512_movepi32_mask (x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *r, int *s1)
+{
+ int i;
+ MASK_TYPE res = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] >> 31)
+ res = res | (one << i);
+
+ *r = res;
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_d) src;
+ MASK_TYPE res, res_ref = 0;
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src.a[i] = 2 * i * sign;
+ sign = sign * -1;
+ }
+
+ res = INTRINSIC (_movepi32_mask) (src.x);
+
+ CALC (&res_ref, src.a);
+
+ if (res_ref != res)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpmovm2d\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%xmm\[0-9\]" } } */
+/* { dg-final { scan-assembler "vpmovm2d\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%ymm\[0-9\]" } } */
+/* { dg-final { scan-assembler "vpmovm2d\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%zmm\[0-9\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512dq_test (void)
+{
+ x128 = _mm_movm_epi32 (m8);
+ x256 = _mm256_movm_epi32 (m8);
+ x512 = _mm512_movm_epi32 (m16);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (int *r, MASK_TYPE s)
+{
+ int i;
+ int all_ones = 0xffffffff;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = ((s >> i) & 1) ? all_ones : 0;
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_d) res, res_ref;
+ MASK_TYPE src = (MASK_TYPE) 0x1111;
+
+ res.x = INTRINSIC (_movm_epi32) (src);
+
+ CALC (res_ref.a, src);
+
+ if (UNION_CHECK (AVX512F_LEN, i_d) (res, res_ref.a))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpmovm2q\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%xmm\[0-9\]" } } */
+/* { dg-final { scan-assembler "vpmovm2q\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%ymm\[0-9\]" } } */
+/* { dg-final { scan-assembler "vpmovm2q\[ \\t\]+\[^\n\]*%k\[1-7\]\[^\n\]*%zmm\[0-9\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+
+void extern
+avx512dq_test (void)
+{
+ x128 = _mm_movm_epi64 (m8);
+ x256 = _mm256_movm_epi64 (m8);
+ x512 = _mm512_movm_epi64 (m8);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (long long *r, MASK_TYPE s)
+{
+ int i;
+ long long all_ones = 0xffffffffffffffff;
+
+ for (i = 0; i < SIZE; i++)
+ r[i] = ((s >> i) & 1) ? all_ones : 0;
+}
+
+void
+TEST (void)
+{
+ int i;
+ UNION_TYPE (AVX512F_LEN, i_q) res, res_ref;
+ MASK_TYPE src = (MASK_TYPE) 0xff;
+
+ res.x = INTRINSIC (_movm_epi64) (src);
+
+ CALC (res_ref.a, src);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (res, res_ref.a))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpmovq2m\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+/* { dg-final { scan-assembler "vpmovq2m\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+/* { dg-final { scan-assembler "vpmovq2m\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%k\[1-7\]" } } */
+
+#include <immintrin.h>
+
+volatile __m512i x512;
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+
+void extern
+avx512dq_test (void)
+{
+ m8 = _mm_movepi64_mask (x128);
+ m8 = _mm256_movepi64_mask (x256);
+ m8 = _mm512_movepi64_mask (x512);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (MASK_TYPE *r, long long *s1)
+{
+ int i;
+ MASK_TYPE res = 0;
+ MASK_TYPE one = 1;
+
+ for (i = 0; i < SIZE; i++)
+ if (s1[i] >> 63)
+ res = res | (one << i);
+
+ *r = res;
+}
+
+void
+TEST (void)
+{
+ int i, sign;
+ UNION_TYPE (AVX512F_LEN, i_q) src;
+ MASK_TYPE res, res_ref = 0;
+
+ sign = -1;
+ for (i = 0; i < SIZE; i++)
+ {
+ src.a[i] = 2 * i * sign;
+ sign = sign * -1;
+ }
+
+ res = INTRINSIC (_movepi64_mask) (src.x);
+
+ CALC (&res_ref, src.a);
+
+ if (res_ref != res)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmullq\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512i _x1, _y1, _z1;
+volatile __m256i _x2, _y2, _z2;
+volatile __m128i _x3, _y3, _z3;
+
+void extern
+avx512dq_test (void)
+{
+ _x3 = _mm_mullo_epi64 (_y3, _z3);
+ _x3 = _mm_mask_mullo_epi64 (_x3, 2, _y3, _z3);
+ _x3 = _mm_maskz_mullo_epi64 (2, _y3, _z3);
+ _x2 = _mm256_mullo_epi64 (_y2, _z2);
+ _x2 = _mm256_mask_mullo_epi64 (_x2, 3, _y2, _z2);
+ _x2 = _mm256_maskz_mullo_epi64 (3, _y2, _z2);
+ _x1 = _mm512_mullo_epi64 (_y1, _z1);
+ _x1 = _mm512_mask_mullo_epi64 (_x1, 3, _y1, _z1);
+ _x1 = _mm512_maskz_mullo_epi64 (3, _y1, _z1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (long long *src1, long long *src2, long long *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ dst[i] = src1[i] * src2[i];
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, i_q) src1, src2, dst1, dst2, dst3;
+ long long dst_ref[SIZE];
+ int i;
+ MASK_TYPE mask = MASK_VALUE;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ src1.a[i] = i + 50;
+ src2.a[i] = i + 100;
+ dst2.a[i] = DEFAULT_VALUE;
+ }
+
+ dst1.x = INTRINSIC (_mullo_epi64) (src1.x, src2.x);
+ dst2.x = INTRINSIC (_mask_mullo_epi64) (dst2.x, mask, src1.x, src2.x);
+ dst3.x = INTRINSIC (_maskz_mullo_epi64) (mask, src1.x, src2.x);
+ CALC (src1.a, src2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, i_q) (dst1, dst_ref))
+ abort ();
+
+ MASK_MERGE (i_q) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (dst2, dst_ref))
+ abort ();
+
+ MASK_ZERO (i_q) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, i_q) (dst3, dst_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 6 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrangepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d z;
+volatile __m256d y;
+volatile __m128d x;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_range_round_pd (z, z, 15, _MM_FROUND_NO_EXC);
+ z = _mm512_range_pd (z, z, 15);
+ y = _mm256_range_pd (y, y, 15);
+ x = _mm_range_pd (x, x, 15);
+
+ z = _mm512_mask_range_round_pd (z, m, z, z, 15, _MM_FROUND_NO_EXC);
+ z = _mm512_mask_range_pd (z, m, z, z, 15);
+ y = _mm256_mask_range_pd (y, m, y, y, 15);
+ x = _mm_mask_range_pd (x, m, x, x, 15);
+
+ z = _mm512_maskz_range_round_pd (m, z, z, 15, _MM_FROUND_NO_EXC);
+ z = _mm512_maskz_range_pd (m, z, z, 15);
+ y = _mm256_maskz_range_pd (m, y, y, 15);
+ x = _mm_maskz_range_pd (m, x, x, 15);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+#define IMM 0x02
+
+void
+CALC (double *s1, double *s2, double *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ if (s1[i] < -s2[i])
+ r[i] = -s2[i];
+ else if (s1[i] > s2[i])
+ r[i] = s2[i];
+ else
+ r[i] = s1[i];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ double res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = 234.567 * i * sign;
+ s2.a[i] = 100 * (i + 1);
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_range_pd) (s1.x, s2.x, IMM);
+ res2.x = INTRINSIC (_mask_range_pd) (res2.x, mask, s1.x, s2.x, IMM);
+ res3.x = INTRINSIC (_maskz_range_pd) (mask, s1.x, s2.x, IMM);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 6 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrangeps\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 z;
+volatile __m256 y;
+volatile __m128 x;
+volatile __mmask8 m8;
+volatile __mmask16 m16;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_range_round_ps (z, z, 15, _MM_FROUND_NO_EXC);
+ z = _mm512_range_ps (z, z, 15);
+ y = _mm256_range_ps (y, y, 15);
+ x = _mm_range_ps (x, x, 15);
+
+ z = _mm512_mask_range_round_ps (z, m16, z, z, 15, _MM_FROUND_NO_EXC);
+ z = _mm512_mask_range_ps (z, m16, z, z, 15);
+ y = _mm256_mask_range_ps (y, m8, y, y, 15);
+ x = _mm_mask_range_ps (x, m8, x, x, 15);
+
+ z = _mm512_maskz_range_round_ps (m16, z, z, 15, _MM_FROUND_NO_EXC);
+ z = _mm512_maskz_range_ps (m16, z, z, 15);
+ y = _mm256_maskz_range_ps (m8, y, y, 15);
+ x = _mm_maskz_range_ps (m8, x, x, 15);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+#define IMM 0x02
+
+void
+CALC (float *s1, float *s2, float *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ if (s1[i] < -s2[i])
+ r[i] = -s2[i];
+ else if (s1[i] > s2[i])
+ r[i] = s2[i];
+ else
+ r[i] = s1[i];
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, ) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ float res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s1.a[i] = 234.567 * i * sign;
+ s2.a[i] = 100 * (i + 1);
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_range_ps) (s1.x, s2.x, IMM);
+ res2.x = INTRINSIC (_mask_range_ps) (res2.x, mask, s1.x, s2.x, IMM);
+ res3.x = INTRINSIC (_maskz_range_ps) (mask, s1.x, s2.x, IMM);
+
+ CALC (s1.a, s2.a, res_ref);
+
+ if (UNION_CHECK (AVX512F_LEN, ) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE () (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, ) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO () (res_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN, ) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vrangesd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vrangesd\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128d x1, x2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm_range_sd (x1, x2, 3);
+ x1 = _mm_range_round_sd (x1, x2, 3, _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vrangess\[ \\t\]+\[^\n\]*\{sae\}\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrangess\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m128 x1, x2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm_range_ss (x1, x2, 1);
+ x1 = _mm_range_round_ss (x1, x2, 1, _MM_FROUND_NO_EXC);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vreducepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+#define IMM 123
+
+volatile __m512d x1;
+volatile __m256d x2;
+volatile __m128d x3;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm512_reduce_pd (x1, IMM);
+ x2 = _mm256_reduce_pd (x2, IMM);
+ x3 = _mm_reduce_pd (x3, IMM);
+
+ x1 = _mm512_mask_reduce_pd (x1, m, x1, IMM);
+ x2 = _mm256_mask_reduce_pd (x2, m, x2, IMM);
+ x3 = _mm_mask_reduce_pd (x3, m, x3, IMM);
+
+ x1 = _mm512_maskz_reduce_pd (m, x1, IMM);
+ x2 = _mm256_maskz_reduce_pd (m, x2, IMM);
+ x3 = _mm_maskz_reduce_pd (m, x3, IMM);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+#define IMM 0x23
+
+void
+CALC (double *s, double *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ double tmp = (int) (4 * s[i]) / 4.0;
+ r[i] = s[i] - tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN, d) s, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ double res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123.456 * (i + 2000) * sign;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_reduce_pd) (s.x, IMM);
+ res2.x = INTRINSIC (_mask_reduce_pd) (res2.x, mask, s.x, IMM);
+ res3.x = INTRINSIC (_maskz_reduce_pd) (mask, s.x, IMM);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_FP_CHECK (AVX512F_LEN, d) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE (d) (res_ref, mask, SIZE);
+ if (UNION_FP_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO (d) (res_ref, mask, SIZE);
+ if (UNION_FP_CHECK (AVX512F_LEN, d) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vreduceps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+#define IMM 123
+
+volatile __m512 x1;
+volatile __m256 x2;
+volatile __m128 x3;
+volatile __mmask16 m16;
+volatile __mmask8 m8;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm512_reduce_ps (x1, IMM);
+ x2 = _mm256_reduce_ps (x2, IMM);
+ x3 = _mm_reduce_ps (x3, IMM);
+
+ x1 = _mm512_mask_reduce_ps (x1, m16, x1, IMM);
+ x2 = _mm256_mask_reduce_ps (x2, m8, x2, IMM);
+ x3 = _mm_mask_reduce_ps (x3, m8, x3, IMM);
+
+ x1 = _mm512_maskz_reduce_ps (m16, x1, IMM);
+ x2 = _mm256_maskz_reduce_ps (m8, x2, IMM);
+ x3 = _mm_maskz_reduce_ps (m8, x3, IMM);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+#define IMM 0x23
+
+void
+CALC (float *s, float *r)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ float tmp = (int) (4 * s[i]) / 4.0;
+ r[i] = s[i] - tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN,) s, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ float res_ref[SIZE];
+ int i, sign = 1;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ s.a[i] = 123.456 * (i + 2000) * sign;
+ res2.a[i] = DEFAULT_VALUE;
+ sign = -sign;
+ }
+
+ res1.x = INTRINSIC (_reduce_ps) (s.x, IMM);
+ res2.x = INTRINSIC (_mask_reduce_ps) (res2.x, mask, s.x, IMM);
+ res3.x = INTRINSIC (_maskz_reduce_ps) (mask, s.x, IMM);
+
+ CALC (s.a, res_ref);
+
+ if (UNION_FP_CHECK (AVX512F_LEN,) (res1, res_ref))
+ abort ();
+
+ MASK_MERGE () (res_ref, mask, SIZE);
+ if (UNION_FP_CHECK (AVX512F_LEN,) (res2, res_ref))
+ abort ();
+
+ MASK_ZERO () (res_ref, mask, SIZE);
+ if (UNION_FP_CHECK (AVX512F_LEN,) (res3, res_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vreducesd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128d x1, x2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm_reduce_sd (x1, x2, 123);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vreducess\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128 x1, x2;
+volatile __mmask8 m;
+
+void extern
+avx512dq_test (void)
+{
+ x1 = _mm_reduce_ss (x1, x2, 123);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vxorpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512d z;
+volatile __m256d y;
+volatile __m128d x;
+volatile __mmask8 m;
+
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_xor_pd (z, z);
+ z = _mm512_mask_xor_pd (z, m, z, z);
+ z = _mm512_maskz_xor_pd (m, z, z);
+
+ y = _mm256_mask_xor_pd (y, m, y, y);
+ y = _mm256_maskz_xor_pd (m, y, y);
+
+ x = _mm_mask_xor_pd (x, m, x, x);
+ x = _mm_maskz_xor_pd (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 64)
+#include "avx512f-mask-type.h"
+
+void
+CALC (double *src1, double *src2, double *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ long long tmp = (*(long long *) &src1[i]) ^ (*(long long *) &src2[i]);
+ dst[i] = *(double *) &tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN,d) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ double dst_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++) {
+ s1.a[i] = 132.45 * i;
+ s2.a[i] = 43.6 - i * 4.4;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_xor_pd) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_xor_pd) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_xor_pd) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN,d) (res1, dst_ref))
+ abort ();
+
+ MASK_MERGE (d) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,d) (res2, dst_ref))
+ abort ();
+
+ MASK_ZERO (d) (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,d) (res3, dst_ref))
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%zmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vxorps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m512 z;
+volatile __m256 y;
+volatile __m128 x;
+volatile __mmask16 m;
+volatile __mmask8 n;
+
+void extern
+avx512dq_test (void)
+{
+ z = _mm512_xor_ps (z, z);
+ z = _mm512_mask_xor_ps (z, m, z, z);
+ z = _mm512_maskz_xor_ps (m, z, z);
+
+ y = _mm256_mask_xor_ps (y, n, y, y);
+ y = _mm256_maskz_xor_ps (n, y, y);
+
+ x = _mm_mask_xor_ps (x, n, x, x);
+ x = _mm_maskz_xor_ps (n, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -DAVX512DQ" } */
+/* { dg-require-effective-target avx512dq } */
+
+#include "avx512f-helper.h"
+
+#define SIZE (AVX512F_LEN / 32)
+#include "avx512f-mask-type.h"
+
+void
+CALC (float *src1, float *src2, float *dst)
+{
+ int i;
+
+ for (i = 0; i < SIZE; i++)
+ {
+ int tmp = (*(int *) &src1[i]) ^ (*(int *) &src2[i]);
+ dst[i] = *(float *) &tmp;
+ }
+}
+
+void
+TEST (void)
+{
+ UNION_TYPE (AVX512F_LEN,) s1, s2, res1, res2, res3;
+ MASK_TYPE mask = MASK_VALUE;
+ float dst_ref[SIZE];
+ int i;
+
+ for (i = 0; i < SIZE; i++) {
+ s1.a[i] = 132.45 * i;
+ s2.a[i] = 43.6 - i * 4.4;
+ res2.a[i] = DEFAULT_VALUE;
+ }
+
+ res1.x = INTRINSIC (_xor_ps) (s1.x, s2.x);
+ res2.x = INTRINSIC (_mask_xor_ps) (res2.x, mask, s1.x, s2.x);
+ res3.x = INTRINSIC (_maskz_xor_ps) (mask, s1.x, s2.x);
+
+ CALC (s1.a, s2.a, dst_ref);
+
+ if (UNION_CHECK (AVX512F_LEN,) (res1, dst_ref))
+ abort ();
+
+ MASK_MERGE () (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,) (res2, dst_ref))
+ abort ();
+
+ MASK_ZERO () (dst_ref, mask, SIZE);
+ if (UNION_CHECK (AVX512F_LEN,) (res3, dst_ref))
+ abort ();
+}
/* This file is used to reduce a number of runtime tests for AVX512F
- instructions. Idea is to create one file per instruction -
+ and AVX512VL instructions. Idea is to create one file per instruction -
avx512f-insn-2.c - using defines from this file instead of intrinsic
name, vector length etc. Then dg-options are set with appropriate
-Dwhatever options in that .c file producing tests for specific
length. */
-#if defined (AVX512F)
+#ifndef AVX512F_HELPER_INCLUDED
+#define AVX512F_HELPER_INCLUDED
+
+#if defined (AVX512F) && !defined (AVX512VL)
#include "avx512f-check.h"
#elif defined (AVX512ER)
#include "avx512er-check.h"
#elif defined (AVX512CD)
#include "avx512cd-check.h"
+#elif defined (AVX512DQ)
+#include "avx512dq-check.h"
+#elif defined (AVX512BW)
+#include "avx512bw-check.h"
+#elif defined (AVX512VL)
+#include "avx512vl-check.h"
#endif
/* Macros expansion. */
#define MASK_ZERO(TYPE) zero_masking_##TYPE
-/* Intrinsic being tested. */
-#define INTRINSIC(NAME) EVAL(_mm, AVX512F_LEN, NAME)
+
/* Unions used for testing (for example union512d, union256d etc.). */
#define UNION_TYPE(SIZE, NAME) EVAL(union, SIZE, NAME)
/* Corresponding union check. */
/* Function which calculates result. */
#define CALC EVAL(calc_, AVX512F_LEN,)
+#ifndef AVX512VL
#define AVX512F_LEN 512
#define AVX512F_LEN_HALF 256
+void test_512 ();
+#endif
void test_512 ();
+void test_256 ();
+void test_128 ();
-#if defined (AVX512F)
+#if defined (AVX512F) && !defined (AVX512VL)
void
avx512f_test (void) { test_512 (); }
#elif defined (AVX512CD)
#elif defined (AVX512ER)
void
avx512er_test (void) { test_512 (); }
+#elif defined (AVX512DQ)
+void
+avx512dq_test (void) { test_512 (); }
+#elif defined (AVX512BW)
+void
+avx512bw_test (void) { test_512 (); }
+#elif defined (AVX512VL)
+void
+avx512vl_test (void) { test_256 (); test_128 (); }
+#endif
+
+#endif /* AVX512F_HELPER_INCLUDED */
+
+/* Intrinsic being tested. It has different deffinitions,
+ depending on AVX512F_LEN, so it's outside include guards
+ and in undefed away to silence warnings. */
+#if defined INTRINSIC
+#undef INTRINSIC
+#endif
+
+#if AVX512F_LEN != 128
+#define INTRINSIC(NAME) EVAL(_mm, AVX512F_LEN, NAME)
+#else
+#define INTRINSIC(NAME) _mm ## NAME
#endif
/* Type of mask. */
#if SIZE <= 8
+#undef MASK_TYPE
+#undef MASK_VALUE
+#undef MASK_ALL_ONES
#define MASK_TYPE __mmask8
#define MASK_VALUE 0xB9
#define MASK_ALL_ONES 0xFF
#elif SIZE <= 16
+#undef MASK_TYPE
+#undef MASK_VALUE
+#undef MASK_ALL_ONES
#define MASK_TYPE __mmask16
#define MASK_VALUE 0xA6BA
#define MASK_ALL_ONES 0xFFFF
+#elif SIZE <= 32
+#undef MASK_TYPE
+#undef MASK_VALUE
+#undef MASK_ALL_ONES
+#define MASK_TYPE __mmask32
+#define MASK_VALUE 0xA6BAAB6A
+#define MASK_ALL_ONES 0xFFFFFFFFu
+#elif SIZE <= 64
+#undef MASK_TYPE
+#undef MASK_VALUE
+#undef MASK_ALL_ONES
+#define MASK_TYPE __mmask64
+#define MASK_VALUE 0xA6BAA6BAB6AB6ABB
+#define MASK_ALL_ONES 0xFFFFFFFFFFFFFFFFull
#endif
if ((dst_ref & mask) != dst2) abort();
#endif
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 4; i++) \
+ { \
+ dst_ref = (((int) rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_pd(s1); \
+ source2.x = _mm256_loadu_pd(s2); \
+ dst1 = _mm256_cmp_pd_mask(source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_pd_mask(mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((dst_ref & mask) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 2; i++) \
+ { \
+ dst_ref = (((int) rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_pd(s1); \
+ source2.x = _mm_loadu_pd(s2); \
+ dst1 = _mm_cmp_pd_mask(source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_pd_mask(mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((dst_ref & mask) != dst2) abort();
+#endif
+
void
TEST ()
{
#include "avx512f-mask-type.h"
#if AVX512F_LEN == 512
+#undef CMP
#define CMP(imm, rel) \
dst_ref = 0; \
for (i = 0; i < 16; i++) \
if ((dst_ref & mask) != dst2) abort();
#endif
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 8; i++) \
+ { \
+ dst_ref = (((int) rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_ps(s1); \
+ source2.x = _mm256_loadu_ps(s2); \
+ dst1 = _mm256_cmp_ps_mask(source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_ps_mask(mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((dst_ref & mask) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 4; i++) \
+ { \
+ dst_ref = (((int) rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_ps(s1); \
+ source2.x = _mm_loadu_ps(s2); \
+ dst1 = _mm_cmp_ps_mask(source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_ps_mask(mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((dst_ref & mask) != dst2) abort();
+#endif
+
void
TEST ()
{
int i;
for (i = 0; i < SIZE; i++)
e[i] = (float) s1.a[i];
+ for (i = SIZE; i < AVX512F_LEN_HALF / 32; i++)
+ e[i] = 0.0;
}
void
UNION_TYPE (AVX512F_LEN, d) s1;
UNION_TYPE (AVX512F_LEN_HALF,) u1, u2, u3;
MASK_TYPE mask = MASK_VALUE;
- float e[SIZE];
+ float e[AVX512F_LEN_HALF / 32];
int i;
for (i = 0; i < SIZE; i++)
{
int i;
- for (i = 0; i < DST_SIZE; i++)
+ for (i = 0; i < SRC_SIZE; i++)
{
r[i] = (unsigned) (s[i] + 0.5);
}
+ for (i = SRC_SIZE; i < DST_SIZE; i++)
+ {
+ r[i] = 0;
+ }
}
void
val.a[15] = 0xc800;
#endif
- res1.x = _mm512_cvtph_ps (val.x);
- res2.x = _mm512_mask_cvtph_ps (res2.x, mask, val.x);
- res3.x = _mm512_maskz_cvtph_ps (mask, val.x);
+ res1.x = INTRINSIC (_cvtph_ps) (val.x);
+ res2.x = INTRINSIC (_mask_cvtph_ps) (res2.x, mask, val.x);
+ res3.x = INTRINSIC (_maskz_cvtph_ps) (mask, val.x);
if (UNION_CHECK (AVX512F_LEN,) (res1, exp))
abort ();
UNION_TYPE (AVX512F_LEN,) val;
UNION_TYPE (AVX512F_LEN_HALF, i_w) res1,res2,res3;
MASK_TYPE mask = MASK_VALUE;
+#if AVX512F_LEN == 128
+ short exp[SIZE * 2];
+#else
short exp[SIZE];
+#endif
int i;
for (i = 0; i < SIZE; i++)
exp[5] = 0xc000;
exp[6] = 0xc400;
exp[7] = 0xc800;
+#else
+ exp[4] = 0;
+ exp[5] = 0;
+ exp[6] = 0;
+ exp[7] = 0;
#endif
#if AVX512F_LEN > 256
exp[8] = 0x3c00;
exp[15] = 0xc800;
#endif
- res1.x = _mm512_cvtps_ph (val.x, 0);
- res2.x = _mm512_mask_cvtps_ph (res2.x, mask, val.x, 0);
- res3.x = _mm512_maskz_cvtps_ph (mask, val.x, 0);
+ res1.x = INTRINSIC (_cvtps_ph (val.x, 0));
+ res2.x = INTRINSIC (_mask_cvtps_ph (res2.x, mask, val.x, 0));
+ res3.x = INTRINSIC (_maskz_cvtps_ph (mask, val.x, 0));
if (UNION_CHECK (AVX512F_LEN_HALF, i_w) (res1, exp))
abort ();
{
int i;
- for (i = 0; i < DST_SIZE; i++)
- {
+ for (i = 0; i < SRC_SIZE; i++)
r[i] = (unsigned) s[i];
- }
+ for (i = SRC_SIZE; i < DST_SIZE; i++)
+ r[i] = 0;
}
void
res3.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_and_si512) (s1.x, s2.x);
res2.x = INTRINSIC (_and_epi32) (s1.x, s2.x);
+#endif
res3.x = INTRINSIC (_mask_and_epi32) (res3.x, mask, s1.x, s2.x);
res4.x = INTRINSIC (_maskz_and_epi32) (mask, s1.x, s2.x);
CALC (s1.a, s2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
abort ();
+#endif
MASK_MERGE (i_d) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
res3.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_andnot_si512) (s1.x, s2.x);
res2.x = INTRINSIC (_andnot_epi32) (s1.x, s2.x);
+#endif
res3.x = INTRINSIC (_mask_andnot_epi32) (res3.x, mask, s1.x, s2.x);
res4.x = INTRINSIC (_maskz_andnot_epi32) (mask, s1.x, s2.x);
CALC (s1.a, s2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
abort ();
+#endif
MASK_MERGE (i_d) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
res2.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_andnot_epi64) (s1.x, s2.x);
+#endif
res2.x = INTRINSIC (_mask_andnot_epi64) (res2.x, mask, s1.x, s2.x);
res3.x = INTRINSIC (_maskz_andnot_epi64) (mask, s1.x, s2.x);
CALC (s1.a, s2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
abort ();
+#endif
MASK_MERGE (i_q) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
res2.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_and_epi64) (s1.x, s2.x);
+#endif
res2.x = INTRINSIC (_mask_and_epi64) (res2.x, mask, s1.x, s2.x);
res3.x = INTRINSIC (_maskz_and_epi64) (mask, s1.x, s2.x);
CALC (s1.a, s2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
abort ();
+#endif
MASK_MERGE (i_q) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
if (UNION_CHECK (AVX512F_LEN, i_q) (res3, res_ref))
abort ();
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_set1_epi64) (src.a[0]);
+#endif
res2.x = INTRINSIC (_mask_set1_epi64) (res2.x, mask, src.a[0]);
res3.x = INTRINSIC (_maskz_set1_epi64) (mask, src.a[0]);
#include "avx512f-mask-type.h"
#if AVX512F_LEN == 512
+#undef CMP
#define CMP(imm, rel) \
dst_ref = 0; \
for (i = 0; i < 16; i++) \
if ((mask & dst_ref) != dst2) abort();
#endif
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 8; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_si256 ((__m256i*)s1); \
+ source2.x = _mm256_loadu_si256 ((__m256i*)s2); \
+ dst1 = _mm256_cmp_epi32_mask (source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_epi32_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 4; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_si128 ((__m128i*)s1); \
+ source2.x = _mm_loadu_si128 ((__m128i*)s2); \
+ dst1 = _mm_cmp_epi32_mask (source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_epi32_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
void
TEST ()
{
__mmask8 dst_ref;
+#if AVX512F_LEN == 512
+#undef CMP
#define CMP(imm, rel) \
dst_ref = 0; \
for (i = 0; i < 8; i++) \
dst2 = _mm512_mask_cmp_epi64_mask (mask, source1.x, source2.x, imm);\
if (dst_ref != dst1) abort(); \
if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 4; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_si256 ((__m256i*)s1); \
+ source2.x = _mm256_loadu_si256 ((__m256i*)s2); \
+ dst1 = _mm256_cmp_epi64_mask (source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_epi64_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 2; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_si128 ((__m128i*)s1); \
+ source2.x = _mm_loadu_si128 ((__m128i*)s2); \
+ dst1 = _mm_cmp_epi64_mask (source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_epi64_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
void
TEST ()
#include "avx512f-mask-type.h"
#if AVX512F_LEN == 512
+#undef CMP
#define CMP(imm, rel) \
dst_ref = 0; \
for (i = 0; i < 16; i++) \
if ((mask & dst_ref) != dst2) abort();
#endif
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 8; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_si256 ((__m256i*)s1); \
+ source2.x = _mm256_loadu_si256 ((__m256i*)s2); \
+ dst1 = _mm256_cmp_epu32_mask (source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_epu32_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 4; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_si128 ((__m128i*)s1); \
+ source2.x = _mm_loadu_si128 ((__m128i*)s2); \
+ dst1 = _mm_cmp_epu32_mask (source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_epu32_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
void
TEST ()
{
#include "avx512f-mask-type.h"
#if AVX512F_LEN == 512
+#undef CMP
#define CMP(imm, rel) \
dst_ref = 0; \
for (i = 0; i < 8; i++) \
if ((mask & dst_ref) != dst2) abort();
#endif
+#if AVX512F_LEN == 256
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 4; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm256_loadu_si256 ((__m256i*)s1); \
+ source2.x = _mm256_loadu_si256 ((__m256i*)s2); \
+ dst1 = _mm256_cmp_epu64_mask (source1.x, source2.x, imm);\
+ dst2 = _mm256_mask_cmp_epu64_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
+#if AVX512F_LEN == 128
+#undef CMP
+#define CMP(imm, rel) \
+ dst_ref = 0; \
+ for (i = 0; i < 2; i++) \
+ { \
+ dst_ref = ((rel) << i) | dst_ref; \
+ } \
+ source1.x = _mm_loadu_si128 ((__m128i*)s1); \
+ source2.x = _mm_loadu_si128 ((__m128i*)s2); \
+ dst1 = _mm_cmp_epu64_mask (source1.x, source2.x, imm);\
+ dst2 = _mm_mask_cmp_epu64_mask (mask, source1.x, source2.x, imm);\
+ if (dst_ref != dst1) abort(); \
+ if ((mask & dst_ref) != dst2) abort();
+#endif
+
void
TEST ()
{
for (i = 0; i < SIZE; i++)
{
+#if AVX512F_LEN == 512
dst[i] = src1[mask[i] & 15];
+#else
+ dst[i] = src1[mask[i] & 7];
