re PR target/90991 (_mm_loadu_ps instrinsic translates to vmovaps in combination...
authorJakub Jelinek <jakub@redhat.com>
Wed, 26 Jun 2019 08:26:18 +0000 (10:26 +0200)
committerJakub Jelinek <jakub@gcc.gnu.org>
Wed, 26 Jun 2019 08:26:18 +0000 (10:26 +0200)
commitd55c1ffd497193b56ee9f7540d5b0eadf6717c42
tree33f8ef728b08279bcf026ded097617d64610277f
parentfdfbed383e48f9a6fce6ef1e0e0defea0363ac3f
re PR target/90991 (_mm_loadu_ps instrinsic translates to vmovaps in combination with _mm512_insertf32x4)

PR target/90991
* config/i386/sse.md
(*<extract_type>_vinsert<shuffletype><extract_suf>_0): Use vmovupd,
vmovups, vmovdqu, vmovdqu32 or vmovdqu64 instead of the aligned
insns if operands[2] is misaligned_operand.

* gcc.target/i386/avx512dq-pr90991-1.c: New test.

From-SVN: r272674
gcc/ChangeLog
gcc/config/i386/sse.md
gcc/testsuite/ChangeLog
gcc/testsuite/gcc.target/i386/avx512dq-pr90991-1.c [new file with mode: 0644]