+#endif
}
}
res3.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_permutexvar_epi32) (src1.x, src2.x);
+#endif
res2.x = INTRINSIC (_maskz_permutexvar_epi32) (mask, src1.x, src2.x);
res3.x = INTRINSIC (_mask_permutexvar_epi32) (res3.x, mask, src1.x, src2.x);
CALC (src1.a, src2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
abort ();
+#endif
MASK_ZERO (i_d) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
#define SIZE (AVX512F_LEN / 64)
#include "avx512f-mask-type.h"
-#ifndef CTRL
-#define CTRL 129
+#if AVX512F_LEN > 256
+ #define CTRL 129
+#elif AVX512F_LEN > 128
+ #define CTRL 5
+#else
+ #define CTRL 1
#endif
static void
if (UNION_CHECK (AVX512F_LEN, d) (res3, res_ref))
abort ();
}
+
+#undef CTRL
res3.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_permutex_epi64) (src1.x, IMM_MASK);
+#endif
res2.x = INTRINSIC (_maskz_permutex_epi64) (mask, src1.x, IMM_MASK);
res3.x = INTRINSIC (_mask_permutex_epi64) (res3.x, mask, src1.x, IMM_MASK);
CALC (src1.a, IMM_MASK, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
abort ();
+#endif
MASK_ZERO (i_q) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
for (i = 0; i < SIZE; i++)
{
+#if AVX512F_LEN == 512
dst[i] = src1[mask[i] & 7];
+#else
+ dst[i] = src1[mask[i] & 3];
+#endif
}
}
res3.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_permutexvar_epi64) (src1.x, src2.x);
+#endif
res2.x = INTRINSIC (_maskz_permutexvar_epi64) (mask, src1.x, src2.x);
res3.x = INTRINSIC (_mask_permutexvar_epi64) (res3.x, mask, src1.x, src2.x);
CALC (src1.a, src2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
abort ();
+#endif
MASK_ZERO (i_q) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
#define SIZE (AVX512F_LEN / 32)
#include "avx512f-mask-type.h"
-static void
-CALC (char *r, int *s)
+void
+CALC (char *r, int *s, int mem)
{
int i;
- for (i = 0; i < 16; i++)
+ int len = mem ? SIZE : 16;
+ for (i = 0; i < len; i++)
{
r[i] = (i < SIZE) ? (char) s[i] : 0;
}
UNION_TYPE (AVX512F_LEN, i_d) src;
MASK_TYPE mask = MASK_VALUE;
char res_ref[16];
+ char res_ref2[16];
sign = -1;
for (i = 0; i < SIZE; i++)
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < 16; i++)
+ {
+ res4[i] = DEFAULT_VALUE * 2;
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtepi32_epi8) (src.x);
res2.x = INTRINSIC (_mask_cvtepi32_epi8) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtepi32_epi8) (mask, src.x);
- INTRINSIC (_mask_cvtepi32_storeu_epi8) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (128, i_b) (res1, res_ref))
abort ();
if (UNION_CHECK (128, i_b) (res2, res_ref))
abort ();
- if (checkVc (res4, res_ref, 16))
- abort ();
-
MASK_ZERO (i_b) (res_ref, mask, SIZE);
if (UNION_CHECK (128, i_b) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtepi32_storeu_epi8) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_b) (res_ref2, mask, SIZE);
+ if (checkVc (res4, res_ref2, 16))
+ abort ();
}
#include "avx512f-mask-type.h"
#define SIZE_HALF (AVX512F_LEN_HALF / 16)
-void static
-CALC (short *r, int *s)
+void
+CALC (short *r, int *s, int mem)
{
int i;
- for (i = 0; i < SIZE_HALF; i++)
+ int len = mem ? SIZE : SIZE_HALF;
+ for (i = 0; i < len; i++)
{
r[i] = (i < SIZE) ? (short) s[i] : 0;
}
UNION_TYPE (AVX512F_LEN, i_d) src;
MASK_TYPE mask = MASK_VALUE;
short res_ref[SIZE_HALF];
+ short res_ref2[SIZE_HALF];
sign = -1;
for (i = 0; i < SIZE; i++)
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < SIZE_HALF; i++)
+ {
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ res4[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtepi32_epi16) (src.x);
res2.x = INTRINSIC (_mask_cvtepi32_epi16) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtepi32_epi16) (mask, src.x);
- INTRINSIC (_mask_cvtepi32_storeu_epi16) (res4, mask, src.x);
-
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (AVX512F_LEN_HALF, i_w) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN_HALF, i_w) (res2, res_ref))
abort ();
- if (checkVs (res4, res_ref, SIZE_HALF))
- abort ();
-
MASK_ZERO (i_w) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN_HALF, i_w) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtepi32_storeu_epi16) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_w) (res_ref2, mask, SIZE);
+ if (checkVs (res4, res_ref2, SIZE_HALF))
+ abort ();
}
{
int i;
/* Don't zero out upper half if destination is memory. */
- int len = mem ? 8 : 16;
+ int len = mem ? SIZE : 16;
for (i = 0; i < len; i++)
{
r[i] = (i < SIZE) ? (char) s[i] : 0;
#include "avx512f-mask-type.h"
#define SIZE_HALF (AVX512F_LEN_HALF / 32)
-void static
-CALC (int *r, long long *s)
+void
+CALC (int *r, long long *s, int mem)
{
int i;
- for (i = 0; i < SIZE_HALF; i++)
+ int len = mem ? SIZE : SIZE_HALF;
+ for (i = 0; i < len; i++)
{
r[i] = (i < SIZE) ? (int) s[i] : 0;
}
UNION_TYPE (AVX512F_LEN, i_q) src;
MASK_TYPE mask = MASK_VALUE;
int res_ref[SIZE_HALF];
+ int res_ref2[SIZE_HALF];
sign = -1;
for (i = 0; i < SIZE; i++)
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < SIZE_HALF; i++)
+ {
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ res4[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtepi64_epi32) (src.x);
res2.x = INTRINSIC (_mask_cvtepi64_epi32) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtepi64_epi32) (mask, src.x);
- INTRINSIC (_mask_cvtepi64_storeu_epi32) (res4, mask, src.x);
-
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res2, res_ref))
abort ();
- if (checkVi (res4, res_ref, SIZE_HALF))
- abort ();
-
MASK_ZERO (i_d) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtepi64_storeu_epi32) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_d) (res_ref2, mask, SIZE);
+ if (checkVi (res4, res_ref2, SIZE_HALF))
+ abort ();
}
#define SIZE (AVX512F_LEN / 64)
#include "avx512f-mask-type.h"
-static void
-CALC (short *r, long long *s)
+void
+CALC (short *r, long long *s, int mem)
{
int i;
- for (i = 0; i < 8; i++)
+ int len = mem ? SIZE : 8;
+ for (i = 0; i < len; i++)
{
r[i] = (i < SIZE) ? (short) s[i] : 0;
}
UNION_TYPE (AVX512F_LEN, i_q) src;
MASK_TYPE mask = MASK_VALUE;
short res_ref[8];
+ short res_ref2[8];
sign = -1;
for (i = 0; i < SIZE; i++)
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < 8; i++)
+ {
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ res4[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtepi64_epi16) (src.x);
res2.x = INTRINSIC (_mask_cvtepi64_epi16) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtepi64_epi16) (mask, src.x);
- INTRINSIC (_mask_cvtepi64_storeu_epi16) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (128, i_w) (res1, res_ref))
abort ();
if (UNION_CHECK (128, i_w) (res2, res_ref))
abort ();
- if (checkVs (res4, res_ref, 8))
- abort ();
-
MASK_ZERO (i_w) (res_ref, mask, SIZE);
if (UNION_CHECK (128, i_w) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtepi64_storeu_epi16) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_w) (res_ref2, mask, SIZE);
+ if (checkVs (res4, res_ref2, 8))
+ abort ();
}
#include "avx512f-mask-type.h"
#include <limits.h>
-static void
-CALC (char *r, int *s)
+void
+CALC (char *r, int *s, int mem)
{
int i;
- for (i = 0; i < 16; i++)
+ int len = mem ? SIZE : 16;
+ for (i = 0; i < len; i++)
{
- if (s[i] < CHAR_MIN)
- r[i] = CHAR_MIN;
- else if (s[i] > CHAR_MAX)
- r[i] = CHAR_MAX;
+ if (i < SIZE)
+ {
+ if (s[i] < CHAR_MIN)
+ r[i] = CHAR_MIN;
+ else if (s[i] > CHAR_MAX)
+ r[i] = CHAR_MAX;
+ else
+ r[i] = s[i];
+ }
else
- r[i] = s[i];
- r[i] = (i < SIZE) ? r[i] : 0;
+ r[i] = 0;
}
}
UNION_TYPE (AVX512F_LEN, i_d) src;
MASK_TYPE mask = MASK_VALUE;
char res_ref[16];
+ char res_ref2[16];
sign = -1;
for (i = 0; i < SIZE; i++)
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < 16; i++)
+ {
+ res4[i] = DEFAULT_VALUE * 2;
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtsepi32_epi8) (src.x);
res2.x = INTRINSIC (_mask_cvtsepi32_epi8) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtsepi32_epi8) (mask, src.x);
- INTRINSIC (_mask_cvtsepi32_storeu_epi8) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (128, i_b) (res1, res_ref))
abort ();
if (UNION_CHECK (128, i_b) (res2, res_ref))
abort ();
- if (checkVc (res4, res_ref, 16))
- abort ();
-
MASK_ZERO (i_b) (res_ref, mask, SIZE);
if (UNION_CHECK (128, i_b) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtsepi32_storeu_epi8) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_b) (res_ref2, mask, SIZE);
+ if (checkVc (res4, res_ref2, 16))
+ abort ();
}
#define SIZE_HALF (AVX512F_LEN_HALF / 16)
#include <limits.h>
-static void
-CALC (short *r, int *s)
+void
+CALC (short *r, int *s, int mem)
{
int i;
- for (i = 0; i < SIZE_HALF; i++)
+ int len = mem ? SIZE : SIZE_HALF;
+ for (i = 0; i < len; i++)
{
- if (s[i] < SHRT_MIN)
- r[i] = SHRT_MIN;
- else if (s[i] > SHRT_MAX)
- r[i] = SHRT_MAX;
+ if (i < SIZE)
+ {
+ if (s[i] < SHRT_MIN)
+ r[i] = SHRT_MIN;
+ else if (s[i] > SHRT_MAX)
+ r[i] = SHRT_MAX;
+ else
+ r[i] = s[i];
+ }
else
- r[i] = s[i];
- r[i] = (i < SIZE) ? r[i] : 0;
+ r[i] = 0;
}
}
UNION_TYPE (AVX512F_LEN, i_d) src;
MASK_TYPE mask = MASK_VALUE;
short res_ref[SIZE_HALF];
+ short res_ref2[SIZE_HALF];
sign = -1;
for (i = 0; i < SIZE; i++)
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < SIZE_HALF; i++)
+ {
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ res4[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtsepi32_epi16) (src.x);
res2.x = INTRINSIC (_mask_cvtsepi32_epi16) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtsepi32_epi16) (mask, src.x);
- INTRINSIC (_mask_cvtsepi32_storeu_epi16) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (AVX512F_LEN_HALF, i_w) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN_HALF, i_w) (res2, res_ref))
abort ();
- if (checkVs (res4, res_ref, SIZE_HALF))
- abort ();
-
MASK_ZERO (i_w) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN_HALF, i_w) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtsepi32_storeu_epi16) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_w) (res_ref2, mask, SIZE);
+ if (checkVs (res4, res_ref2, SIZE_HALF))
+ abort ();
}
CALC (char *r, long long *s, int mem)
{
int i;
- int len = mem ? 8 : 16;
+ int len = mem ? SIZE : 16;
for (i = 0; i < len; i++)
{
if (s[i] < CHAR_MIN)
#define SIZE_HALF (AVX512F_LEN_HALF / 32)
#include <limits.h>
-static void
-CALC (int *r, long long *s)
+void
+CALC (int *r, long long *s, int mem)
{
int i;
- for (i = 0; i < SIZE_HALF; i++)
+ int len = mem ? SIZE : SIZE_HALF;
+ for (i = 0; i < len; i++)
{
- if (s[i] < INT_MIN)
- r[i] = INT_MIN;
- else if (s[i] > INT_MAX)
- r[i] = INT_MAX;
+ if (i < SIZE)
+ {
+ if (s[i] < INT_MIN)
+ r[i] = INT_MIN;
+ else if (s[i] > INT_MAX)
+ r[i] = INT_MAX;
+ else
+ r[i] = s[i];
+ }
else
- r[i] = s[i];
- r[i] = (i < SIZE) ? r[i] : 0;
+ r[i] = 0;
}
}
UNION_TYPE (AVX512F_LEN, i_q) src;
MASK_TYPE mask = MASK_VALUE;
int res_ref[SIZE_HALF];
+ int res_ref2[SIZE_HALF];
sign = -1;
for (i = 0; i < SIZE; i++)
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < SIZE_HALF; i++)
+ {
+ res4[i] = DEFAULT_VALUE * 2;
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtsepi64_epi32) (src.x);
res2.x = INTRINSIC (_mask_cvtsepi64_epi32) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtsepi64_epi32) (mask, src.x);
- INTRINSIC (_mask_cvtsepi64_storeu_epi32) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res2, res_ref))
abort ();
- if (checkVi (res4, res_ref, SIZE_HALF))
- abort ();
-
MASK_ZERO (i_d) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN_HALF, i_d) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtsepi64_storeu_epi32) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_d) (res_ref2, mask, SIZE);
+ if (checkVi (res4, res_ref2, SIZE_HALF))
+ abort ();
}
#include "avx512f-mask-type.h"
#include <limits.h>
-void static
-CALC (short *r, long long *s)
+void
+CALC (short *r, long long *s, int mem)
{
int i;
- for (i = 0; i < 8; i++)
+ int len = mem ? SIZE : 8;
+ for (i = 0; i < len; i++)
{
if (s[i] < SHRT_MIN)
r[i] = SHRT_MIN;
UNION_TYPE (AVX512F_LEN, i_q) src;
MASK_TYPE mask = MASK_VALUE;
short res_ref[8];
+ short res_ref2[8];
sign = -1;
for (i = 0; i < SIZE; i++)
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < 8; i++)
+ {
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ res4[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtsepi64_epi16) (src.x);
res2.x = INTRINSIC (_mask_cvtsepi64_epi16) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtsepi64_epi16) (mask, src.x);
- INTRINSIC (_mask_cvtsepi64_storeu_epi16) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (128, i_w) (res1, res_ref))
abort ();
if (UNION_CHECK (128, i_w) (res2, res_ref))
abort ();
- if (checkVs (res4, res_ref, 8))
- abort ();
-
MASK_ZERO (i_w) (res_ref, mask, SIZE);
if (UNION_CHECK (128, i_w) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtsepi64_storeu_epi16) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_w) (res_ref2, mask, SIZE);
+ if (checkVs (res4, res_ref2, 8))
+ abort ();
}
#include "avx512f-mask-type.h"
#include <limits.h>
-static void
-CALC (unsigned char *r, unsigned int *s)
+void
+CALC (unsigned char *r, unsigned int *s, int mem)
{
int i;
- for (i = 0; i < 16; i++)
- {
- r[i] = (s[i] > UCHAR_MAX) ? UCHAR_MAX : s[i];
- r[i] = (i < SIZE) ? r[i] : 0;
- }
+ int len = mem ? SIZE : 16;
+ for (i = 0; i < len; i++)
+ r[i] = (i < SIZE) ? ((s[i] > UCHAR_MAX) ? UCHAR_MAX : s[i]) : 0;
}
void
UNION_TYPE (AVX512F_LEN, i_ud) src;
MASK_TYPE mask = MASK_VALUE;
unsigned char res_ref[16];
+ unsigned char res_ref2[16];
for (i = 0; i < SIZE; i++)
{
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < 16; i++)
+ {
+ res4[i] = DEFAULT_VALUE * 2;
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtusepi32_epi8) (src.x);
res2.x = INTRINSIC (_mask_cvtusepi32_epi8) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtusepi32_epi8) (mask, src.x);
- INTRINSIC (_mask_cvtusepi32_storeu_epi8) (res4, mask, src.x);
-
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (128, i_ub) (res1, res_ref))
abort ();
if (UNION_CHECK (128, i_ub) (res2, res_ref))
abort ();
- if (checkVuc (res4, res_ref, 16))
- abort ();
-
MASK_ZERO (i_ub) (res_ref, mask, SIZE);
if (UNION_CHECK (128, i_ub) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtusepi32_storeu_epi8) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_b) (res_ref2, mask, SIZE);
+ if (checkVc (res4, res_ref2, 16))
+ abort ();
}
#define SIZE_HALF (AVX512F_LEN_HALF / 16)
#include <limits.h>
-static void
-CALC (unsigned short *r, unsigned int *s)
+void
+CALC (unsigned short *r, unsigned int *s, int mem)
{
int i;
- for (i = 0; i < SIZE_HALF; i++)
- {
- r[i] = (s[i] > USHRT_MAX) ? USHRT_MAX : s[i];
- r[i] = (i < SIZE) ? r[i] : 0;
- }
+ int len = mem ? SIZE : SIZE_HALF;
+ for (i = 0; i < len; i++)
+ r[i] = (i < SIZE) ? ((s[i] > USHRT_MAX) ? USHRT_MAX : s[i]) : 0;
}
void
UNION_TYPE (AVX512F_LEN, i_ud) src;
MASK_TYPE mask = MASK_VALUE;
unsigned short res_ref[SIZE_HALF];
+ unsigned short res_ref2[SIZE_HALF];
for (i = 0; i < SIZE; i++)
{
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < SIZE_HALF; i++)
+ {
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ res4[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtusepi32_epi16) (src.x);
res2.x = INTRINSIC (_mask_cvtusepi32_epi16) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtusepi32_epi16) (mask, src.x);
- INTRINSIC (_mask_cvtusepi32_storeu_epi16) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (AVX512F_LEN_HALF, i_uw) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN_HALF, i_uw) (res2, res_ref))
abort ();
- if (checkVus (res4, res_ref, SIZE_HALF))
- abort ();
-
MASK_ZERO (i_uw) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN_HALF, i_uw) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtusepi32_storeu_epi16) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_w) (res_ref2, mask, SIZE);
+ if (checkVs (res4, res_ref2, SIZE_HALF))
+ abort ();
}
CALC (unsigned char *r, unsigned long long *s, int mem)
{
int i;
- int len = mem ? 8 : 16;
+ int len = mem ? SIZE : 16;
for (i = 0; i < len; i++)
- {
- r[i] = (s[i] > UCHAR_MAX) ? UCHAR_MAX : s[i];
- r[i] = (i < SIZE) ? r[i] : 0;
- }
+ r[i] = (i < SIZE) ? ((s[i] > UCHAR_MAX) ? UCHAR_MAX : s[i]) : 0;
}
void
#define SIZE_HALF (AVX512F_LEN_HALF / 32)
#include <limits.h>
-static void
-CALC (unsigned int *r, unsigned long long *s)
+void
+CALC (unsigned int *r, unsigned long long *s, int mem)
{
int i;
- for (i = 0; i < SIZE_HALF; i++)
- {
- r[i] = (s[i] > UINT_MAX) ? UINT_MAX : s[i];
- r[i] = (i < SIZE) ? r[i] : 0;
- }
+ int len = mem ? SIZE : SIZE_HALF;
+ for (i = 0; i < len; i++)
+ r[i] = (i < SIZE) ? ((s[i] > UINT_MAX) ? UINT_MAX : s[i]) : 0;
}
void
UNION_TYPE (AVX512F_LEN, i_uq) src;
MASK_TYPE mask = MASK_VALUE;
unsigned int res_ref[SIZE_HALF];
+ unsigned int res_ref2[SIZE_HALF];
for (i = 0; i < SIZE; i++)
{
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < SIZE_HALF; i++)
+ {
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ res4[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtusepi64_epi32) (src.x);
res2.x = INTRINSIC (_mask_cvtusepi64_epi32) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtusepi64_epi32) (mask, src.x);
- INTRINSIC (_mask_cvtusepi64_storeu_epi32) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (AVX512F_LEN_HALF, i_ud) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN_HALF, i_ud) (res2, res_ref))
abort ();
- if (checkVui (res4, res_ref, SIZE_HALF))
- abort ();
-
MASK_ZERO (i_ud) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN_HALF, i_ud) (res3, res_ref))
abort ();
+
+ CALC (res_ref2, src.a, 1);
+ INTRINSIC (_mask_cvtusepi64_storeu_epi32) (res4, mask, src.x);
+
+ MASK_MERGE (i_d) (res_ref2, mask, SIZE);
+ if (checkVi (res4, res_ref2, SIZE_HALF))
+ abort ();
}
#include "avx512f-mask-type.h"
#include <limits.h>
-static void
-CALC (unsigned short *r, unsigned long long *s)
+void
+CALC (unsigned short *r, unsigned long long *s, int mem)
{
int i;
- for (i = 0; i < 8; i++)
- {
- r[i] = (s[i] > USHRT_MAX) ? USHRT_MAX : s[i];
- r[i] = (i < SIZE) ? r[i] : 0;
- }
+ int len = mem ? SIZE : 8;
+ for (i = 0; i < len; i++)
+ r[i] = (i < SIZE) ? ((s[i] > USHRT_MAX) ? USHRT_MAX : s[i]) : 0;
}
void
UNION_TYPE (AVX512F_LEN, i_uq) src;
MASK_TYPE mask = MASK_VALUE;
unsigned short res_ref[8];
+ unsigned short res_ref2[8];
for (i = 0; i < SIZE; i++)
{
res4[i] = DEFAULT_VALUE;
}
+ for (i = SIZE; i < 8; i++)
+ {
+ res_ref2[i] = DEFAULT_VALUE * 2;
+ res4[i] = DEFAULT_VALUE * 2;
+ }
+
res1.x = INTRINSIC (_cvtusepi64_epi16) (src.x);
res2.x = INTRINSIC (_mask_cvtusepi64_epi16) (res2.x, mask, src.x);
res3.x = INTRINSIC (_maskz_cvtusepi64_epi16) (mask, src.x);
- INTRINSIC (_mask_cvtusepi64_storeu_epi16) (res4, mask, src.x);
- CALC (res_ref, src.a);
+ CALC (res_ref, src.a, 0);
if (UNION_CHECK (128, i_uw) (res1, res_ref))
abort ();
if (UNION_CHECK (128, i_uw) (res2, res_ref))
abort ();
- if (checkVus (res4, res_ref, 8))
- abort ();
-
MASK_ZERO (i_uw) (res_ref, mask, SIZE);
if (UNION_CHECK (128, i_uw) (res3, res_ref))
abort ();
+
+ INTRINSIC (_mask_cvtusepi64_storeu_epi16) (res4, mask, src.x);
+ CALC (res_ref2, src.a, 1);
+
+ MASK_MERGE (i_w) (res_ref2, mask, SIZE);
+ if (checkVs (res4, res_ref2, 8))
+ abort ();
}
res3.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_or_si512) (s1.x, s2.x);
res2.x = INTRINSIC (_or_epi32) (s1.x, s2.x);
+#endif
res3.x = INTRINSIC (_mask_or_epi32) (res3.x, mask, s1.x, s2.x);
res4.x = INTRINSIC (_maskz_or_epi32) (mask, s1.x, s2.x);
CALC (s1.a, s2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
abort ();
+#endif
MASK_MERGE (i_d) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
sign = -sign;
res2.a[i] = DEFAULT_VALUE;
}
-
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_or_epi64) (s1.x, s2.x);
+#endif
res2.x = INTRINSIC (_mask_or_epi64) (res2.x, mask, s1.x, s2.x);
res3.x = INTRINSIC (_maskz_or_epi64) (mask, s1.x, s2.x);
CALC (s1.a, s2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
abort ();
+#endif
MASK_MERGE (i_q) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
res3.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_xor_si512) (s1.x, s2.x);
res2.x = INTRINSIC (_xor_epi32) (s1.x, s2.x);
+#endif
res3.x = INTRINSIC (_mask_xor_epi32) (res3.x, mask, s1.x, s2.x);
res4.x = INTRINSIC (_maskz_xor_epi32) (mask, s1.x, s2.x);
CALC (s1.a, s2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
abort ();
if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref))
abort ();
+#endif
MASK_MERGE (i_d) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref))
res2.a[i] = DEFAULT_VALUE;
}
+#if AVX512F_LEN == 512
res1.x = INTRINSIC (_xor_epi64) (s1.x, s2.x);
+#endif
res2.x = INTRINSIC (_mask_xor_epi64) (res2.x, mask, s1.x, s2.x);
res3.x = INTRINSIC (_maskz_xor_epi64) (mask, s1.x, s2.x);
CALC (s1.a, s2.a, res_ref);
+#if AVX512F_LEN == 512
if (UNION_CHECK (AVX512F_LEN, i_q) (res1, res_ref))
- abort ();
+ abort ();
+#endif
MASK_MERGE (i_q) (res_ref, mask, SIZE);
if (UNION_CHECK (AVX512F_LEN, i_q) (res2, res_ref))
case 1:
imm = _MM_FROUND_FLOOR;
res1.x = INTRINSIC (_floor_pd) (s.x);
- res2.x = INTRINSIC (_mask_floor_pd) (res2.x, mask, s.x);
+ #if AVX512F_LEN == 512
+ res2.x = INTRINSIC (_mask_floor_pd) (res2.x, mask, s.x);
+ #endif
break;
case 2:
imm = _MM_FROUND_CEIL;
res1.x = INTRINSIC (_ceil_pd) (s.x);
- res2.x = INTRINSIC (_mask_ceil_pd) (res2.x, mask, s.x);
+ #if AVX512F_LEN == 512
+ res2.x = INTRINSIC (_mask_ceil_pd) (res2.x, mask, s.x);
+ #endif
break;
}
MASK_MERGE(d) (res_ref,mask,SIZE );
- if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
- abort ();
+ #if AVX512F_LEN == 512
+ if (UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+ #else
+ if (!i && UNION_CHECK (AVX512F_LEN, d) (res2, res_ref))
+ abort ();
+ #endif
MASK_ZERO(d) (res_ref,mask,SIZE );
UNION_TYPE (AVX512F_LEN,) res1, res2, res3, s;
float res_ref[SIZE];
- MASK_TYPE mask = 6 ^ (0xffff >> SIZE);
+ MASK_TYPE mask = MASK_VALUE;
imm = _MM_FROUND_FLOOR | (7 << 4);
case 1:
imm = _MM_FROUND_FLOOR;
res1.x = INTRINSIC (_floor_ps) (s.x);
- res2.x = INTRINSIC (_mask_floor_ps) (res2.x, mask, s.x);
+ #if AVX512F_LEN == 512
+ res2.x = INTRINSIC (_mask_floor_ps) (res2.x, mask, s.x);
+ #endif
break;
case 2:
imm = _MM_FROUND_CEIL;
res1.x = INTRINSIC (_ceil_ps) (s.x);
- res2.x = INTRINSIC (_mask_ceil_ps) (res2.x, mask, s.x);
+ #if AVX512F_LEN == 512
+ res2.x = INTRINSIC (_mask_ceil_ps) (res2.x, mask, s.x);
+ #endif
break;
}
MASK_MERGE ()(res_ref, mask, SIZE);
- if (UNION_CHECK (AVX512F_LEN,) (res2, res_ref))
- abort ();
+ #if AVX512F_LEN == 512
+ if (UNION_CHECK (AVX512F_LEN,) (res2, res_ref))
+ abort ();
+ #else
+ if (!i && UNION_CHECK (AVX512F_LEN,) (res2, res_ref))
+ abort ();
+ #endif
MASK_ZERO ()(res_ref, mask, SIZE);
MASK_TYPE mask = MASK_VALUE;
float e[SIZE];
int i;
+#if AVX512F_LEN == 512
int imm = 203;
+#else
+ int imm = 1;
+#endif
for (i = 0; i < SIZE; i++)
{
MASK_TYPE mask = MASK_VALUE;
double e[SIZE];
int i;
+#if AVX512F_LEN == 512
int imm = 203;
+#else
+ int imm = 1;
+#endif
for (i = 0; i < SIZE; i++)
{
MASK_TYPE mask = MASK_VALUE;
int e[SIZE];
int i;
+#if AVX512F_LEN == 512
int imm = 203;
+#else
+ int imm = 1;
+#endif
for (i = 0; i < SIZE; i++)
{
MASK_TYPE mask = MASK_VALUE;
long long e[SIZE];
int i;
+#if AVX512F_LEN == 512
int imm = 203;
+#else
+ int imm = 1;
+#endif
for (i = 0; i < SIZE; i++)
{
#define SIZE (AVX512F_LEN / 64)
#include "avx512f-mask-type.h"
+
+#if AVX512F_LEN > 256
+ #define IMM_VAR 120
+#elif AVX512F_LEN > 128
+ #define IMM_VAR 5
+#else
+ #define IMM_VAR 1
+#endif
+
void static
CALC (double *e, UNION_TYPE (AVX512F_LEN, d) s1,
UNION_TYPE (AVX512F_LEN, d) s2, int imm)
u3.a[i] = DEFAULT_VALUE;
}
- u1.x = INTRINSIC (_shuffle_pd) (s1.x, s2.x, 120);
- u2.x = INTRINSIC (_mask_shuffle_pd) (u2.x, mask, s1.x, s2.x, 120);
- u3.x = INTRINSIC (_maskz_shuffle_pd) (mask, s1.x, s2.x, 120);
- CALC (e, s1, s2, 120);
+ u1.x = INTRINSIC (_shuffle_pd) (s1.x, s2.x, IMM_VAR);
+ u2.x = INTRINSIC (_mask_shuffle_pd) (u2.x, mask, s1.x, s2.x, IMM_VAR);
+ u3.x = INTRINSIC (_maskz_shuffle_pd) (mask, s1.x, s2.x, IMM_VAR);
+ CALC (e, s1, s2, IMM_VAR);
if (UNION_CHECK (AVX512F_LEN, d) (u1, e))
abort ();
if (UNION_CHECK (AVX512F_LEN, d) (u3, e))
abort ();
}
+
+#undef IMM_VAR
--- /dev/null
+#include <stdlib.h>
+#include "cpuid.h"
+#include "m512-check.h"
+#include "avx512f-os-support.h"
+
+static void avx512vl_test (void);
+
+static void __attribute__ ((noinline)) do_test (void)
+{
+ avx512vl_test ();
+}
+
+int
+main ()
+{
+ unsigned int eax, ebx, ecx, edx;
+
+ if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx))
+ return 0;
+
+ /* Run avx512vl test only if host has avx512vl support. */
+ if ((ecx & bit_OSXSAVE) == (bit_OSXSAVE))
+ {
+ if (__get_cpuid_max (0, NULL) < 7)
+ return 0;
+
+ __cpuid_count (7, 0, eax, ebx, ecx, edx);
+
+ if ((avx512f_os_support ())
+ && ((ebx & bit_AVX512F) == bit_AVX512F)
+ && ((ebx & bit_AVX512VL) == bit_AVX512VL))
+ {
+ do_test ();
+#ifdef DEBUG
+ printf ("PASSED\n");
+#endif
+ return 0;
+ }
+#ifdef DEBUG
+ printf ("SKIPPED\n");
+#endif
+ }
+#ifdef DEBUG
+ else
+ printf ("SKIPPED\n");
+#endif
+
+ return 0;
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-require-effective-target avx512vl } */
+/* { dg-options "-O3 -mavx512vl" } */
+
+#include "avx512vl-check.h"
+
+#define N 12
+float vf1[N+4], vf2[N];
+double vd1[N+4], vd2[N];
+int vi1[N+4], vi2[N], k[N];
+long long vl1[N+4], vl2[N];
+long l[N];
+
+__attribute__((noinline, noclone)) void
+f1 (void)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vf2[i] = vf1[k[i]];
+}
+
+__attribute__((noinline, noclone)) void
+f2 (void)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vi2[i] = vi1[k[i]];
+}
+
+__attribute__((noinline, noclone)) void
+f3 (int x)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vf2[i] = vf1[k[i] + x];
+}
+
+__attribute__((noinline, noclone)) void
+f4 (int x)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vi2[i] = vi1[k[i] + x];
+}
+
+__attribute__((noinline, noclone)) void
+f5 (void)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vd2[i] = vd1[k[i]];
+}
+
+__attribute__((noinline, noclone)) void
+f6 (void)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vl2[i] = vl1[k[i]];
+}
+
+__attribute__((noinline, noclone)) void
+f7 (int x)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vd2[i] = vd1[k[i] + x];
+}
+
+__attribute__((noinline, noclone)) void
+f8 (int x)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vl2[i] = vl1[k[i] + x];
+}
+
+__attribute__((noinline, noclone)) void
+f9 (void)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vf2[i] = vf1[l[i]];
+}
+
+__attribute__((noinline, noclone)) void
+f10 (void)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vi2[i] = vi1[l[i]];
+}
+
+__attribute__((noinline, noclone)) void
+f11 (int x)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vf2[i] = vf1[l[i] + x];
+}
+
+__attribute__((noinline, noclone)) void
+f12 (int x)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vi2[i] = vi1[l[i] + x];
+}
+
+__attribute__((noinline, noclone)) void
+f13 (void)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vd2[i] = vd1[l[i]];
+}
+
+__attribute__((noinline, noclone)) void
+f14 (void)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vl2[i] = vl1[l[i]];
+}
+
+__attribute__((noinline, noclone)) void
+f15 (int x)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vd2[i] = vd1[l[i] + x];
+}
+
+__attribute__((noinline, noclone)) void
+f16 (int x)
+{
+ int i;
+ for (i = 0; i < N; i++)
+ vl2[i] = vl1[l[i] + x];
+}
+
+static void
+avx512vl_test (void)
+{
+ int i;
+
+ for (i = 0; i < N + 4; i++)
+ {
+ asm ("");
+ vf1[i] = 17.0f + i;
+ vd1[i] = 19.0 + i;
+ vi1[i] = 21 + i;
+ vl1[i] = 23L + i;
+ }
+ for (i = 0; i < N; i++)
+ {
+ asm ("");
+ k[i] = (i * 731) & (N - 1);
+ l[i] = (i * 657) & (N - 1);
+ }
+
+ f1 ();
+ f2 ();
+ for (i = 0; i < N; i++)
+ if (vf2[i] != ((i * 731) & (N - 1)) + 17
+ || vi2[i] != ((i * 731) & (N - 1)) + 21)
+ abort ();
+
+ f3 (1);
+ f4 (2);
+ for (i = 0; i < N; i++)
+ if (vf2[i] != ((i * 731) & (N - 1)) + 17 + 1
+ || vi2[i] != ((i * 731) & (N - 1)) + 21 + 2)
+ abort ();
+
+ f5 ();
+ f6 ();
+ for (i = 0; i < N; i++)
+ if (vd2[i] != ((i * 731) & (N - 1)) + 19
+ || vl2[i] != ((i * 731) & (N - 1)) + 23)
+ abort ();
+
+ f7 (3);
+ f8 (2);
+ for (i = 0; i < N; i++)
+ if (vd2[i] != ((i * 731) & (N - 1)) + 19 + 3
+ || vl2[i] != ((i * 731) & (N - 1)) + 23 + 2)
+ abort ();
+
+ f9 ();
+ f10 ();
+ for (i = 0; i < N; i++)
+ if (vf2[i] != ((i * 657) & (N - 1)) + 17
+ || vi2[i] != ((i * 657) & (N - 1)) + 21)
+ abort ();
+
+ f11 (4);
+ f12 (1);
+ for (i = 0; i < N; i++)
+ if (vf2[i] != ((i * 657) & (N - 1)) + 17 + 4
+ || vi2[i] != ((i * 657) & (N - 1)) + 21 + 1)
+ abort ();
+
+ f13 ();
+ f14 ();
+ for (i = 0; i < N; i++)
+ if (vd2[i] != ((i * 657) & (N - 1)) + 19
+ || vl2[i] != ((i * 657) & (N - 1)) + 23)
+ abort ();
+
+ f15 (2);
+ f16 (4);
+ for (i = 0; i < N; i++)
+ if (vd2[i] != ((i * 657) & (N - 1)) + 19 + 2
+ || vl2[i] != ((i * 657) & (N - 1)) + 23 + 4)
+ abort ();
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpgatherdd\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*ymm\[0-9\]{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vpgatherdd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x1, idx1;
+volatile __m128i x2, idx2;
+volatile __mmask8 m8;
+int *base;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mmask_i32gather_epi32 (x1, 0xFF, idx1, base, 8);
+ x1 = _mm256_mmask_i32gather_epi32 (x1, m8, idx1, base, 8);
+ x2 = _mm_mmask_i32gather_epi32 (x2, 0xFF, idx2, base, 8);
+ x2 = _mm_mmask_i32gather_epi32 (x2, m8, idx2, base, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vgatherdpd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*ymm\[0-9\]{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vgatherdpd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __m128i idx;
+volatile __mmask8 m8;
+double *base;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mmask_i32gather_pd (x1, 0xFF, idx, base, 8);
+ x1 = _mm256_mmask_i32gather_pd (x1, m8, idx, base, 8);
+ x2 = _mm_mmask_i32gather_pd (x2, 0xFF, idx, base, 8);
+ x2 = _mm_mmask_i32gather_pd (x2, m8, idx, base, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vgatherdps\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*ymm\[0-9\]{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vgatherdps\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __m256i idx1;
+volatile __m128i idx2;
+volatile __mmask8 m8;
+float *base;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mmask_i32gather_ps (x1, 0xFF, idx1, base, 8);
+ x1 = _mm256_mmask_i32gather_ps (x1, m8, idx1, base, 8);
+ x2 = _mm_mmask_i32gather_ps (x2, 0xFF, idx2, base, 8);
+ x2 = _mm_mmask_i32gather_ps (x2, m8, idx2, base, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpgatherdq\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*ymm\[0-9\]{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vpgatherdq\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x1;
+volatile __m128i x2, idx;
+volatile __mmask8 m8;
+long long *base;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mmask_i32gather_epi64 (x1, 0xFF, idx, base, 8);
+ x1 = _mm256_mmask_i32gather_epi64 (x1, m8, idx, base, 8);
+ x2 = _mm_mmask_i32gather_epi64 (x2, 0xFF, idx, base, 8);
+ x2 = _mm_mmask_i32gather_epi64 (x2, m8, idx, base, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpscatterdd\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*ymm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vpscatterdd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256i src1, idx1;
+volatile __m128i src2, idx2;
+volatile __mmask8 m8;
+int *addr;
+
+void extern
+avx512vl_test (void)
+{
+ _mm256_i32scatter_epi32 (addr, idx1, src1, 8);
+ _mm256_mask_i32scatter_epi32 (addr, m8, idx1, src1, 8);
+
+ _mm_i32scatter_epi32 (addr, idx2, src2, 8);
+ _mm_mask_i32scatter_epi32 (addr, m8, idx2, src2, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vscatterdpd\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vscatterdpd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256d src1;
+volatile __m128d src2;
+volatile __m128i idx;
+volatile __mmask8 m8;
+double *addr;
+
+void extern
+avx512vl_test (void)
+{
+ _mm256_i32scatter_pd (addr, idx, src1, 8);
+ _mm256_mask_i32scatter_pd (addr, m8, idx, src1, 8);
+
+ _mm_i32scatter_pd (addr, idx, src2, 8);
+ _mm_mask_i32scatter_pd (addr, m8, idx, src2, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vscatterdps\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*ymm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vscatterdps\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256 src1;
+volatile __m128 src2;
+volatile __m256i idx1;
+volatile __m128i idx2;
+volatile __mmask8 m8;
+float *addr;
+
+void extern
+avx512vl_test (void)
+{
+ _mm256_i32scatter_ps (addr, idx1, src1, 8);
+ _mm256_mask_i32scatter_ps (addr, m8, idx1, src1, 8);
+
+ _mm_i32scatter_ps (addr, idx2, src2, 8);
+ _mm_mask_i32scatter_ps (addr, m8, idx2, src2, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpscatterdq\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vpscatterdq\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256i src1;
+volatile __m128i src2, idx;
+volatile __mmask8 m8;
+long long *addr;
+
+void extern
+avx512vl_test (void)
+{
+ _mm256_i32scatter_epi64 (addr, idx, src1, 8);
+ _mm256_mask_i32scatter_epi64 (addr, m8, idx, src1, 8);
+
+ _mm_i32scatter_epi64 (addr, idx, src2, 8);
+ _mm_mask_i32scatter_epi64 (addr, m8, idx, src2, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpgatherqd\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vpgatherqd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256i idx1;
+volatile __m128i idx2, x;
+volatile __mmask8 m8;
+int *base;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mmask_i64gather_epi32 (x, 0xFF, idx1, base, 8);
+ x = _mm256_mmask_i64gather_epi32 (x, m8, idx1, base, 8);
+ x = _mm_mmask_i64gather_epi32 (x, 0xFF, idx2, base, 8);
+ x = _mm_mmask_i64gather_epi32 (x, m8, idx2, base, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vgatherqpd\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*ymm\[0-9\]{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vgatherqpd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __m256i idx1;
+volatile __m128i idx2;
+volatile __mmask8 m8;
+double *base;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mmask_i64gather_pd (x1, 0xFF, idx1, base, 8);
+ x1 = _mm256_mmask_i64gather_pd (x1, m8, idx1, base, 8);
+ x2 = _mm_mmask_i64gather_pd (x2, 0xFF, idx2, base, 8);
+ x2 = _mm_mmask_i64gather_pd (x2, m8, idx2, base, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vgatherqps\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vgatherqps\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m128 x;
+volatile __m256i idx1;
+volatile __m128i idx2;
+volatile __mmask8 m8;
+float *base;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mmask_i64gather_ps (x, 0xFF, idx1, base, 8);
+ x = _mm256_mmask_i64gather_ps (x, m8, idx1, base, 8);
+ x = _mm_mmask_i64gather_ps (x, 0xFF, idx2, base, 8);
+ x = _mm_mmask_i64gather_ps (x, m8, idx2, base, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpgatherqq\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*ymm\[0-9\]{%k\[1-7\]}" 2} } */
+/* { dg-final { scan-assembler-times "vpgatherqq\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x1, idx1;
+volatile __m128i x2, idx2;
+volatile __mmask8 m8;
+long long *base;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mmask_i64gather_epi64 (x1, 0xFF, idx1, base, 8);
+ x1 = _mm256_mmask_i64gather_epi64 (x1, m8, idx1, base, 8);
+ x2 = _mm_mmask_i64gather_epi64 (x2, 0xFF, idx2, base, 8);
+ x2 = _mm_mmask_i64gather_epi64 (x2, m8, idx2, base, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpscatterqd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*ymm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vpscatterqd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256i idx1;
+volatile __m128i idx2, src;
+volatile __mmask8 m8;
+int *addr;
+
+void extern
+avx512vl_test (void)
+{
+ _mm256_i64scatter_epi32 (addr, idx1, src, 8);
+ _mm256_mask_i64scatter_epi32 (addr, m8, idx1, src, 8);
+
+ _mm_i64scatter_epi32 (addr, idx2, src, 8);
+ _mm_mask_i64scatter_epi32 (addr, m8, idx2, src, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vscatterqpd\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*ymm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vscatterqpd\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256d src1;
+volatile __m128d src2;
+volatile __m256i idx1;
+volatile __m128i idx2;
+volatile __mmask8 m8;
+double *addr;
+
+void extern
+avx512vl_test (void)
+{
+ _mm256_i64scatter_pd (addr, idx1, src1, 8);
+ _mm256_mask_i64scatter_pd (addr, m8, idx1, src1, 8);
+
+ _mm_i64scatter_pd (addr, idx2, src2, 8);
+ _mm_mask_i64scatter_pd (addr, m8, idx2, src2, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vscatterqps\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*ymm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vscatterqps\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m128 src;
+volatile __m256i idx1;
+volatile __m128i idx2;
+volatile __mmask8 m8;
+float *addr;
+
+void extern
+avx512vl_test (void)
+{
+ _mm256_i64scatter_ps (addr, idx1, src, 8);
+ _mm256_mask_i64scatter_ps (addr, m8, idx1, src, 8);
+
+ _mm_i64scatter_ps (addr, idx2, src, 8);
+ _mm_mask_i64scatter_ps (addr, m8, idx2, src, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpscatterqq\[ \\t\]+\[^\n\]*ymm\[0-9\]\[^\n\]*ymm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+/* { dg-final { scan-assembler-times "vpscatterqq\[ \\t\]+\[^\n\]*xmm\[0-9\]\[^\n\]*xmm\[0-9\]\[^\n\]*{%k\[1-7\]}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256i src1, idx1;
+volatile __m128i src2, idx2;
+volatile __mmask8 m8;
+long long *addr;
+
+void extern
+avx512vl_test (void)
+{
+ _mm256_i64scatter_epi64 (addr, idx1, src1, 8);
+ _mm256_mask_i64scatter_epi64 (addr, m8, idx1, src1, 8);
+
+ _mm_i64scatter_epi64 (addr, idx2, src2, 8);
+ _mm_mask_i64scatter_epi64 (addr, m8, idx2, src2, 8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vaddpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vaddpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vaddpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vaddpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128d x128;
+volatile __m256d x256;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x128 = _mm_mask_add_pd (x128, m, x128, x128);
+ x128 = _mm_maskz_add_pd (m, x128, x128);
+
+ x256 = _mm256_mask_add_pd (x256, m, x256, x256);
+ x256 = _mm256_maskz_add_pd (m, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vaddpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vaddpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vaddps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vaddps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vaddps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vaddps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128 x128;
+volatile __m256 x256;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x128 = _mm_mask_add_ps (x128, m, x128, x128);
+ x128 = _mm_maskz_add_ps (m, x128, x128);
+
+ x256 = _mm256_mask_add_ps (x256, m, x256, x256);
+ x256 = _mm256_maskz_add_ps (m, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vaddps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vaddps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "valignd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "valignd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "valignd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "valignd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "valignd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "valignd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_alignr_epi32 (y, y, 3);
+ y = _mm256_mask_alignr_epi32 (y, m, y, y, 3);
+ y = _mm256_maskz_alignr_epi32 (m, y, y, 3);
+
+ x = _mm_alignr_epi32 (x, x, 3);
+ x = _mm_mask_alignr_epi32 (x, m, x, x, 3);
+ x = _mm_maskz_alignr_epi32 (m, x, x, 3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-valignd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-valignd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "valignq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "valignq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "valignq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "valignq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "valignq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "valignq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_alignr_epi64 (y, y, 3);
+ y = _mm256_mask_alignr_epi64 (y, m, y, y, 3);
+ y = _mm256_maskz_alignr_epi64 (m, y, y, 3);
+
+ x = _mm_alignr_epi64 (x, x, 1);
+ x = _mm_mask_alignr_epi64 (x, m, x, x, 1);
+ x = _mm_maskz_alignr_epi64 (m, x, x, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-valignq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-valignq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vandnpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vandnpd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vandnps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vandnps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vandpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vandpd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vandps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vandps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "(vblendmpd|vmovapd)\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "(vblendmpd|vmovapd)\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256d x;
+volatile __m128d xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_blend_pd (m, x, x);
+ xx = _mm_mask_blend_pd (m, xx, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vblendmpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vblendmpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "(vblendmps|vmovaps)\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "(vblendmps|vmovaps)\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __m128 xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_blend_ps (m, x, x);
+ xx = _mm_mask_blend_ps (m, xx, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vblendmps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vblendmps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vbroadcastf32x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]|vshuff32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vshuff32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastf32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}|vshuff32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __m128 y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_broadcast_f32x4 (y);
+ x = _mm256_mask_broadcast_f32x4 (x, m, y);
+ x = _mm256_maskz_broadcast_f32x4 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -mavx512dq -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vbroadcastf32x4-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vbroadcastf64x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vbroadcasti32x2-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vbroadcasti32x2-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512dq -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]|vshufi32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vshufi32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcasti32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}|vshufi32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_broadcast_i32x4 (y);
+ x = _mm256_mask_broadcast_i32x4 (x, m, y);
+ x = _mm256_maskz_broadcast_i32x4 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -mavx512dq -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vbroadcasti32x4-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vbroadcasti64x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcastsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x;
+volatile __m128d y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_broadcastsd_pd (y);
+ x = _mm256_mask_broadcastsd_pd (x, m, y);
+ x = _mm256_maskz_broadcastsd_pd (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vbroadcastsd-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vbroadcastss\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastss\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastss\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastss\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastss\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vbroadcastss\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __m128 y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_broadcastss_ps (y);
+ x = _mm256_mask_broadcastss_ps (x, m, y);
+ x = _mm256_maskz_broadcastss_ps (m, y);
+ y = _mm_broadcastss_ps (y);
+ y = _mm_mask_broadcastss_ps (y, m, y);
+ y = _mm_maskz_broadcastss_ps (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vbroadcastss-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vbroadcastss-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler "vcmppd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vcmppd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vcmppd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vcmppd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m256d x;
+volatile __m128d xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm256_cmp_pd_mask (x, x, _CMP_FALSE_OQ);
+ m = _mm256_mask_cmp_pd_mask (m, x, x, _CMP_FALSE_OQ);
+ m = _mm_cmp_pd_mask (xx, xx, _CMP_FALSE_OQ);
+ m = _mm_mask_cmp_pd_mask (m, xx, xx, _CMP_FALSE_OQ);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcmppd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcmppd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler "vcmpps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vcmpps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vcmpps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vcmpps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __m128 xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm256_cmp_ps_mask (x, x, _CMP_FALSE_OQ);
+ m = _mm256_mask_cmp_ps_mask (m, x, x, _CMP_FALSE_OQ);
+ m = _mm_cmp_ps_mask (xx, xx, _CMP_FALSE_OQ);
+ m = _mm_mask_cmp_ps_mask (m, xx, xx, _CMP_FALSE_OQ);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcmpps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcmpps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcompresspd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcompresspd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcompresspd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcompresspd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcompresspd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcompresspd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+double *p;
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_compress_pd (x1, m, x1);
+ x2 = _mm_mask_compress_pd (x2, m, x2);
+
+ x1 = _mm256_maskz_compress_pd (m, x1);
+ x2 = _mm_maskz_compress_pd (m, x2);
+
+ _mm256_mask_compressstoreu_pd (p, m, x1);
+ _mm_mask_compressstoreu_pd (p, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcompresspd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcompresspd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcompressps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcompressps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcompressps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcompressps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcompressps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcompressps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+float *p;
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_compress_ps (x1, m, x1);
+ x2 = _mm_mask_compress_ps (x2, m, x2);
+
+ x1 = _mm256_maskz_compress_ps (m, x1);
+ x2 = _mm_maskz_compress_ps (m, x2);
+
+ _mm256_mask_compressstoreu_ps (p, m, x1);
+ _mm_mask_compressstoreu_ps (p, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcompressps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcompressps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtdq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtdq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtdq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtdq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s;
+volatile __m256d res1;
+volatile __m128d res2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepi32_pd (res1, m, s);
+ res2 = _mm_mask_cvtepi32_pd (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepi32_pd (m, s);
+ res2 = _mm_maskz_cvtepi32_pd (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtdq2pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtdq2pd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtdq2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtdq2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtdq2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtdq2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i s1;
+volatile __m128i s2;
+volatile __m256 res1;
+volatile __m128 res2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepi32_ps (res1, m, s1);
+ res2 = _mm_mask_cvtepi32_ps (res2, m, s2);
+
+ res1 = _mm256_maskz_cvtepi32_ps (m, s1);
+ res2 = _mm_maskz_cvtepi32_ps (m, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtdq2ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtdq2ps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtpd2dqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2dqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2dqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2dqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d s1;
+volatile __m128d s2;
+volatile __m128i res;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res = _mm256_mask_cvtpd_epi32 (res, m, s1);
+ res = _mm_mask_cvtpd_epi32 (res, m, s2);
+
+ res = _mm256_maskz_cvtpd_epi32 (m, s1);
+ res = _mm_maskz_cvtpd_epi32 (m, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtpd2dq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtpd2dq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vcvtpd2psy\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2psy\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2psx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2psx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x;
+volatile __m128d xx;
+volatile __m128 y;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_cvtpd_ps (y, 4, x);
+ y = _mm256_maskz_cvtpd_ps (6, x);
+ y = _mm_mask_cvtpd_ps (y, 4, xx);
+ y = _mm_maskz_cvtpd_ps (6, xx);
+}
+
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtpd2ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtpd2ps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtpd2qq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtpd2qq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtpd2udqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2udqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2udqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2udqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2udqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtpd2udqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d s1;
+volatile __m128d s2;
+volatile __m128i res;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res = _mm256_cvtpd_epu32 (s1);
+ res = _mm_cvtpd_epu32 (s2);
+
+ res = _mm256_mask_cvtpd_epu32 (res, m, s1);
+ res = _mm_mask_cvtpd_epu32 (res, m, s2);
+
+ res = _mm256_maskz_cvtpd_epu32 (m, s1);
+ res = _mm_maskz_cvtpd_epu32 (m, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtpd2udq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtpd2udq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtpd2uqq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtpd2uqq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vcvtph2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtph2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x;
+volatile __m256 y;
+volatile __m128i xx;
+volatile __m128 yy;
+
+void extern
+avx512bw_test (void)
+{
+ y = _mm256_mask_cvtph_ps (y, 4, x);
+ y = _mm256_maskz_cvtph_ps (6, x);
+ yy = _mm_mask_cvtph_ps (yy, 4, xx);
+ yy = _mm_maskz_cvtph_ps (6, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mf16c -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtph2ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtph2ps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtps2dq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2dq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2dq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2dq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __m256i z1;
+volatile __m128i z2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z1 = _mm256_mask_cvtps_epi32 (z1, m, x1);
+ z1 = _mm256_maskz_cvtps_epi32 (m, x1);
+ z2 = _mm_mask_cvtps_epi32 (z2, m, x2);
+ z2 = _mm_maskz_cvtps_epi32 (m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtps2dq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtps2dq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtps2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128 s;
+volatile __m256d res1;
+volatile __m128d res2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtps_pd (res1, m, s);
+ res2 = _mm_mask_cvtps_pd (res2, m, s);
+
+ res1 = _mm256_maskz_cvtps_pd (m, s);
+ res2 = _mm_maskz_cvtps_pd (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtps2pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtps2pd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vcvtps2ph\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vcvtps2ph\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __m128i y;
+volatile __m128 xx;
+volatile __m128i yy;
+
+void extern
+avx512bw_test (void)
+{
+ y = _mm256_maskz_cvtps_ph (4, x, 0);
+ y = _mm256_mask_cvtps_ph (y, 2, x, 0);
+ yy = _mm_maskz_cvtps_ph (4, xx, 0);
+ yy = _mm_mask_cvtps_ph (yy, 2, xx, 0);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mf16c -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtps2ph-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtps2ph-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtps2qq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtps2qq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtps2udq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2udq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2udq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2udq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2udq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtps2udq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __m256i z1;
+volatile __m128i z2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z1 = _mm256_cvtps_epu32 (x1);
+ z1 = _mm256_mask_cvtps_epu32 (z1, m, x1);
+ z1 = _mm256_maskz_cvtps_epu32 (m, x1);
+ z2 = _mm_cvtps_epu32 (x2);
+ z2 = _mm_mask_cvtps_epu32 (z2, m, x2);
+ z2 = _mm_maskz_cvtps_epu32 (m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtps2udq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtps2udq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtps2uqq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtps2uqq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtqq2pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtqq2pd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtqq2ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtqq2ps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvttpd2dqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2dqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2dqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2dqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d s1;
+volatile __m128d s2;
+volatile __m128i res;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res = _mm256_mask_cvttpd_epi32 (res, m, s1);
+ res = _mm_mask_cvttpd_epi32 (res, m, s2);
+
+ res = _mm256_maskz_cvttpd_epi32 (m, s1);
+ res = _mm_maskz_cvttpd_epi32 (m, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvttpd2dq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvttpd2dq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvttpd2qq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvttpd2qq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvttpd2udqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2udqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2udqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2udqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2udqy\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttpd2udqx\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d s1;
+volatile __m128d s2;
+volatile __m128i res;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res = _mm256_cvttpd_epu32 (s1);
+ res = _mm_cvttpd_epu32 (s2);
+
+ res = _mm256_mask_cvttpd_epu32 (res, m, s1);
+ res = _mm_mask_cvttpd_epu32 (res, m, s2);
+
+ res = _mm256_maskz_cvttpd_epu32 (m, s1);
+ res = _mm_maskz_cvttpd_epu32 (m, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvttpd2udq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvttpd2udq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvttpd2uqq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvttpd2uqq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvttps2dq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2dq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2dq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2dq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 s1;
+volatile __m128 s2;
+volatile __m256i res1;
+volatile __m128i res2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvttps_epi32 (res1, m, s1);
+ res1 = _mm256_maskz_cvttps_epi32 (m, s1);
+ res2 = _mm_mask_cvttps_epi32 (res2, m, s2);
+ res2 = _mm_maskz_cvttps_epi32 (m, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvttps2dq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvttps2dq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvttps2qq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvttps2qq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvttps2udq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2udq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2udq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2udq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2udq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvttps2udq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 s1;
+volatile __m128 s2;
+volatile __m256i res1;
+volatile __m128i res2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_cvttps_epu32 (s1);
+ res1 = _mm256_mask_cvttps_epu32 (res1, m, s1);
+ res1 = _mm256_maskz_cvttps_epu32 (m, s1);
+ res2 = _mm_cvttps_epu32 (s2);
+ res2 = _mm_mask_cvttps_epu32 (res2, m, s2);
+ res2 = _mm_maskz_cvttps_epu32 (m, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvttps2udq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvttps2udq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvttps2uqq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvttps2uqq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtudq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s;
+volatile __m256d res1;
+volatile __m128d res2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_cvtepu32_pd (s);
+ res2 = _mm_cvtepu32_pd (s);
+
+ res1 = _mm256_mask_cvtepu32_pd (res1, m, s);
+ res2 = _mm_mask_cvtepu32_pd (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepu32_pd (m, s);
+ res2 = _mm_maskz_cvtepu32_pd (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtudq2pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtudq2pd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vcvtudq2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vcvtudq2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i s1;
+volatile __m128i s2;
+volatile __m256 res1;
+volatile __m128 res2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_cvtepu32_ps (s1);
+ res2 = _mm_cvtepu32_ps (s2);
+
+ res1 = _mm256_mask_cvtepu32_ps (res1, m, s1);
+ res2 = _mm_mask_cvtepu32_ps (res2, m, s2);
+
+ res1 = _mm256_maskz_cvtepu32_ps (m, s1);
+ res2 = _mm_maskz_cvtepu32_ps (m, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtudq2ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vcvtudq2ps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtuqq2pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtuqq2pd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtuqq2ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vcvtuqq2ps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vdbpsadbw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vdbpsadbw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vdivpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vdivpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdivpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vdivpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vdivpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdivpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x256;
+volatile __m128d x128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_div_pd (x256, m, x256, x256);
+ x256 = _mm256_maskz_div_pd (m, x256, x256);
+ x128 = _mm_mask_div_pd (x128, m, x128, x128);
+ x128 = _mm_maskz_div_pd (m, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vdivpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vdivpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vdivps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vdivps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdivps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vdivps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vdivps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vdivps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x256;
+volatile __m128 x128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_div_ps (x256, m, x256, x256);
+ x256 = _mm256_maskz_div_ps (m, x256, x256);
+ x128 = _mm_mask_div_ps (x128, m, x128, x128);
+ x128 = _mm_maskz_div_ps (m, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vdivps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vdivps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vexpandpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vexpandpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vexpandpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vexpandpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+
+#include <immintrin.h>
+
+double *p;
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_expand_pd (x1, m, x1);
+ x2 = _mm_mask_expand_pd (x2, m, x2);
+
+ x1 = _mm256_maskz_expand_pd (m, x1);
+ x2 = _mm_maskz_expand_pd (m, x2);
+
+ x1 = _mm256_mask_expandloadu_pd (x1, m, p);
+ x2 = _mm_mask_expandloadu_pd (x2, m, p);
+
+ x1 = _mm256_maskz_expandloadu_pd (m, p);
+ x2 = _mm_maskz_expandloadu_pd (m, p);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vexpandpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vexpandpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vexpandps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vexpandps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vexpandps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vexpandps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+
+#include <immintrin.h>
+
+float *p;
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_expand_ps (x1, m, x1);
+ x2 = _mm_mask_expand_ps (x2, m, x2);
+
+ x1 = _mm256_maskz_expand_ps (m, x1);
+ x2 = _mm_maskz_expand_ps (m, x2);
+
+ x1 = _mm256_mask_expandloadu_ps (x1, m, p);
+ x2 = _mm_mask_expandloadu_ps (x2, m, p);
+
+ x1 = _mm256_maskz_expandloadu_ps (m, p);
+ x2 = _mm_maskz_expandloadu_ps (m, p);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vexpandps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vexpandps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vextractf32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vextractf32x4\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vextractf32x4\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __m128 y;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_extractf32x4_ps (x, 1);
+ y = _mm256_mask_extractf32x4_ps (y, 2, x, 1);
+ y = _mm256_maskz_extractf32x4_ps (2, x, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vextractf32x4-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vextractf64x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vextracti32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vextracti32x4\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vextracti32x4\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_extracti32x4_epi32 (x, 1);
+ y = _mm256_mask_extracti32x4_epi32 (y, 2, x, 1);
+ y = _mm256_maskz_extracti32x4_epi32 (2, x, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vextracti32x4-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vextracti64x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler "vfixupimmpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" } } */
+/* { dg-final { scan-assembler "vfixupimmpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" } } */
+/* { dg-final { scan-assembler "vfixupimmpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" } } */
+/* { dg-final { scan-assembler "vfixupimmpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256d xx;
+volatile __m256i yy;
+volatile __m128d x2;
+volatile __m128i y2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ xx = _mm256_fixupimm_pd (xx, xx, yy, 3);
+ xx = _mm256_mask_fixupimm_pd (xx, m, xx, yy, 3);
+ xx = _mm256_maskz_fixupimm_pd (m, xx, xx, yy, 3);
+ x2 = _mm_fixupimm_pd (x2, x2, y2, 3);
+ x2 = _mm_mask_fixupimm_pd (x2, m, x2, y2, 3);
+ x2 = _mm_maskz_fixupimm_pd (m, x2, x2, y2, 3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfixupimmpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfixupimmpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler "vfixupimmps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" } } */
+/* { dg-final { scan-assembler "vfixupimmps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" } } */
+/* { dg-final { scan-assembler "vfixupimmps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" } } */
+/* { dg-final { scan-assembler "vfixupimmps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256 xx;
+volatile __m256i yy;
+volatile __m128 x2;
+volatile __m128i y2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ xx = _mm256_fixupimm_ps (xx, xx, yy, 3);
+ xx = _mm256_mask_fixupimm_ps (xx, m, xx, yy, 3);
+ xx = _mm256_maskz_fixupimm_ps (m, xx, xx, yy, 3);
+ x2 = _mm_fixupimm_ps (x2, x2, y2, 3);
+ x2 = _mm_mask_fixupimm_ps (x2, m, x2, y2, 3);
+ x2 = _mm_maskz_fixupimm_ps (m, x2, x2, y2, 3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfixupimmps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfixupimmps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmadd...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmadd...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmadd231pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmadd231pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmadd...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfmadd...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2, y3;
+volatile __m128d xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fmadd_pd (yy, m, y2, y3);
+ xx = _mm_mask_fmadd_pd (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fmadd_pd (yy, y2, y3, m);
+ x3 = _mm_mask3_fmadd_pd (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fmadd_pd (m, yy, y2, y3);
+ xx = _mm_maskz_fmadd_pd (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmaddXXXpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmaddXXXpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmadd...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmadd...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmadd231ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmadd231ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmadd...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfmadd...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 yy, y2, y3;
+volatile __m128 xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fmadd_ps (yy, m, y2, y3);
+ xx = _mm_mask_fmadd_ps (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fmadd_ps (yy, y2, y3, m);
+ x3 = _mm_mask3_fmadd_ps (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fmadd_ps (m, yy, y2, y3);
+ xx = _mm_maskz_fmadd_ps (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmaddXXXps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmaddXXXps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmaddsub...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub231pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub231pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2, y3;
+volatile __m128d xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fmaddsub_pd (yy, m, y2, y3);
+ xx = _mm_mask_fmaddsub_pd (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fmaddsub_pd (yy, y2, y3, m);
+ x3 = _mm_mask3_fmaddsub_pd (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fmaddsub_pd (m, yy, y2, y3);
+ xx = _mm_maskz_fmaddsub_pd (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmaddsubXXXpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmaddsubXXXpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmaddsub...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub231ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub231ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 yy, y2, y3;
+volatile __m128 xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fmaddsub_ps (yy, m, y2, y3);
+ xx = _mm_mask_fmaddsub_ps (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fmaddsub_ps (yy, y2, y3, m);
+ x3 = _mm_mask3_fmaddsub_ps (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fmaddsub_ps (m, yy, y2, y3);
+ xx = _mm_maskz_fmaddsub_ps (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmaddsubXXXps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmaddsubXXXps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmsub...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsub...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsub231pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsub231pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsub...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsub...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2, y3;
+volatile __m128d xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fmsub_pd (yy, m, y2, y3);
+ xx = _mm_mask_fmsub_pd (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fmsub_pd (yy, y2, y3, m);
+ x3 = _mm_mask3_fmsub_pd (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fmsub_pd (m, yy, y2, y3);
+ xx = _mm_maskz_fmsub_pd (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmsubXXXpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmsubXXXpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmsub...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsub...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsub231ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsub231ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsub...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsub...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 yy, y2, y3;
+volatile __m128 xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fmsub_ps (yy, m, y2, y3);
+ xx = _mm_mask_fmsub_ps (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fmsub_ps (yy, y2, y3, m);
+ x3 = _mm_mask3_fmsub_ps (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fmsub_ps (m, yy, y2, y3);
+ xx = _mm_maskz_fmsub_ps (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmsubXXXps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmsubXXXps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmsubadd...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd231pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd231pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2, y3;
+volatile __m128d xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fmsubadd_pd (yy, m, y2, y3);
+ xx = _mm_mask_fmsubadd_pd (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fmsubadd_pd (yy, y2, y3, m);
+ x3 = _mm_mask3_fmsubadd_pd (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fmsubadd_pd (m, yy, y2, y3);
+ xx = _mm_maskz_fmsubadd_pd (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmsubaddXXXpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmsubaddXXXpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfmsubadd...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd231ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd231ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 yy, y2, y3;
+volatile __m128 xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fmsubadd_ps (yy, m, y2, y3);
+ xx = _mm_mask_fmsubadd_ps (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fmsubadd_ps (yy, y2, y3, m);
+ x3 = _mm_mask3_fmsubadd_ps (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fmsubadd_ps (m, yy, y2, y3);
+ xx = _mm_maskz_fmsubadd_ps (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmsubaddXXXps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfmsubaddXXXps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfnmadd...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmadd231pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmadd231pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2, y3;
+volatile __m128d xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fnmadd_pd (yy, m, y2, y3);
+ xx = _mm_mask_fnmadd_pd (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fnmadd_pd (yy, y2, y3, m);
+ x3 = _mm_mask3_fnmadd_pd (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fnmadd_pd (m, yy, y2, y3);
+ xx = _mm_maskz_fnmadd_pd (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfnmaddXXXpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfnmaddXXXpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfnmadd...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmadd231ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmadd231ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 yy, y2, y3;
+volatile __m128 xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fnmadd_ps (yy, m, y2, y3);
+ xx = _mm_mask_fnmadd_ps (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fnmadd_ps (yy, y2, y3, m);
+ x3 = _mm_mask3_fnmadd_ps (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fnmadd_ps (m, yy, y2, y3);
+ xx = _mm_maskz_fnmadd_ps (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfnmaddXXXps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfnmaddXXXps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfnmsub...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmsub231pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmsub231pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2, y3;
+volatile __m128d xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fnmsub_pd (yy, m, y2, y3);
+ xx = _mm_mask_fnmsub_pd (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fnmsub_pd (yy, y2, y3, m);
+ x3 = _mm_mask3_fnmsub_pd (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fnmsub_pd (m, yy, y2, y3);
+ xx = _mm_maskz_fnmsub_pd (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfnmsubXXXpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfnmsubXXXpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vfnmsub...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmsub231ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmsub231ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 yy, y2, y3;
+volatile __m128 xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_fnmsub_ps (yy, m, y2, y3);
+ xx = _mm_mask_fnmsub_ps (xx, m, x2, x3);
+
+ y3 = _mm256_mask3_fnmsub_ps (yy, y2, y3, m);
+ x3 = _mm_mask3_fnmsub_ps (xx, x2, x3, m);
+
+ yy = _mm256_maskz_fnmsub_ps (m, yy, y2, y3);
+ xx = _mm_maskz_fnmsub_ps (m, xx, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfnmsubXXXps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vfnmsubXXXps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vfpclasspd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vfpclasspd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vfpclassps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vfpclassps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vgetexppd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3} } */
+/* { dg-final { scan-assembler-times "vgetexppd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1} } */
+/* { dg-final { scan-assembler-times "vgetexppd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1} } */
+/* { dg-final { scan-assembler-times "vgetexppd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3} } */
+/* { dg-final { scan-assembler-times "vgetexppd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1} } */
+/* { dg-final { scan-assembler-times "vgetexppd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1} } */
+
+#include <immintrin.h>
+
+volatile __m256d xx;
+volatile __m128d x2;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ xx = _mm256_getexp_pd (xx);
+ xx = _mm256_mask_getexp_pd (xx, m8, xx);
+ xx = _mm256_maskz_getexp_pd (m8, xx);
+ x2 = _mm_getexp_pd (x2);
+ x2 = _mm_mask_getexp_pd (x2, m8, x2);
+ x2 = _mm_maskz_getexp_pd (m8, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vgetexppd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vgetexppd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vgetexpps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3} } */
+/* { dg-final { scan-assembler-times "vgetexpps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1} } */
+/* { dg-final { scan-assembler-times "vgetexpps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1} } */
+/* { dg-final { scan-assembler-times "vgetexpps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3} } */
+/* { dg-final { scan-assembler-times "vgetexpps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1} } */
+/* { dg-final { scan-assembler-times "vgetexpps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1} } */
+
+#include <immintrin.h>
+
+volatile __m256 xx;
+volatile __m128 x2;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ xx = _mm256_getexp_ps (xx);
+ xx = _mm256_mask_getexp_ps (xx, m8, xx);
+ xx = _mm256_maskz_getexp_ps (m8, xx);
+ x2 = _mm_getexp_ps (x2);
+ x2 = _mm_mask_getexp_ps (x2, m8, x2);
+ x2 = _mm_maskz_getexp_ps (m8, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vgetexpps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vgetexpps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vgetmantpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x, y;
+volatile __m128d a, b;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_getmant_pd (y, _MM_MANT_NORM_p75_1p5, _MM_MANT_SIGN_src);
+ x = _mm256_mask_getmant_pd (x, m, y, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ x = _mm256_maskz_getmant_pd (m, y, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ a = _mm_getmant_pd (b, _MM_MANT_NORM_p75_1p5, _MM_MANT_SIGN_src);
+ a = _mm_mask_getmant_pd (a, m, b, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ a = _mm_maskz_getmant_pd (m, b, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vgetmantpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vgetmantpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vgetmantps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vgetmantps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x, y;
+volatile __m128 a, b;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_getmant_ps (y, _MM_MANT_NORM_p75_1p5, _MM_MANT_SIGN_src);
+ x = _mm256_mask_getmant_ps (x, m, y, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ x = _mm256_maskz_getmant_ps (m, y, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ a = _mm_getmant_ps (b, _MM_MANT_NORM_p75_1p5, _MM_MANT_SIGN_src);
+ a = _mm_mask_getmant_ps (a, m, b, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+ a = _mm_maskz_getmant_ps (m, b, _MM_MANT_NORM_p75_1p5,
+ _MM_MANT_SIGN_src);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vgetmantps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vgetmantps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vinsertf32x4\[^\n\]*ymm" 3 } } */
+/* { dg-final { scan-assembler-times "vinsertf32x4\[^\n\]*\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vinsertf32x4\[^\n\]*\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __m128 y;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_insertf32x4 (x, y, 1);
+ x = _mm256_mask_insertf32x4 (x, 2, x, y, 1);
+ x = _mm256_maskz_insertf32x4 (2, x, y, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vinsertf32x4-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vinsertf64x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vinserti32x4\[^\n\]*ymm" 3 } } */
+/* { dg-final { scan-assembler-times "vinserti32x4\[^\n\]*\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vinserti32x4\[^\n\]*\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_inserti32x4 (x, y, 1);
+ x = _mm256_mask_inserti32x4 (x, 2, x, y, 1);
+ x = _mm256_maskz_inserti32x4 (2, x, y, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vinserti32x4-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vinserti64x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vmaxpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vmaxpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmaxpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmaxpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vmaxpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmaxpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x256;
+volatile __m128d x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_max_pd (x256, m8, x256, x256);
+ x256 = _mm256_maskz_max_pd (m8, x256, x256);
+ x128 = _mm_mask_max_pd (x128, m8, x128, x128);
+ x128 = _mm_maskz_max_pd (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmaxpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmaxpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vmaxps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vmaxps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmaxps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmaxps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vmaxps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmaxps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x256;
+volatile __m128 x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_max_ps (x256, m8, x256, x256);
+ x256 = _mm256_maskz_max_ps (m8, x256, x256);
+ x128 = _mm_mask_max_ps (x128, m8, x128, x128);
+ x128 = _mm_maskz_max_ps (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmaxps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmaxps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vminpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vminpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vminpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vminpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vminpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vminpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x256;
+volatile __m128d x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_min_pd (x256, m8, x256, x256);
+ x256 = _mm256_maskz_min_pd (m8, x256, x256);
+ x128 = _mm_mask_min_pd (x128, m8, x128, x128);
+ x128 = _mm_maskz_min_pd (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vminpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vminpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vminps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vminps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vminps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vminps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vminps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vminps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x256;
+volatile __m128 x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_min_ps (x256, m8, x256, x256);
+ x256 = _mm256_maskz_min_ps (m8, x256, x256);
+ x128 = _mm_mask_min_ps (x128, m8, x128, x128);
+ x128 = _mm_maskz_min_ps (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vminps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vminps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovapd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+double *p;
+volatile __m256d yy, y2;
+volatile __m128d xx, x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_mov_pd (yy, m, y2);
+ xx = _mm_mask_mov_pd (xx, m, x2);
+
+ yy = _mm256_maskz_mov_pd (m, y2);
+ xx = _mm_maskz_mov_pd (m, x2);
+
+ yy = _mm256_mask_load_pd (yy, m, p);
+ xx = _mm_mask_load_pd (xx, m, p);
+
+ yy = _mm256_maskz_load_pd (m, p);
+ xx = _mm_maskz_load_pd (m, p);
+
+ _mm256_mask_store_pd (p, m, yy);
+ _mm_mask_store_pd (p, m, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovapd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovapd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovaps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+float __attribute__ ((aligned (32))) *p;
+volatile __m256 yy, y2;
+volatile __m128 xx, x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_mov_ps (yy, m, y2);
+ xx = _mm_mask_mov_ps (xx, m, x2);
+
+ yy = _mm256_maskz_mov_ps (m, y2);
+ xx = _mm_maskz_mov_ps (m, x2);
+
+ yy = _mm256_mask_load_ps (yy, m, p);
+ xx = _mm_mask_load_ps (xx, m, p);
+
+ yy = _mm256_maskz_load_ps (m, p);
+ xx = _mm_maskz_load_ps (m, p);
+
+ _mm256_mask_store_ps (p, m, yy);
+ _mm_mask_store_ps (p, m, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovaps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovaps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovddup\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vunpcklpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovddup\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]|vunpcklpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovddup\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}|vunpcklpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovddup\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}|vunpcklpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2;
+volatile __m128d xx, x2;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_movedup_pd (yy, m8, y2);
+ yy = _mm256_maskz_movedup_pd (m8, y2);
+
+ xx = _mm_mask_movedup_pd (xx, m8, x2);
+ xx = _mm_maskz_movedup_pd (m8, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovddup-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovddup-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa32\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+int *p;
+volatile __m256i yy, y2;
+volatile __m128i xx, x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_mov_epi32 (yy, m, y2);
+ xx = _mm_mask_mov_epi32 (xx, m, x2);
+
+ yy = _mm256_maskz_mov_epi32 (m, y2);
+ xx = _mm_maskz_mov_epi32 (m, x2);
+
+ yy = _mm256_mask_load_epi32 (yy, m, p);
+ xx = _mm_mask_load_epi32 (xx, m, p);
+
+ yy = _mm256_maskz_load_epi32 (m, p);
+ xx = _mm_maskz_load_epi32 (m, p);
+
+ _mm256_mask_store_epi32 (p, m, yy);
+ _mm_mask_store_epi32 (p, m, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovdqa32-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovdqa32-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\\(\[^\n\]*%ymm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\\(\[^\n\]*%xmm\[0-9\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^x^y\]*\\(" 2 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^x^y\]*\\(" 2 { xfail *-*-* } } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqa64\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+long long *p;
+__m256i const *p1;
+__m128i const *p2;
+volatile __m256i yy, y2;
+volatile __m128i xx, x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_mov_epi64 (yy, m, y2);
+ xx = _mm_mask_mov_epi64 (xx, m, x2);
+
+ yy = _mm256_maskz_mov_epi64 (m, y2);
+ xx = _mm_maskz_mov_epi64 (m, x2);
+
+ yy = _mm256_load_si256 (p1);
+ xx = _mm_load_si128 (p2);
+
+ yy = _mm256_mask_load_epi64 (yy, m, p);
+ xx = _mm_mask_load_epi64 (xx, m, p);
+
+ yy = _mm256_maskz_load_epi64 (m, p);
+ xx = _mm_maskz_load_epi64 (m, p);
+
+ _mm256_store_epi64 (p, yy);
+ _mm_store_epi64 (p, xx);
+
+ _mm256_mask_store_epi64 (p, m, yy);
+ _mm_mask_store_epi64 (p, m, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovdqa64-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovdqa64-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vmovdqu16-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vmovdqu16-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovdqu32\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu32\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu32\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu32\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu32\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu32\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+int *p;
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_loadu_epi32 (x1, m, p);
+ x2 = _mm_mask_loadu_epi32 (x2, m, p);
+
+ x1 = _mm256_maskz_loadu_epi32 (m, p);
+ x2 = _mm_maskz_loadu_epi32 (m, p);
+
+ _mm256_mask_storeu_epi32 (p, m, x1);
+ _mm_mask_storeu_epi32 (p, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovdqu32-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovdqu32-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovdqu64\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu64\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu64\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu64\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu64\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovdqu64\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+long long *p;
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_loadu_epi64 (x1, m, p);
+ x2 = _mm_mask_loadu_epi64 (x2, m, p);
+
+ x1 = _mm256_maskz_loadu_epi64 (m, p);
+ x2 = _mm_maskz_loadu_epi64 (m, p);
+
+ _mm256_mask_storeu_epi64 (p, m, x1);
+ _mm_mask_storeu_epi64 (p, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovdqu64-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovdqu64-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vmovdqu8-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vmovdqu8-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vmovntdqa\[ \\t\]+\[^\n\]*%zmm\[0-9\]"} } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+__m512i *y;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm512_stream_load_si512 (y);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovshdup\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovshdup\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovshdup\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovshdup\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 y;
+volatile __m128 x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_movehdup_ps (y, m, y);
+ x = _mm_mask_movehdup_ps (x, m, x);
+
+ y = _mm256_maskz_movehdup_ps (m, y);
+ x = _mm_maskz_movehdup_ps (m, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovshdup-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovshdup-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovsldup\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovsldup\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovsldup\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovsldup\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 y;
+volatile __m128 x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_moveldup_ps (y, m, y);
+ x = _mm_mask_moveldup_ps (x, m, x);
+
+ y = _mm256_maskz_moveldup_ps (m, y);
+ x = _mm_maskz_moveldup_ps (m, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovsldup-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovsldup-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovupd\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovupd\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovupd\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovupd\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovupd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovupd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+double *p;
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_loadu_pd (x1, m, p);
+ x2 = _mm_mask_loadu_pd (x2, m, p);
+
+ x1 = _mm256_maskz_loadu_pd (m, p);
+ x2 = _mm_maskz_loadu_pd (m, p);
+
+ _mm256_mask_storeu_pd (p, m, x1);
+ _mm_mask_storeu_pd (p, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovupd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovupd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vmovups\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovups\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovups\[ \\t\]+\[^\n\]*\\)\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovups\[ \\t\]+\[^\n\]*\\)\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmovups\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmovups\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+float *p;
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_loadu_ps (x1, m, p);
+ x2 = _mm_mask_loadu_ps (x2, m, p);
+
+ x1 = _mm256_maskz_loadu_ps (m, p);
+ x2 = _mm_maskz_loadu_ps (m, p);
+
+ _mm256_mask_storeu_ps (p, m, x1);
+ _mm_mask_storeu_ps (p, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovups-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmovups-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vmulpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vmulpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmulpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmulpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vmulpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmulpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x256;
+volatile __m128d x128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_mul_pd (x256, m, x256, x256);
+ x256 = _mm256_maskz_mul_pd (m, x256, x256);
+ x128 = _mm_mask_mul_pd (x128, m, x128, x128);
+ x128 = _mm_maskz_mul_pd (m, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmulpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmulpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vmulps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vmulps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmulps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmulps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vmulps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vmulps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x256;
+volatile __m128 x128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_mul_ps (x256, m, x256, x256);
+ x256 = _mm256_maskz_mul_ps (m, x256, x256);
+ x128 = _mm_mask_mul_ps (x128, m, x128, x128);
+ x128 = _mm_maskz_mul_ps (m, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmulps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vmulps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vorpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vorpd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vorps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vorps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpabsb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpabsb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpabsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i xx;
+volatile __m128i x2;
+
+void extern
+avx512vl_test (void)
+{
+ xx = _mm256_mask_abs_epi32 (xx, 2,xx);
+ xx = _mm256_maskz_abs_epi32 (2, xx);
+ x2 = _mm_mask_abs_epi32 (x2, 2, x2);
+ x2 = _mm_maskz_abs_epi32 (2, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpabsd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpabsd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpabsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpabsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpabsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpabsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i xx;
+volatile __m128i x2;
+
+void extern
+avx512vl_test (void)
+{
+ xx = _mm256_abs_epi64 (xx);
+ xx = _mm256_mask_abs_epi64 (xx, 2, xx);
+ xx = _mm256_maskz_abs_epi64 (2, xx);
+ x2 = _mm_abs_epi64 (x2);
+ x2 = _mm_mask_abs_epi64 (x2, 2, x2);
+ x2 = _mm_maskz_abs_epi64 (2, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpabsq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpabsq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpabsw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpabsw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpackssdw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpackssdw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpacksswb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpacksswb-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpackusdw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpackusdw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpackuswb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpackuswb-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpaddd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_add_epi32 (x256, m8, x256, x256);
+ x256 = _mm256_maskz_add_epi32 (m8, x256, x256);
+ x128 = _mm_mask_add_epi32 (x128, m8, x128, x128);
+ x128 = _mm_maskz_add_epi32 (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpaddd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpaddd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpaddq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpaddq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpaddq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_add_epi64 (x256, m8, x256, x256);
+ x256 = _mm256_maskz_add_epi64 (m8, x256, x256);
+ x128 = _mm_mask_add_epi64 (x128, m8, x128, x128);
+ x128 = _mm_maskz_add_epi64 (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpaddq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpaddq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddsb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddsb-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddsw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddsw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddusb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddusb-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddusw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddusw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpaddw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpalignr-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpalignr-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpandd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpandd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpandd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpandd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpandd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpandd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_and_epi32 (y, m, y, y);
+ y = _mm256_maskz_and_epi32 (m, y, y);
+
+ x = _mm_mask_and_epi32 (x, m, x, x);
+ x = _mm_maskz_and_epi32 (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpandd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpandd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpandnd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpandnd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpandnd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpandnd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpandnd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpandnd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_andnot_epi32 (y, m, y, y);
+ y = _mm256_maskz_andnot_epi32 (m, y, y);
+
+ x = _mm_mask_andnot_epi32 (x, m, x, x);
+ x = _mm_maskz_andnot_epi32 (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpandnd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpandnd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpandnq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpandnq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpandnq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpandnq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpandnq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpandnq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_andnot_epi64 (y, m, y, y);
+ y = _mm256_maskz_andnot_epi64 (m, y, y);
+
+ x = _mm_mask_andnot_epi64 (x, m, x, x);
+ x = _mm_maskz_andnot_epi64 (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpandnq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpandnq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpandq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpandq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpandq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpandq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpandq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpandq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_and_epi64 (y, m, y, y);
+ y = _mm256_maskz_and_epi64 (m, y, y);
+
+ x = _mm_mask_and_epi64 (x, m, x, x);
+ x = _mm_maskz_and_epi64 (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpandq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpandq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpavgb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpavgb-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpavgw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpavgw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpblendmb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpblendmb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "(vpblendmd|vmovdqa32)\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "(vpblendmd|vmovdqa32)\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_blend_epi32 (m, x, x);
+ xx = _mm_mask_blend_epi32 (m, xx, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpblendmd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpblendmd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "(vpblendmq|vmovdqa64)\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "(vpblendmq|vmovdqa64)\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_blend_epi64 (m, x, x);
+ xx = _mm_mask_blend_epi64 (m, xx, xx);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpblendmq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpblendmq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpblendmw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpblendmw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpbroadcastb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpbroadcastb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpbroadcastd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastd\[ \\t\]+%e\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 { target { ! { ia32 } } } } } */
+/* { dg-final { scan-assembler-times "vpbroadcastd\[ \\t\]+%e\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 { target { ! { ia32 } } } } } */
+/* { dg-final { scan-assembler-times "vpbroadcastd\[ \\t\]+%e\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 { target { ! { ia32 } } } } } */
+/* { dg-final { scan-assembler-times "vpbroadcastd\[ \\t\]+%e\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 { target { ! { ia32 } } } } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile int z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_broadcastd_epi32 (x, m, y);
+ x = _mm256_maskz_broadcastd_epi32 (m, y);
+ y = _mm_mask_broadcastd_epi32 (y, m, y);
+ y = _mm_maskz_broadcastd_epi32 (m, y);
+
+ x = _mm256_mask_set1_epi32 (x, m, z);
+ x = _mm256_maskz_set1_epi32 (m, z);
+ y = _mm_mask_set1_epi32 (y, m, z);
+ y = _mm_maskz_set1_epi32 (m, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpbroadcastd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpbroadcastd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512cd -O2" } */
+/* { dg-final { scan-assembler "vpbroadcastmb2q\[ \\t\]+\[^\n\]*k\[1-7\]\[^\n\]*%xmm\[0-7\]" } } */
+/* { dg-final { scan-assembler "vpbroadcastmb2q\[ \\t\]+\[^\n\]*k\[1-7\]\[^\n\]*%ymm\[0-7\]" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x128 = _mm_broadcastmb_epi64 (m8);
+ x256 = _mm256_broadcastmb_epi64 (m8);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512cd -O2" } */
+/* { dg-final { scan-assembler "vpbroadcastmw2d\[ \\t\]+\[^\n\]*k\[1-7\]\[^\n\]*%xmm\[0-7\]" } } */
+/* { dg-final { scan-assembler "vpbroadcastmw2d\[ \\t\]+\[^\n\]*k\[1-7\]\[^\n\]*%ymm\[0-7\]" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __mmask16 m16;
+
+void extern
+avx512vl_test (void)
+{
+ x128 = _mm_broadcastmw_epi32 (m16);
+ x256 = _mm256_broadcastmw_epi32 (m16);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpbroadcastq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcastq\[ \\t\]+%r\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 { target { ! { ia32 } } } } } */
+/* { dg-final { scan-assembler-times "vpbroadcastq\[ \\t\]+%r\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 { target { ! { ia32 } } } } } */
+/* { dg-final { scan-assembler-times "vpbroadcastq\[ \\t\]+%r\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 { target { ! { ia32 } } } } } */
+/* { dg-final { scan-assembler-times "vpbroadcastq\[ \\t\]+%r\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 { target { ! { ia32 } } } } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile long long z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_broadcastq_epi64 (x, m, y);
+ x = _mm256_maskz_broadcastq_epi64 (m, y);
+ y = _mm_mask_broadcastq_epi64 (y, m, y);
+ y = _mm_maskz_broadcastq_epi64 (m, y);
+
+ x = _mm256_mask_set1_epi64 (x, m, z);
+ x = _mm256_maskz_set1_epi64 (m, z);
+ y = _mm_mask_set1_epi64 (y, m, z);
+ y = _mm_maskz_set1_epi64 (m, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpbroadcastq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpbroadcastq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpbroadcastw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpbroadcastw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler "vpcmpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm256_cmp_epi32_mask (x, x, _MM_CMPINT_EQ);
+ m = _mm256_mask_cmp_epi32_mask (m, x, x, _MM_CMPINT_EQ);
+ m = _mm_cmp_epi32_mask (xx, xx, _MM_CMPINT_EQ);
+ m = _mm_mask_cmp_epi32_mask (m, xx, xx, _MM_CMPINT_EQ);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpeqb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpeqb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpcmpeqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpeqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpeqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcmpeqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm_cmpeq_epi32_mask (x128, x128);
+ m = _mm256_cmpeq_epi32_mask (x256, x256);
+ m = _mm_mask_cmpeq_epi32_mask (3, x128, x128);
+ m = _mm256_mask_cmpeq_epi32_mask (3, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpeqd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpeqd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpcmpeqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpeqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpeqq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpeqq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm_cmpeq_epi64_mask (x128, x128);
+ m = _mm256_cmpeq_epi64_mask (x256, x256);
+ m = _mm_mask_cmpeq_epi64_mask (3, x128, x128);
+ m = _mm256_mask_cmpeq_epi64_mask (3, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpeqq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpeqq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpeqw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpeqw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpged-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpged-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgeq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgeq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgeud-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgeud-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgeuq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgeuq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpgtb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpgtb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpcmpgtd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpgtd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpgtd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpgtd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm_cmpgt_epi32_mask (x128, x128);
+ m = _mm256_cmpgt_epi32_mask (x256, x256);
+ m = _mm_mask_cmpgt_epi32_mask (3, x128, x128);
+ m = _mm256_mask_cmpgt_epi32_mask (3, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgtd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgtd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vpcmpgtq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpgtq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpgtq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vpcmpgtq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%k\[1-7\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm_cmpgt_epi64_mask (x128, x128);
+ m = _mm256_cmpgt_epi64_mask (x256, x256);
+ m = _mm_mask_cmpgt_epi64_mask (3, x128, x128);
+ m = _mm256_mask_cmpgt_epi64_mask (3, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgtq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpgtq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpgtw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpgtw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpled-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpled-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpleq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpleq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpleud-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpleud-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpleuq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpleuq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpltd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpltd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpltq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpltq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpltud-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpltud-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpltuq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpltuq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpneqd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpneqd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpneqq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpneqq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpnequd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpnequd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpnequq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpnequq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler "vpcmpq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask8 m;
+
+void extern
+avx512bw_test (void)
+{
+ m = _mm256_cmp_epi64_mask (x, x, _MM_CMPINT_UNUSED);
+ m = _mm256_mask_cmp_epi64_mask (m, x, x, _MM_CMPINT_NE);
+ m = _mm_cmp_epi64_mask (xx, xx, _MM_CMPINT_NLT);
+ m = _mm_mask_cmp_epi64_mask (m, xx, xx, _MM_CMPINT_GE);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpub-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpub-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler "vpcmpud\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpud\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpud\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpud\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm256_cmp_epu32_mask (x, x, _MM_CMPINT_LE);
+ m = _mm256_mask_cmp_epu32_mask (m, x, x, _MM_CMPINT_UNUSED);
+ m = _mm_cmp_epu32_mask (xx, xx, _MM_CMPINT_NE);
+ m = _mm_mask_cmp_epu32_mask (m, xx, xx, _MM_CMPINT_NLT);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpud-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpud-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler "vpcmpuq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpuq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+/* { dg-final { scan-assembler "vpcmpuq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vpcmpuq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n^k\]*%k\[1-7\]\{" } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i xx;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm256_cmp_epu64_mask (x, x, _MM_CMPINT_NLE);
+ m = _mm256_mask_cmp_epu64_mask (m, x, x, _MM_CMPINT_GT);
+ m = _mm_cmp_epu64_mask (xx, xx, _MM_CMPINT_EQ);
+ m = _mm_mask_cmp_epu64_mask (m, xx, xx, _MM_CMPINT_LT);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpuq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcmpuq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpuw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpuw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpcmpw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpcompressd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+int *p;
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_compress_epi32 (x1, m, x1);
+ x2 = _mm_mask_compress_epi32 (x2, m, x2);
+
+ x1 = _mm256_maskz_compress_epi32 (m, x1);
+ x2 = _mm_maskz_compress_epi32 (m, x2);
+
+ _mm256_mask_compressstoreu_epi32 (p, m, x1);
+ _mm_mask_compressstoreu_epi32 (p, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcompressd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcompressd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpcompressq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpcompressq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*\\)\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+long long *p;
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_compress_epi64 (x1, m, x1);
+ x2 = _mm_mask_compress_epi64 (x2, m, x2);
+
+ x1 = _mm256_maskz_compress_epi64 (m, x1);
+ x2 = _mm_maskz_compress_epi64 (m, x2);
+
+ _mm256_mask_compressstoreu_epi64 (p, m, x1);
+ _mm_mask_compressstoreu_epi64 (p, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcompressq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpcompressq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512cd -O2" } */
+/* { dg-final { scan-assembler-times "vpconflictd\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpconflictd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpconflictd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpconflictd\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpconflictd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpconflictd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i s;
+volatile __m256i res;
+volatile __m128i s2;
+volatile __m128i res2;
+
+void extern
+avx512vl_test (void)
+{
+ res = _mm256_conflict_epi32 (s);
+ res = _mm256_mask_conflict_epi32 (res, 2, s);
+ res = _mm256_maskz_conflict_epi32 (2, s);
+ res2 = _mm_conflict_epi32 (s2);
+ res2 = _mm_mask_conflict_epi32 (res2, 2, s2);
+ res2 = _mm_maskz_conflict_epi32 (2, s2);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512cd -O2" } */
+/* { dg-final { scan-assembler-times "vpconflictq\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpconflictq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpconflictq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpconflictq\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpconflictq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpconflictq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i s;
+volatile __m256i res;
+volatile __m128i s2;
+volatile __m128i res2;
+
+void extern
+avx512vl_test (void)
+{
+ res = _mm256_conflict_epi64 (s);
+ res = _mm256_mask_conflict_epi64 (res, 2, s);
+ res = _mm256_maskz_conflict_epi64 (2, s);
+ res2 = _mm_conflict_epi64 (s2);
+ res2 = _mm_mask_conflict_epi64 (res2, 2, s2);
+ res2 = _mm_maskz_conflict_epi64 (2, s2);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_maskz_permutexvar_epi32 (m, x, x);
+ x = _mm256_mask_permutexvar_epi32 (x, m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermd-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermi2d\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermi2d\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask2_permutex2var_epi32 (x1, y, m, x1);
+ x2 = _mm_mask2_permutex2var_epi32 (x2, z, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermi2d-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermi2d-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermi2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermi2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask2_permutex2var_pd (x1, y, m, x1);
+ x2 = _mm_mask2_permutex2var_pd (x2, z, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermi2pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermi2pd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermi2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermi2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask2_permutex2var_ps (x1, y, m, x1);
+ x2 = _mm_mask2_permutex2var_ps (x2, z, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermi2ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermi2ps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermi2q\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermi2q\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask2_permutex2var_epi64 (x1, y, m, x1);
+ x2 = _mm_mask2_permutex2var_epi64 (x2, z, m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermi2q-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermi2q-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpermi2w-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpermi2w-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d y;
+volatile __m256i c;
+volatile __m128d x;
+volatile __m128i k;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_permutevar_pd (y, m, y, c);
+ y = _mm256_maskz_permutevar_pd (m, y, c);
+ x = _mm_mask_permutevar_pd (x, m, x, k);
+ x = _mm_maskz_permutevar_pd (m, x, k);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermilpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermilpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 { xfail *-*-* } } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 { xfail *-*-* } } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 { xfail *-*-* } } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*3\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*3\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilpd\[ \\t\]+\[^\n\]*3\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d y;
+volatile __m128d x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_permute_pd (y, m, y, 13);
+ y = _mm256_maskz_permute_pd (m, y, 13);
+ x = _mm_mask_permute_pd (x, m, x, 3);
+ x = _mm_maskz_permute_pd (m, x, 3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermilpdi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermilpdi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 y;
+volatile __m128 x;
+volatile __m256i c;
+volatile __m128i k;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_permutevar_ps (y, m, y, c);
+ y = _mm256_maskz_permutevar_ps (m, y, c);
+ x = _mm_mask_permutevar_ps (x, m, x, k);
+ x = _mm_maskz_permutevar_ps (m, x, k);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermilps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermilps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*13\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*13\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*13\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermilps\[ \\t\]+\[^\n\]*13\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 y;
+volatile __m128 x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_permute_ps (y, m, y, 13);
+ y = _mm256_maskz_permute_ps (m, y, 13);
+ x = _mm_mask_permute_ps (x, m, x, 13);
+ x = _mm_maskz_permute_ps (m, x, 13);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermilpsi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermilpsi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m256d y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_permutexvar_pd (x, y);
+ y = _mm256_mask_permutexvar_pd (y, m, x, y);
+ y = _mm256_maskz_permutexvar_pd (m, x, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermpd-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_permutex_pd (x, 13);
+ x = _mm256_mask_permutex_pd (x, m, x, 13);
+ x = _mm256_maskz_permutex_pd (m, x, 13);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermpdi-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m256 y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_permutexvar_ps (x, y);
+ y = _mm256_mask_permutexvar_ps (y, m, x, y);
+ y = _mm256_maskz_permutexvar_ps (m, x, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermps-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_permutex_epi64 (x, m, x, 13);
+ x = _mm256_maskz_permutex_epi64 (m, x, 13);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermq-imm-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_maskz_permutexvar_epi64 (m, x, x);
+ x = _mm256_mask_permutexvar_epi64 (x, m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermq-var-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermt2d\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2d\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2d\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2d\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2d\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2d\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_permutex2var_epi32 (x1, y, x1);
+ x1 = _mm256_mask_permutex2var_epi32 (x1, m, y, x1);
+ x1 = _mm256_maskz_permutex2var_epi32 (m, x1, y, x1);
+ x2 = _mm_permutex2var_epi32 (x2, z, x2);
+ x2 = _mm_mask_permutex2var_epi32 (x2, m, z, x2);
+ x2 = _mm_maskz_permutex2var_epi32 (m, x2, z, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermt2d-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermt2d-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermt2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_permutex2var_pd (x1, y, x1);
+ x1 = _mm256_mask_permutex2var_pd (x1, m, y, x1);
+ x1 = _mm256_maskz_permutex2var_pd (m, x1, y, x1);
+ x2 = _mm_permutex2var_pd (x2, z, x2);
+ x2 = _mm_mask_permutex2var_pd (x2, m, z, x2);
+ x2 = _mm_maskz_permutex2var_pd (m, x2, z, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermt2pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermt2pd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermt2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_permutex2var_ps (x1, y, x1);
+ x1 = _mm256_mask_permutex2var_ps (x1, m, y, x1);
+ x1 = _mm256_maskz_permutex2var_ps (m, x1, y, x1);
+ x2 = _mm_permutex2var_ps (x2, z, x2);
+ x2 = _mm_mask_permutex2var_ps (x2, m, z, x2);
+ x2 = _mm_maskz_permutex2var_ps (m, x2, z, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermt2ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermt2ps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpermt2q\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2q\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpermt2q\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2q\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2q\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpermt2q\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_permutex2var_epi64 (x1, y, x1);
+ x1 = _mm256_mask_permutex2var_epi64 (x1, m, y, x1);
+ x1 = _mm256_maskz_permutex2var_epi64 (m, x1, y, x1);
+ x2 = _mm_permutex2var_epi64 (x2, z, x2);
+ x2 = _mm_mask_permutex2var_epi64 (x2, m, z, x2);
+ x2 = _mm_maskz_permutex2var_epi64 (m, x2, z, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermt2q-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpermt2q-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpermt2w-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpermt2w-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpermw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpermw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpexpandd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpexpandd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpexpandd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpexpandd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+
+#include <immintrin.h>
+
+int *p;
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_expand_epi32 (x1, m, x1);
+ x2 = _mm_mask_expand_epi32 (x2, m, x2);
+
+ x1 = _mm256_maskz_expand_epi32 (m, x1);
+ x2 = _mm_maskz_expand_epi32 (m, x2);
+
+ x1 = _mm256_mask_expandloadu_epi32 (x1, m, p);
+ x2 = _mm_mask_expandloadu_epi32 (x2, m, p);
+
+ x1 = _mm256_maskz_expandloadu_epi32 (m, p);
+ x2 = _mm_maskz_expandloadu_epi32 (m, p);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpexpandd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpexpandd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpexpandq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpexpandq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpexpandq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpexpandq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 2 } } */
+
+#include <immintrin.h>
+
+long long *p;
+volatile __m256i x1;
+volatile __m128i x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_expand_epi64 (x1, m, x1);
+ x2 = _mm_mask_expand_epi64 (x2, m, x2);
+
+ x1 = _mm256_maskz_expand_epi64 (m, x1);
+ x2 = _mm_maskz_expand_epi64 (m, x2);
+
+ x1 = _mm256_mask_expandloadu_epi64 (x1, m, p);
+ x2 = _mm_mask_expandloadu_epi64 (x2, m, p);
+
+ x1 = _mm256_maskz_expandloadu_epi64 (m, p);
+ x2 = _mm_maskz_expandloadu_epi64 (m, p);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpexpandq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpexpandq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512cd -O2" } */
+/* { dg-final { scan-assembler-times "vplzcntd\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vplzcntd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1} } */
+/* { dg-final { scan-assembler-times "vplzcntd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vplzcntd\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vplzcntd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1} } */
+/* { dg-final { scan-assembler-times "vplzcntd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i s;
+volatile __m256i res;
+volatile __m128i s2;
+volatile __m128i res2;
+
+void extern
+avx512vl_test (void)
+{
+ res = _mm256_lzcnt_epi32 (s);
+ res = _mm256_mask_lzcnt_epi32 (res, 2, s);
+ res = _mm256_maskz_lzcnt_epi32 (2, s);
+ res2 = _mm_lzcnt_epi32 (s2);
+ res2 = _mm_mask_lzcnt_epi32 (res2, 2, s2);
+ res2 = _mm_maskz_lzcnt_epi32 (2, s2);
+}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512cd -O2" } */
+/* { dg-final { scan-assembler-times "vplzcntq\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vplzcntq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vplzcntq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vplzcntq\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vplzcntq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vplzcntq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i s;
+volatile __m256i res;
+volatile __m128i s2;
+volatile __m128i res2;
+
+void extern
+avx512vl_test (void)
+{
+ res = _mm256_lzcnt_epi64 (s);
+ res = _mm256_maskz_lzcnt_epi64 (2, s);
+ res = _mm256_mask_lzcnt_epi64 (res, 2, s);
+ res2 = _mm_lzcnt_epi64 (s2);
+ res2 = _mm_maskz_lzcnt_epi64 (2, s2);
+ res2 = _mm_mask_lzcnt_epi64 (res2, 2, s2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaddubsw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaddubsw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaddwd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaddwd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaxsb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaxsb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmaxsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_max_epi32 (x, m, x, x);
+ x = _mm256_maskz_max_epi32 (m, x, x);
+ y = _mm_mask_max_epi32 (y, m, y, y);
+ y = _mm_maskz_max_epi32 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmaxsd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmaxsd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512bw -O2" } */
+/* { dg-final { scan-assembler-times "vpmaxsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaxsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaxsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_max_epi64 (x, x);
+ x = _mm256_mask_max_epi64 (x, m, x, x);
+ x = _mm256_maskz_max_epi64 (m, x, x);
+ y = _mm_max_epi64 (y, y);
+ y = _mm_mask_max_epi64 (y, m, y, y);
+ y = _mm_maskz_max_epi64 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmaxsq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmaxsq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaxsw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaxsw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaxub-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaxub-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmaxud\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxud\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxud\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxud\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_max_epu32 (x, m, x, x);
+ x = _mm256_maskz_max_epu32 (m, x, x);
+ y = _mm_mask_max_epu32 (y, m, y, y);
+ y = _mm_maskz_max_epu32 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmaxud-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmaxud-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512bw -O2" } */
+/* { dg-final { scan-assembler-times "vpmaxuq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaxuq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxuq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxuq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpmaxuq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmaxuq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_max_epu64 (x, x);
+ x = _mm256_mask_max_epu64 (x, m, x, x);
+ x = _mm256_maskz_max_epu64 (m, x, x);
+ y = _mm_max_epu64 (y, y);
+ y = _mm_mask_max_epu64 (y, m, y, y);
+ y = _mm_maskz_max_epu64 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmaxuq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmaxuq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaxuw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmaxuw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpminsb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpminsb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpminsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_min_epi32 (x, m, x, x);
+ x = _mm256_maskz_min_epi32 (m, x, x);
+ y = _mm_mask_min_epi32 (y, m, y, y);
+ y = _mm_maskz_min_epi32 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpminsd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpminsd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512bw -O2" } */
+/* { dg-final { scan-assembler-times "vpminsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpminsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpminsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminsq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_min_epi64 (x, x);
+ x = _mm256_mask_min_epi64 (x, m, x, x);
+ x = _mm256_maskz_min_epi64 (m, x, x);
+ y = _mm_min_epi64 (y, y);
+ y = _mm_mask_min_epi64 (y, m, y, y);
+ y = _mm_maskz_min_epi64 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpminsq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpminsq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpminsw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpminsw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpminub-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpminub-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpminud\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminud\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminud\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminud\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_min_epu32 (x, m, x, x);
+ x = _mm256_maskz_min_epu32 (m, x, x);
+ y = _mm_mask_min_epu32 (y, m, y, y);
+ y = _mm_maskz_min_epu32 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpminud-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpminud-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512bw -O2" } */
+/* { dg-final { scan-assembler-times "vpminuq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpminuq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminuq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpminuq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpminuq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpminuq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_min_epu64 (x, x);
+ x = _mm256_mask_min_epu64 (x, m, x, x);
+ x = _mm256_maskz_min_epu64 (m, x, x);
+ y = _mm_min_epu64 (y, y);
+ y = _mm_mask_min_epu64 (y, m, y, y);
+ y = _mm_maskz_min_epu64 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpminuq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpminuq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpminuw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpminuw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovb2m-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovb2m-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmovd2m-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmovd2m-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtepi32_epi8 (x);
+ z = _mm_mask_cvtepi32_epi8 (z, m, x);
+ z = _mm_maskz_cvtepi32_epi8 (m, x);
+ z = _mm256_cvtepi32_epi8 (y);
+ z = _mm256_mask_cvtepi32_epi8 (z, m, y);
+ z = _mm256_maskz_cvtepi32_epi8 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovdb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovdb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtepi32_epi16 (x);
+ z = _mm_mask_cvtepi32_epi16 (z, m, x);
+ z = _mm_maskz_cvtepi32_epi16 (m, x);
+ z = _mm256_cvtepi32_epi16 (y);
+ z = _mm256_mask_cvtepi32_epi16 (z, m, y);
+ z = _mm256_maskz_cvtepi32_epi16 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovdw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovdw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovm2b-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovm2b-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmovm2d-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmovm2d-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmovm2q-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmovm2q-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovm2w-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovm2w-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmovq2m-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmovq2m-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtepi64_epi8 (x);
+ z = _mm_mask_cvtepi64_epi8 (z, m, x);
+ z = _mm_maskz_cvtepi64_epi8 (m, x);
+ z = _mm256_cvtepi64_epi8 (y);
+ z = _mm256_mask_cvtepi64_epi8 (z, m, y);
+ z = _mm256_maskz_cvtepi64_epi8 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovqb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovqb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtepi64_epi32 (x);
+ z = _mm_mask_cvtepi64_epi32 (z, m, x);
+ z = _mm_maskz_cvtepi64_epi32 (m, x);
+ z = _mm256_cvtepi64_epi32 (y);
+ z = _mm256_mask_cvtepi64_epi32 (z, m, y);
+ z = _mm256_maskz_cvtepi64_epi32 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovqd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovqd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtepi64_epi16 (x);
+ z = _mm_mask_cvtepi64_epi16 (z, m, x);
+ z = _mm_maskz_cvtepi64_epi16 (m, x);
+ z = _mm256_cvtepi64_epi16 (y);
+ z = _mm256_mask_cvtepi64_epi16 (z, m, y);
+ z = _mm256_maskz_cvtepi64_epi16 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovqw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovqw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtsepi32_epi8 (x);
+ z = _mm_mask_cvtsepi32_epi8 (z, m, x);
+ z = _mm_maskz_cvtsepi32_epi8 (m, x);
+ z = _mm256_cvtsepi32_epi8 (y);
+ z = _mm256_mask_cvtsepi32_epi8 (z, m, y);
+ z = _mm256_maskz_cvtsepi32_epi8 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsdb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsdb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtsepi32_epi16 (x);
+ z = _mm_mask_cvtsepi32_epi16 (z, m, x);
+ z = _mm_maskz_cvtsepi32_epi16 (m, x);
+ z = _mm256_cvtsepi32_epi16 (y);
+ z = _mm256_mask_cvtsepi32_epi16 (z, m, y);
+ z = _mm256_maskz_cvtsepi32_epi16 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsdw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsdw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtsepi64_epi8 (x);
+ z = _mm_mask_cvtsepi64_epi8 (z, m, x);
+ z = _mm_maskz_cvtsepi64_epi8 (m, x);
+ z = _mm256_cvtsepi64_epi8 (y);
+ z = _mm256_mask_cvtsepi64_epi8 (z, m, y);
+ z = _mm256_maskz_cvtsepi64_epi8 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsqb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsqb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtsepi64_epi32 (x);
+ z = _mm_mask_cvtsepi64_epi32 (z, m, x);
+ z = _mm_maskz_cvtsepi64_epi32 (m, x);
+ z = _mm256_cvtsepi64_epi32 (y);
+ z = _mm256_mask_cvtsepi64_epi32 (z, m, y);
+ z = _mm256_maskz_cvtsepi64_epi32 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsqd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsqd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtsepi64_epi16 (x);
+ z = _mm_mask_cvtsepi64_epi16 (z, m, x);
+ z = _mm_maskz_cvtsepi64_epi16 (m, x);
+ z = _mm256_cvtsepi64_epi16 (y);
+ z = _mm256_mask_cvtsepi64_epi16 (z, m, y);
+ z = _mm256_maskz_cvtsepi64_epi16 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsqw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsqw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovswb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovswb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsxbd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepi8_epi32 (res1, m, s);
+ res2 = _mm_mask_cvtepi8_epi32 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepi8_epi32 (m, s);
+ res2 = _mm_maskz_cvtepi8_epi32 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxbd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxbd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsxbq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxbq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepi8_epi64 (res1, m, s);
+ res2 = _mm_mask_cvtepi8_epi64 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepi8_epi64 (m, s);
+ res2 = _mm_maskz_cvtepi8_epi64 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxbq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxbq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovsxbw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovsxbw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsxdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepi32_epi64 (res1, m, s);
+ res2 = _mm_mask_cvtepi32_epi64 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepi32_epi64 (m, s);
+ res2 = _mm_maskz_cvtepi32_epi64 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxdq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxdq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsxwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepi16_epi32 (res1, m, s);
+ res2 = _mm_mask_cvtepi16_epi32 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepi16_epi32 (m, s);
+ res2 = _mm_maskz_cvtepi16_epi32 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxwd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxwd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovsxwq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxwq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxwq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovsxwq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepi16_epi64 (res1, m, s);
+ res2 = _mm_mask_cvtepi16_epi64 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepi16_epi64 (m, s);
+ res2 = _mm_maskz_cvtepi16_epi64 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxwq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovsxwq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovusdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtusepi32_epi8 (x);
+ z = _mm_mask_cvtusepi32_epi8 (z, m, x);
+ z = _mm_maskz_cvtusepi32_epi8 (m, x);
+ z = _mm256_cvtusepi32_epi8 (y);
+ z = _mm256_mask_cvtusepi32_epi8 (z, m, y);
+ z = _mm256_maskz_cvtusepi32_epi8 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusdb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusdb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovusdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusdw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtusepi32_epi16 (x);
+ z = _mm_mask_cvtusepi32_epi16 (z, m, x);
+ z = _mm_maskz_cvtusepi32_epi16 (m, x);
+ z = _mm256_cvtusepi32_epi16 (y);
+ z = _mm256_mask_cvtusepi32_epi16 (z, m, y);
+ z = _mm256_maskz_cvtusepi32_epi16 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusdw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusdw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovusqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqb\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqb\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtusepi64_epi8 (x);
+ z = _mm_mask_cvtusepi64_epi8 (z, m, x);
+ z = _mm_maskz_cvtusepi64_epi8 (m, x);
+ z = _mm256_cvtusepi64_epi8 (y);
+ z = _mm256_mask_cvtusepi64_epi8 (z, m, y);
+ z = _mm256_maskz_cvtusepi64_epi8 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusqb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusqb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovusqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtusepi64_epi32 (x);
+ z = _mm_mask_cvtusepi64_epi32 (z, m, x);
+ z = _mm_maskz_cvtusepi64_epi32 (m, x);
+ z = _mm256_cvtusepi64_epi32 (y);
+ z = _mm256_mask_cvtusepi64_epi32 (z, m, y);
+ z = _mm256_maskz_cvtusepi64_epi32 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusqd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusqd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovusqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqw\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\[\\n\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovusqw\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i x, z;
+volatile __m256i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ z = _mm_cvtusepi64_epi16 (x);
+ z = _mm_mask_cvtusepi64_epi16 (z, m, x);
+ z = _mm_maskz_cvtusepi64_epi16 (m, x);
+ z = _mm256_cvtusepi64_epi16 (y);
+ z = _mm256_mask_cvtusepi64_epi16 (z, m, y);
+ z = _mm256_maskz_cvtusepi64_epi16 (m, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusqw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovusqw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovuswb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovuswb-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovw2m-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovw2m-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovwb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovwb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovzxbd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepu8_epi32 (res1, m, s);
+ res2 = _mm_mask_cvtepu8_epi32 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepu8_epi32 (m, s);
+ res2 = _mm_maskz_cvtepu8_epi32 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxbd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxbd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovzxbq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxbq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepu8_epi64 (res1, m, s);
+ res2 = _mm_mask_cvtepu8_epi64 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepu8_epi64 (m, s);
+ res2 = _mm_maskz_cvtepu8_epi64 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxbq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxbq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovzxbw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmovzxbw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovzxdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepu32_epi64 (res1, m, s);
+ res2 = _mm_mask_cvtepu32_epi64 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepu32_epi64 (m, s);
+ res2 = _mm_maskz_cvtepu32_epi64 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxdq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxdq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovzxwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxwd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepu16_epi32 (res1, m, s);
+ res2 = _mm_mask_cvtepu16_epi32 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepu16_epi32 (m, s);
+ res2 = _mm_maskz_cvtepu16_epi32 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxwd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxwd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmovzxwq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxwq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxwq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmovzxwq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128i s, res2;
+volatile __m256i res1;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ res1 = _mm256_mask_cvtepu16_epi64 (res1, m, s);
+ res2 = _mm_mask_cvtepu16_epi64 (res2, m, s);
+
+ res1 = _mm256_maskz_cvtepu16_epi64 (m, s);
+ res2 = _mm_maskz_cvtepu16_epi64 (m, s);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxwq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmovzxwq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmuldq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmuldq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmuldq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmuldq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_mul_epi32 (x, m, x, x);
+ x = _mm256_maskz_mul_epi32 (m, x, x);
+ y = _mm_mask_mul_epi32 (y, m, y, y);
+ y = _mm_maskz_mul_epi32 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmuldq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmuldq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmulhrsw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmulhrsw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmulhuw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmulhuw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmulhw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmulhw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmulld\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulld\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulld\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmulld\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i z;
+volatile __mmask8 myz;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_mullo_epi32 (y, myz, y, y);
+ y = _mm256_maskz_mullo_epi32 (myz, y, y);
+ z = _mm_mask_mullo_epi32 (z, myz, z, z);
+ z = _mm_maskz_mullo_epi32 (myz, z, z);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmulld-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmulld-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmullq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vpmullq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmullw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpmullw-2.c"
--- /dev/null
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpmuludq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmuludq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmuludq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpmuludq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}{z}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_mul_epu32 (x, m, x, x);
+ x = _mm256_maskz_mul_epu32 (m, x, x);
+ y = _mm_mask_mul_epu32 (y, m, y, y);
+ y = _mm_maskz_mul_epu32 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmuludq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpmuludq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_or_epi32 (y, m, y, y);
+ y = _mm256_maskz_or_epi32 (m, y, y);
+
+ x = _mm_mask_or_epi32 (x, m, x, x);
+ x = _mm_maskz_or_epi32 (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpord-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpord-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vporq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vporq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vporq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vporq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vporq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vporq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_or_epi64 (y, m, y, y);
+ y = _mm256_maskz_or_epi64 (m, y, y);
+
+ x = _mm_mask_or_epi64 (x, m, x, x);
+ x = _mm_maskz_or_epi64 (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vporq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vporq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vprold\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprold\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprold\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vprold\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprold\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprold\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_rol_epi32 (x, 11);
+ x = _mm256_mask_rol_epi32 (x, m, x, 11);
+ x = _mm256_maskz_rol_epi32 (m, x, 11);
+
+ y = _mm_rol_epi32 (y, 12);
+ y = _mm_mask_rol_epi32 (y, m, y, 12);
+ y = _mm_maskz_rol_epi32 (m, y, 12);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprold-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprold-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vprolq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprolq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprolq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vprolq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprolq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprolq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_rol_epi64 (x, 11);
+ x = _mm256_mask_rol_epi64 (x, m, x, 11);
+ x = _mm256_maskz_rol_epi64 (m, x, 11);
+
+ y = _mm_rol_epi64 (y, 12);
+ y = _mm_mask_rol_epi64 (y, m, y, 12);
+ y = _mm_maskz_rol_epi64 (m, y, 12);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprolq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprolq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vprolvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprolvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprolvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vprolvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprolvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprolvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_rolv_epi32 (x, x);
+ x = _mm256_mask_rolv_epi32 (x, m, x, x);
+ x = _mm256_maskz_rolv_epi32 (m, x, x);
+
+ y = _mm_rolv_epi32 (y, y);
+ y = _mm_mask_rolv_epi32 (y, m, y, y);
+ y = _mm_maskz_rolv_epi32 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprolvd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprolvd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vprolvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprolvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprolvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vprolvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprolvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprolvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_rolv_epi64 (x, x);
+ x = _mm256_mask_rolv_epi64 (x, m, x, x);
+ x = _mm256_maskz_rolv_epi64 (m, x, x);
+
+ y = _mm_rolv_epi64 (y, y);
+ y = _mm_mask_rolv_epi64 (y, m, y, y);
+ y = _mm_maskz_rolv_epi64 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprolvq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprolvq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vprord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vprord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_ror_epi32 (x, 11);
+ x = _mm256_mask_ror_epi32 (x, m, x, 11);
+ x = _mm256_maskz_ror_epi32 (m, x, 11);
+
+ y = _mm_ror_epi32 (y, 12);
+ y = _mm_mask_ror_epi32 (y, m, y, 12);
+ y = _mm_maskz_ror_epi32 (m, y, 12);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprord-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprord-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vprorq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprorq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprorq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vprorq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprorq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprorq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_ror_epi64 (x, 11);
+ x = _mm256_mask_ror_epi64 (x, m, x, 11);
+ x = _mm256_maskz_ror_epi64 (m, x, 11);
+
+ y = _mm_ror_epi64 (y, 12);
+ y = _mm_mask_ror_epi64 (y, m, y, 12);
+ y = _mm_maskz_ror_epi64 (m, y, 12);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprorq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprorq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vprorvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprorvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprorvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vprorvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprorvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprorvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_rorv_epi32 (x, x);
+ x = _mm256_mask_rorv_epi32 (x, m, x, x);
+ x = _mm256_maskz_rorv_epi32 (m, x, x);
+
+ y = _mm_rorv_epi32 (y, y);
+ y = _mm_mask_rorv_epi32 (y, m, y, y);
+ y = _mm_maskz_rorv_epi32 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprorvd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprorvd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vprorvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprorvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprorvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vprorvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vprorvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vprorvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_rorv_epi64 (x, x);
+ x = _mm256_mask_rorv_epi64 (x, m, x, x);
+ x = _mm256_maskz_rorv_epi64 (m, x, x);
+
+ y = _mm_rorv_epi64 (y, y);
+ y = _mm_mask_rorv_epi64 (y, m, y, y);
+ y = _mm_maskz_rorv_epi64 (m, y, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprorvq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vprorvq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -mavx512bw -O2" } */
+/* { dg-final { scan-assembler "vpsadbw\[ \\t\]+\[^\n\]*%zmm\[0-9\]"} } */
+
+#include <immintrin.h>
+
+volatile __m512i x;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm512_sad_epu8 (x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpshufb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpshufb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpshufd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpshufd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_shuffle_epi32 (x, m, x, _MM_PERM_AADB);
+ x = _mm256_maskz_shuffle_epi32 (m, x, _MM_PERM_AADB);
+ y = _mm_mask_shuffle_epi32 (y, m, y, _MM_PERM_AADB);
+ y = _mm_maskz_shuffle_epi32 (m, y, _MM_PERM_AADB);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpshufd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpshufd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpshufhw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpshufhw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpshuflw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpshuflw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 4 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_sll_epi32 (x256, m256, x256, x128);
+ x256 = _mm256_maskz_sll_epi32 (m256, x256, x128);
+ x128 = _mm_mask_sll_epi32 (x128, m128, x128, x128);
+ x128 = _mm_maskz_sll_epi32 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpslld-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpslld-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpslld\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m256;
+volatile __mmask8 m128;
+#define y 7
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_slli_epi32 (x256, m256, x256, y);
+ x256 = _mm256_maskz_slli_epi32 (m256, x256, y);
+ x128 = _mm_mask_slli_epi32 (x128, m128, x128, y);
+ x128 = _mm_maskz_slli_epi32 (m128, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpslldi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpslldi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 4 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m256;
+volatile __mmask8 m128;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_sll_epi64 (x256, m256, x256, x128);
+ x256 = _mm256_maskz_sll_epi64 (m256, x256, x128);
+ x128 = _mm_mask_sll_epi64 (x128, m128, x128, x128);
+ x128 = _mm_maskz_sll_epi64 (m128, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsllq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsllq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m256;
+volatile __mmask8 m128;
+#define y 7
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_slli_epi64 (x256, m256, x256, y);
+ x256 = _mm256_maskz_slli_epi64 (m256, x256, y);
+ x128 = _mm_mask_slli_epi64 (x128, m128, x128, y);
+ x128 = _mm_maskz_slli_epi64 (m128, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsllqi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsllqi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsllvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256, y256;
+volatile __m128i x128, y128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_sllv_epi32 (x256, m, x256, y256);
+ x256 = _mm256_maskz_sllv_epi32 (m, x256, y256);
+ x128 = _mm_mask_sllv_epi32 (x128, m, x128, y128);
+ x128 = _mm_maskz_sllv_epi32 (m, x128, y128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsllvd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsllvd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsllvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsllvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsllvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256, y256;
+volatile __m128i x128, y128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_sllv_epi64 (x256, m, x256, y256);
+ x256 = _mm256_maskz_sllv_epi64 (m, x256, y256);
+ x128 = _mm_mask_sllv_epi64 (x128, m, x128, y128);
+ x128 = _mm_maskz_sllv_epi64 (m, x128, y128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsllvq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsllvq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsllvw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsllvw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsllw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsllw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsllwi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsllwi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_sra_epi32 (x256, m, x256, y);
+ x256 = _mm256_maskz_sra_epi32 (m, x256, y);
+ x128 = _mm_mask_sra_epi32 (x128, m, x128, y);
+ x128 = _mm_maskz_sra_epi32 (m, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrad-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrad-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrad\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+#define y 7
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_srai_epi32 (x256, m, x256, y);
+ x256 = _mm256_maskz_srai_epi32 (m, x256, y);
+ x128 = _mm_mask_srai_epi32 (x128, m, x128, y);
+ x128 = _mm_maskz_srai_epi32 (m, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsradi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsradi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_sra_epi64 (x256, y);
+ x256 = _mm256_mask_sra_epi64 (x256, m, x256, y);
+ x256 = _mm256_maskz_sra_epi64 (m, x256, y);
+ x128 = _mm_sra_epi64 (x128, y);
+ x128 = _mm_mask_sra_epi64 (x128, m, x128, y);
+ x128 = _mm_maskz_sra_epi64 (m, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsraq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsraq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsraq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+#define y 7
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_srai_epi64 (x256, y);
+ x256 = _mm256_mask_srai_epi64 (x256, m, x256, y);
+ x256 = _mm256_maskz_srai_epi64 (m, x256, y);
+ x128 = _mm_srai_epi64 (x128, y);
+ x128 = _mm_mask_srai_epi64 (x128, m, x128, y);
+ x128 = _mm_maskz_srai_epi64 (m, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsraqi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsraqi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsravd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsravd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsravd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256, y256;
+volatile __m128i x128, y128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_srav_epi32 (x256, m, x256, y256);
+ x256 = _mm256_maskz_srav_epi32 (m, x256, y256);
+ x128 = _mm_mask_srav_epi32 (x128, m, x128, y128);
+ x128 = _mm_maskz_srav_epi32 (m, x128, y128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsravd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsravd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsravq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsravq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsravq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsravq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256, y256;
+volatile __m128i x128, y128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_srav_epi64 (x256, y256);
+ x256 = _mm256_mask_srav_epi64 (x256, m, x256, y256);
+ x256 = _mm256_maskz_srav_epi64 (m, x256, y256);
+ x128 = _mm_srav_epi64 (x128, y128);
+ x128 = _mm_mask_srav_epi64 (x128, m, x128, y128);
+ x128 = _mm_maskz_srav_epi64 (m, x128, y128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsravq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsravq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsravw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsravw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsraw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsraw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsrawi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsrawi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_srl_epi32 (x256, m, x256, y);
+ x256 = _mm256_maskz_srl_epi32 (m, x256, y);
+ x128 = _mm_mask_srl_epi32 (x128, m, x128, y);
+ x128 = _mm_maskz_srl_epi32 (m, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrld-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrld-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrld\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+#define y 7
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_srli_epi32 (x256, m, x256, y);
+ x256 = _mm256_maskz_srli_epi32 (m, x256, y);
+ x128 = _mm_mask_srli_epi32 (x128, m, x128, y);
+ x128 = _mm_maskz_srli_epi32 (m, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrldi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrldi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __m128i y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_srl_epi64 (x256, y);
+ x256 = _mm256_mask_srl_epi64 (x256, m, x256, y);
+ x256 = _mm256_maskz_srl_epi64 (m, x256, y);
+ x128 = _mm_srl_epi64 (x128, y);
+ x128 = _mm_mask_srl_epi64 (x128, m, x128, y);
+ x128 = _mm_maskz_srl_epi64 (m, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrlq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrlq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlq\[ \\t\]+\[^\n\]*, %xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+#define y 7
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_srli_epi64 (x256, y);
+ x256 = _mm256_mask_srli_epi64 (x256, m, x256, y);
+ x256 = _mm256_maskz_srli_epi64 (m, x256, y);
+ x128 = _mm_srli_epi64 (x128, y);
+ x128 = _mm_mask_srli_epi64 (x128, m, x128, y);
+ x128 = _mm_maskz_srli_epi64 (m, x128, y);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrlqi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrlqi-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsrlvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrlvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrlvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256, y256;
+volatile __m128i x128, y128;
+volatile __mmask16 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_srlv_epi32 (x256, m, x256, y256);
+ x256 = _mm256_maskz_srlv_epi32 (m, x256, y256);
+ x128 = _mm_mask_srlv_epi32 (x128, m, x128, y128);
+ x128 = _mm_maskz_srlv_epi32 (m, x128, y128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrlvd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrlvd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsrlvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrlvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsrlvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsrlvq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256, y256;
+volatile __m128i x128, y128;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_srlv_epi64 (x256, m, x256, y256);
+ x256 = _mm256_maskz_srlv_epi64 (m, x256, y256);
+ x128 = _mm_mask_srlv_epi64 (x128, m, x128, y128);
+ x128 = _mm_maskz_srlv_epi64 (m, x128, y128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrlvq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsrlvq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsrlvw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsrlvw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsrlw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsrlw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsrlwi-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsrlwi-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsubd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_sub_epi32 (x256, m8, x256, x256);
+ x256 = _mm256_maskz_sub_epi32 (m8, x256, x256);
+ x128 = _mm_mask_sub_epi32 (x128, m8, x128, x128);
+ x128 = _mm_maskz_sub_epi32 (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsubd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsubd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vpsubq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpsubq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpsubq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x256;
+volatile __m128i x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_mask_sub_epi64 (x256, m8, x256, x256);
+ x256 = _mm256_maskz_sub_epi64 (m8, x256, x256);
+ x128 = _mm_mask_sub_epi64 (x128, m8, x128, x128);
+ x128 = _mm_maskz_sub_epi64 (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsubq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpsubq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubsb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubsb-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubsw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubsw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubusb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubusb-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubusw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubusw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpsubw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpternlogd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpternlogd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpternlogd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpternlogd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpternlogd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpternlogd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y, y2, y3;
+volatile __m128i x, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_ternarylogic_epi32 (y, y2, y3, 0xF0);
+ y = _mm256_mask_ternarylogic_epi32 (y, m, y2, y3, 0xF0);
+ y = _mm256_maskz_ternarylogic_epi32 (m, y, y2, y3, 0xF0);
+
+ x = _mm_ternarylogic_epi32 (x, x2, x3, 0xF0);
+ x = _mm_mask_ternarylogic_epi32 (x, m, x2, x3, 0xF0);
+ x = _mm_maskz_ternarylogic_epi32 (m, x, x2, x3, 0xF0);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpternlogd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpternlogd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpternlogq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpternlogq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vpternlogq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpternlogq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpternlogq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpternlogq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y, y2, y3;
+volatile __m128i x, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_ternarylogic_epi64 (y, y2, y3, 0xF0);
+ y = _mm256_mask_ternarylogic_epi64 (y, m, y2, y3, 0xF0);
+ y = _mm256_maskz_ternarylogic_epi64 (m, y, y2, y3, 0xF0);
+
+ x = _mm_ternarylogic_epi64 (x, x2, x3, 0xF0);
+ x = _mm_mask_ternarylogic_epi64 (x, m, x2, x3, 0xF0);
+ x = _mm_maskz_ternarylogic_epi64 (m, x, x2, x3, 0xF0);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpternlogq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpternlogq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vptestmb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vptestmb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vptestmd\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestmd\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestmd\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%*k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestmd\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%*k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm_test_epi32_mask (x128, x128);
+ m = _mm256_test_epi32_mask (x256, x256);
+ m = _mm_mask_test_epi32_mask (3, x128, x128);
+ m = _mm256_mask_test_epi32_mask (3, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vptestmd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vptestmd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vptestmq\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestmq\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestmq\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestmq\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm_test_epi64_mask (x128, x128);
+ m = _mm256_test_epi64_mask (x256, x256);
+ m = _mm_mask_test_epi64_mask (3, x128, x128);
+ m = _mm256_mask_test_epi64_mask (3, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vptestmq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vptestmq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vptestmw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vptestmw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vptestnmb-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vptestnmb-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vptestnmd\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmd\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmd\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestnmd\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm_testn_epi32_mask (x128, x128);
+ m = _mm256_testn_epi32_mask (x256, x256);
+ m = _mm_mask_testn_epi32_mask (3, x128, x128);
+ m = _mm256_mask_testn_epi32_mask (3, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vptestnmd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vptestnmd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler "vptestnmq\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmq\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\[^\{\]" } } */
+/* { dg-final { scan-assembler "vptestnmq\[ \\t\]+\[^\n\]*%xmm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+/* { dg-final { scan-assembler "vptestnmq\[ \\t\]+\[^\n\]*%ymm\[0-7\]\[^\n\]*k\[1-7\]\{%k\[1-7\]\}" } } */
+
+#include <immintrin.h>
+
+volatile __m128i x128;
+volatile __m256i x256;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ m = _mm_testn_epi64_mask (x128, x128);
+ m = _mm256_testn_epi64_mask (x256, x256);
+ m = _mm_mask_testn_epi64_mask (3, x128, x128);
+ m = _mm256_mask_testn_epi64_mask (3, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vptestnmq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vptestnmq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vptestnmw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vptestnmw-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpunpckhbw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpunpckhbw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpunpckhdq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhdq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x, y, z;
+volatile __m128i a, b, c;
+volatile __mmask8 m;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm256_mask_unpackhi_epi32 (x, m, y, z);
+ x = _mm256_maskz_unpackhi_epi32 (m, y, z);
+ a = _mm_mask_unpackhi_epi32 (a, m, b, c);
+ a = _mm_maskz_unpackhi_epi32 (m, b, c);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpunpckhdq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpunpckhdq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpunpckhqdq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhqdq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhqdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckhqdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x, y, z;
+volatile __m128i a, b, c;
+volatile __mmask8 m;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm256_mask_unpackhi_epi64 (x, m, y, z);
+ x = _mm256_maskz_unpackhi_epi64 (m, y, z);
+ a = _mm_mask_unpackhi_epi64 (a, m, b, c);
+ a = _mm_maskz_unpackhi_epi64 (m, b, c);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpunpckhqdq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpunpckhqdq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpunpckhwd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpunpckhwd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpunpcklbw-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpunpcklbw-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpunpckldq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckldq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckldq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpckldq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x, y, z;
+volatile __m128i a, b, c;
+volatile __mmask8 m;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm256_mask_unpacklo_epi32 (x, m, y, z);
+ x = _mm256_maskz_unpacklo_epi32 (m, y, z);
+ a = _mm_mask_unpacklo_epi32 (a, m, b, c);
+ a = _mm_maskz_unpacklo_epi32 (m, b, c);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpunpckldq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpunpckldq-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpunpcklqdq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklqdq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklqdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpunpcklqdq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x, y, z;
+volatile __m128i a, b, c;
+volatile __mmask8 m;
+
+void extern
+avx512bw_test (void)
+{
+ x = _mm256_mask_unpacklo_epi64 (x, m, y, z);
+ x = _mm256_maskz_unpacklo_epi64 (m, y, z);
+ a = _mm_mask_unpacklo_epi64 (a, m, b, c);
+ a = _mm_maskz_unpacklo_epi64 (m, b, c);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpunpcklqdq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpunpcklqdq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512bw -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpunpcklwd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512bw-vpunpcklwd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpxord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpxord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpxord\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpxord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpxord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpxord\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_xor_epi32 (y, m, y, y);
+ y = _mm256_maskz_xor_epi32 (m, y, y);
+
+ x = _mm_mask_xor_epi32 (x, m, x, x);
+ x = _mm_maskz_xor_epi32 (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpxord-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpxord-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vpxorq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpxorq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpxorq\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpxorq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 2 } } */
+/* { dg-final { scan-assembler-times "vpxorq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vpxorq\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i y;
+volatile __m128i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ y = _mm256_mask_xor_epi64 (y, m, y, y);
+ y = _mm256_maskz_xor_epi64 (m, y, y);
+
+ x = _mm_mask_xor_epi64 (x, m, x, x);
+ x = _mm_maskz_xor_epi64 (m, x, x);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpxorq-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vpxorq-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vrangepd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vrangepd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vrangeps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vrangeps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vrcp14pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrcp14pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrcp14pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrcp14pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrcp14pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrcp14pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_rcp14_pd (x1);
+ x2 = _mm_rcp14_pd (x2);
+
+ x1 = _mm256_mask_rcp14_pd (x1, m, x1);
+ x2 = _mm_mask_rcp14_pd (x2, m, x2);
+
+ x1 = _mm256_maskz_rcp14_pd (m, x1);
+ x2 = _mm_maskz_rcp14_pd (m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrcp14pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrcp14pd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vrcp14ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrcp14ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrcp14ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrcp14ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrcp14ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrcp14ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_rcp14_ps (x1);
+ x2 = _mm_rcp14_ps (x2);
+
+ x1 = _mm256_mask_rcp14_ps (x1, m, x1);
+ x2 = _mm_mask_rcp14_ps (x2, m, x2);
+
+ x1 = _mm256_maskz_rcp14_ps (m, x1);
+ x2 = _mm_maskz_rcp14_ps (m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrcp14ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrcp14ps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vreducepd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vreducepd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vreduceps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vreduceps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vrndscalepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 5 } } */
+/* { dg-final { scan-assembler-times "vrndscalepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalepd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 5 } } */
+/* { dg-final { scan-assembler-times "vrndscalepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscalepd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x1;
+volatile __m128d x2;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_roundscale_pd (x1, 0x42);
+ x1 = _mm256_ceil_pd (x1);
+ x1 = _mm256_floor_pd (x1);
+ x1 = _mm256_mask_roundscale_pd (x1, 2, x1, 0x42);
+ x1 = _mm256_maskz_roundscale_pd (2, x1, 0x42);
+ x2 = _mm_roundscale_pd (x2, 0x42);
+ x2 = _mm_ceil_pd (x2);
+ x2 = _mm_floor_pd (x2);
+ x2 = _mm_mask_roundscale_pd (x2, 2, x2, 0x42);
+ x2 = _mm_maskz_roundscale_pd (2, x2, 0x42);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrndscalepd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrndscalepd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vrndscaleps\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 5 } } */
+/* { dg-final { scan-assembler-times "vrndscaleps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscaleps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscaleps\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 5 } } */
+/* { dg-final { scan-assembler-times "vrndscaleps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrndscaleps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_roundscale_ps (x1, 0x42);
+ x1 = _mm256_ceil_ps (x1);
+ x1 = _mm256_floor_ps (x1);
+ x1 = _mm256_mask_roundscale_ps (x1, 2, x1, 0x42);
+ x1 = _mm256_maskz_roundscale_ps (2, x1, 0x42);
+ x2 = _mm_roundscale_ps (x2, 0x42);
+ x2 = _mm_ceil_ps (x2);
+ x2 = _mm_floor_ps (x2);
+ x2 = _mm_mask_roundscale_ps (x2, 2, x2, 0x42);
+ x2 = _mm_maskz_roundscale_ps (2, x2, 0x42);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrndscaleps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrndscaleps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vrsqrt14pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14pd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14pd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_rsqrt14_pd (x1);
+ x1 = _mm256_mask_rsqrt14_pd (x1, m, x1);
+ x1 = _mm256_maskz_rsqrt14_pd (m, x1);
+
+ x2 = _mm_rsqrt14_pd (x2);
+ x2 = _mm_mask_rsqrt14_pd (x2, m, x2);
+ x2 = _mm_maskz_rsqrt14_pd (m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrsqrt14pd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrsqrt14pd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vrsqrt14ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14ps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vrsqrt14ps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_rsqrt14_ps (x1);
+ x1 = _mm256_mask_rsqrt14_ps (x1, m, x1);
+ x1 = _mm256_maskz_rsqrt14_ps (m, x1);
+
+ x2 = _mm_rsqrt14_ps (x2);
+ x2 = _mm_mask_rsqrt14_ps (x2, m, x2);
+ x2 = _mm_maskz_rsqrt14_ps (m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrsqrt14ps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vrsqrt14ps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vscalefpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vscalefpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vscalefpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vscalefpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vscalefpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vscalefpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x256;
+volatile __m128d x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_scalef_pd (x256, x256);
+ x256 = _mm256_mask_scalef_pd (x256, m8, x256, x256);
+ x256 = _mm256_maskz_scalef_pd (m8, x256, x256);
+ x128 = _mm_scalef_pd (x128, x128);
+ x128 = _mm_mask_scalef_pd (x128, m8, x128, x128);
+ x128 = _mm_maskz_scalef_pd (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vscalefpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vscalefpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vscalefps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vscalefps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vscalefps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vscalefps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\[^\{\]" 3 } } */
+/* { dg-final { scan-assembler-times "vscalefps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vscalefps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x256;
+volatile __m128 x128;
+volatile __mmask8 m8;
+
+void extern
+avx512vl_test (void)
+{
+ x256 = _mm256_scalef_ps (x256, x256);
+ x256 = _mm256_mask_scalef_ps (x256, m8, x256, x256);
+ x256 = _mm256_maskz_scalef_ps (m8, x256, x256);
+ x128 = _mm_scalef_ps (x128, x128);
+ x128 = _mm_mask_scalef_ps (x128, m8, x128, x128);
+ x128 = _mm_maskz_scalef_ps (m8, x128, x128);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vscalefps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vscalefps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vshuff32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vshuff32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vshuff32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_shuffle_f32x4 (x, x, 2);
+ x = _mm256_mask_shuffle_f32x4 (x, m, x, x, 2);
+ x = _mm256_maskz_shuffle_f32x4 (m, x, x, 2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vshuff32x4-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vshuff64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vshuff64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vshuff64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_shuffle_f64x2 (x, x, 2);
+ x = _mm256_mask_shuffle_f64x2 (x, m, x, x, 2);
+ x = _mm256_maskz_shuffle_f64x2 (m, x, x, 2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vshuff64x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vshufi32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vshufi32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vshufi32x4\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_shuffle_i32x4 (x, x, 2);
+ x = _mm256_mask_shuffle_i32x4 (x, m, x, x, 2);
+ x = _mm256_maskz_shuffle_i32x4 (m, x, x, 2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vshufi32x4-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512dq -mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vshufi64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vshufi64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vshufi64x2\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256i x;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_shuffle_i64x2 (x, x, 2);
+ x = _mm256_mask_shuffle_i64x2 (x, m, x, x, 2);
+ x = _mm256_maskz_shuffle_i64x2 (m, x, x, 2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vshufi64x2-2.c"
+
+void
+test_128 () {}
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vshufpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vshufpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vshufpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vshufpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x;
+volatile __m128d y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_shuffle_pd (x, m, x, x, 13);
+ x = _mm256_maskz_shuffle_pd (m, x, x, 13);
+ y = _mm_mask_shuffle_pd (y, m, y, y, 1);
+ y = _mm_maskz_shuffle_pd (m, y, y, 1);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vshufpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vshufpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vshufps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vshufps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vshufps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vshufps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x;
+volatile __m128 y;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_mask_shuffle_ps (x, m, x, x, 64);
+ x = _mm256_maskz_shuffle_ps (m, x, x, 64);
+ y = _mm_mask_shuffle_ps (y, m, y, y, 64);
+ y = _mm_maskz_shuffle_ps (m, y, y, 64);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vshufps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vshufps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vsqrtpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d x1;
+volatile __m128d x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_sqrt_pd (x1, m, x1);
+ x1 = _mm256_maskz_sqrt_pd (m, x1);
+
+ x2 = _mm_mask_sqrt_pd (x2, m, x2);
+ x2 = _mm_maskz_sqrt_pd (m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vsqrtpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vsqrtpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vsqrtps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vsqrtps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x1;
+volatile __m128 x2;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x1 = _mm256_mask_sqrt_ps (x1, m, x1);
+ x1 = _mm256_maskz_sqrt_ps (m, x1);
+
+ x2 = _mm_mask_sqrt_ps (x2, m, x2);
+ x2 = _mm_maskz_sqrt_ps (m, x2);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vsqrtps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vsqrtps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vsubpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vsubpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vsubpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vsubpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128d x128;
+volatile __m256d x256;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x128 = _mm_mask_sub_pd (x128, m, x128, x128);
+ x128 = _mm_maskz_sub_pd (m, x128, x128);
+
+ x256 = _mm256_mask_sub_pd (x256, m, x256, x256);
+ x256 = _mm256_maskz_sub_pd (m, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512f -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vsubpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vsubpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512vl" } */
+/* { dg-final { scan-assembler-times "vsubps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vsubps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vsubps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vsubps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m128 x128;
+volatile __m256 x256;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ x128 = _mm_mask_sub_ps (x128, m, x128, x128);
+ x128 = _mm_maskz_sub_ps (m, x128, x128);
+
+ x256 = _mm256_mask_sub_ps (x256, m, x256, x256);
+ x256 = _mm256_maskz_sub_ps (m, x256, x256);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vsubps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vsubps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vunpckhpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vunpckhpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vunpckhpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vunpckhpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2, y3;
+volatile __m128d xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_unpackhi_pd (yy, m, y2, y3);
+ xx = _mm_mask_unpackhi_pd (xx, m, x2, x3);
+
+ yy = _mm256_maskz_unpackhi_pd (m, y2, y3);
+ xx = _mm_maskz_unpackhi_pd (m, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vunpckhpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vunpckhpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vunpckhps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vunpckhps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vunpckhps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vunpckhps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 yy, y2, y3;
+volatile __m128 xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_unpackhi_ps (yy, m, y2, y3);
+ xx = _mm_mask_unpackhi_ps (xx, m, x2, x3);
+
+ yy = _mm256_maskz_unpackhi_ps (m, y2, y3);
+ xx = _mm_maskz_unpackhi_ps (m, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vunpckhps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vunpckhps-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vunpcklpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vunpcklpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vunpcklpd\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vunpcklpd\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256d yy, y2, y3;
+volatile __m128d xx, x2, x3;
+volatile __mmask8 m;
+
+void extern
+avx512vl_test (void)
+{
+ yy = _mm256_mask_unpacklo_pd (yy, m, y2, y3);
+ xx = _mm_mask_unpacklo_pd (xx, m, x2, x3);
+
+ yy = _mm256_maskz_unpacklo_pd (m, y2, y3);
+ xx = _mm_maskz_unpacklo_pd (m, x2, x3);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vunpcklpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vunpcklpd-2.c"
--- /dev/null
+/* { dg-do compile } */
+/* { dg-options "-mavx512vl -O2" } */
+/* { dg-final { scan-assembler-times "vunpcklps\[ \\t\]+\[^\n\]*%ymm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vunpcklps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vunpcklps\[ \\t\]+\[^\n\]*%ymm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+/* { dg-final { scan-assembler-times "vunpcklps\[ \\t\]+\[^\n\]*%xmm\[0-9\]" 3 } } */
+/* { dg-final { scan-assembler-times "vunpcklps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\{z\}" 1 } } */
+/* { dg-final { scan-assembler-times "vunpcklps\[ \\t\]+\[^\n\]*%xmm\[0-9\]\{%k\[1-7\]\}\[^\{\]" 1 } } */
+
+#include <immintrin.h>
+
+volatile __m256 x, y, z;
+volatile __m128 xx, yy, zz;
+
+void extern
+avx512vl_test (void)
+{
+ x = _mm256_unpacklo_ps (y, z);
+ x = _mm256_mask_unpacklo_ps (x, 2, y, z);
+ x = _mm256_maskz_unpacklo_ps (2, y, z);
+ xx = _mm_unpacklo_ps (yy, zz);
+ xx = _mm_mask_unpacklo_ps (xx, 2, yy, zz);
+ xx = _mm_maskz_unpacklo_ps (2, yy, zz);
+}
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vunpcklps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512f-vunpcklps-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vxorpd-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vxorpd-2.c"
--- /dev/null
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512dq -mavx512vl -DAVX512VL" } */
+/* { dg-require-effective-target avx512vl } */
+
+#define AVX512F_LEN 256
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vxorps-2.c"
+
+#undef AVX512F_LEN
+#undef AVX512F_LEN_HALF
+
+#define AVX512F_LEN 128
+#define AVX512F_LEN_HALF 128
+#include "avx512dq-vxorps-2.c"
} "-mavx512f" ]
}
+# Return 1 if avx512vl instructions can be compiled.
+proc check_effective_target_avx512vl { } {
+ return [check_no_compiler_messages avx512vl object {
+ typedef long long __v4di __attribute__ ((__vector_size__ (32)));
+ __v4di
+ mm256_and_epi64 (__v4di __X, __v4di __Y)
+ {
+ __v4di __W;
+ return __builtin_ia32_pandq256_mask (__X, __Y, __W, -1);
+ }
+ } "-mavx512vl" ]
+}
+
# Return 1 if avx512cd instructions can be compiled.
proc check_effective_target_avx512cd { } {
return [check_no_compiler_messages avx512cd_trans object {
_mm512_conflict_epi64 (__v8di __W, __v8di __A)
{
return (__v8di) __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,
- (__v8di) __W,
- -1);
+ (__v8di) __W,
+ -1);
}
} "-Wno-psabi -mavx512cd" ]
}
} "-O2 -msha" ]
}
+# Return 1 if avx512dq instructions can be compiled.
+proc check_effective_target_avx512dq { } {
+ return [check_no_compiler_messages avx512dq object {
+ typedef long long __v8di __attribute__ ((__vector_size__ (64)));
+ __v8di
+ _mm512_mask_mullo_epi64 (__v8di __W, __v8di __A, __v8di __B)
+ {
+ return (__v8di) __builtin_ia32_pmullq512_mask ((__v8di) __A,
+ (__v8di) __B,
+ (__v8di) __W,
+ -1);
+ }
+ } "-mavx512dq" ]
+}
+
+# Return 1 if avx512bw instructions can be compiled.
+proc check_effective_target_avx512bw { } {
+ return [check_no_compiler_messages avx512bw object {
+ typedef short __v32hi __attribute__ ((__vector_size__ (64)));
+ __v32hi
+ _mm512_mask_mulhrs_epi16 (__v32hi __W, __v32hi __A, __v32hi __B)
+ {
+ return (__v32hi) __builtin_ia32_pmulhrsw512_mask ((__v32hi) __A,
+ (__v32hi) __B,
+ (__v32hi) __W,
+ -1);
+ }
+ } "-mavx512bw" ]
+}
+
# If a testcase doesn't have special options, use these.
global DEFAULT_CFLAGS
if ![info exists DEFAULT_CFLAGS] then {
CHECK_ROUGH_EXP (union512, float, "%f")
CHECK_ROUGH_EXP (union512d, double, "%f")
+CHECK_ROUGH_EXP (union256, float, "%f")
+CHECK_ROUGH_EXP (union256d, double, "%f")
+CHECK_ROUGH_EXP (union128, float, "%f")
+CHECK_ROUGH_EXP (union128d, double, "%f")
projects / gcc.git / commitdiff