From: Matthew Malcomson Date: Thu, 7 Nov 2019 16:38:59 +0000 (+0000) Subject: [binutils][aarch64] Bfloat16 enablement [2/X] X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=df6780137dc9d832db8c5525ecc760acdbcde5c9;p=binutils-gdb.git [binutils][aarch64] Bfloat16 enablement [2/X] Hi, This patch is part of a series that adds support for Armv8.6-A (Matrix Multiply and BFloat16 extensions) to binutils. This patch introduces the following BFloat16 instructions to the aarch64 backend: bfdot, bfmmla, bfcvt, bfcvtnt, bfmlal[t/b], bfcvtn2. Committed on behalf of Mihail Ionescu. gas/ChangeLog: 2019-11-07 Mihail Ionescu 2019-11-07 Matthew Malcomson * config/tc-aarch64.c (vectype_to_qualifier): Special case the S_2H operand qualifier. * doc/c-aarch64.texi: Document bf16 and bf16mmla4 extensions. * testsuite/gas/aarch64/bfloat16.d: New test. * testsuite/gas/aarch64/bfloat16.s: New test. * testsuite/gas/aarch64/illegal-bfloat16.d: New test. * testsuite/gas/aarch64/illegal-bfloat16.l: New test. * testsuite/gas/aarch64/illegal-bfloat16.s: New test. * testsuite/gas/aarch64/sve-bfloat-movprfx.s: New test. * testsuite/gas/aarch64/sve-bfloat-movprfx.d: New test. include/ChangeLog: 2019-11-07 Mihail Ionescu 2019-11-07 Matthew Malcomson * opcode/aarch64.h (AARCH64_FEATURE_BFLOAT16): New feature macros. (AARCH64_ARCH_V8_6): Include BFloat16 feature macros. (enum aarch64_opnd_qualifier): Introduce new operand qualifier AARCH64_OPND_QLF_S_2H. (enum aarch64_insn_class): Introduce new class "bfloat16". (BFLOAT16_SVE_INSNC): New feature set for bfloat16 instructions to support the movprfx constraint. opcodes/ChangeLog: 2019-11-07 Mihail Ionescu 2019-11-07 Matthew Malcomson * aarch64-asm.c (aarch64_ins_reglane): Use AARCH64_OPND_QLF_S_2H in reglane special case. * aarch64-dis-2.c (aarch64_opcode_lookup_1, aarch64_find_next_opcode): Account for new instructions. * aarch64-dis.c (aarch64_ext_reglane): Use AARCH64_OPND_QLF_S_2H in reglane special case. * aarch64-opc.c (struct operand_qualifier_data): Add data for new AARCH64_OPND_QLF_S_2H qualifier. * aarch64-tbl.h (QL_BFDOT QL_BFDOT64, QL_BFDOT64I, QL_BFMMLA2, QL_BFCVT64, QL_BFCVTN64, QL_BFCVTN2_64): New qualifiers. (aarch64_feature_bfloat16, aarch64_feature_bfloat16_sve, aarch64_feature_bfloat16_bfmmla4): New feature sets. (BFLOAT_SVE, BFLOAT): New feature set macros. (BFLOAT_SVE_INSN, BFLOAT_BFMMLA4_INSN, BFLOAT_INSN): New macros to define BFloat16 instructions. (aarch64_opcode_table): Define new instructions bfdot, bfmmla, bfcvt, bfcvtnt, bfdot, bfdot, bfcvtn, bfmlal[b/t] bfcvtn2, bfcvt. Regression tested on aarch64-elf. Is it ok for trunk? Regards, Mihail --- diff --git a/gas/ChangeLog b/gas/ChangeLog index 40481d26b4f..05bf4e4621c 100644 --- a/gas/ChangeLog +++ b/gas/ChangeLog @@ -1,3 +1,17 @@ +2019-11-07 Mihail Ionescu +2019-11-07 Matthew Malcomson + + * config/tc-aarch64.c (vectype_to_qualifier): Special case the + S_2H operand qualifier. + * doc/c-aarch64.texi: Document bf16 extension. + * testsuite/gas/aarch64/bfloat16.d: New test. + * testsuite/gas/aarch64/bfloat16.s: New test. + * testsuite/gas/aarch64/illegal-bfloat16.d: New test. + * testsuite/gas/aarch64/illegal-bfloat16.l: New test. + * testsuite/gas/aarch64/illegal-bfloat16.s: New test. + * testsuite/gas/aarch64/sve-bfloat-movprfx.s: New test. + * testsuite/gas/aarch64/sve-bfloat-movprfx.d: New test. + 2019-11-07 Mihail Ionescu 2019-11-07 Matthew Malcomson diff --git a/gas/config/tc-aarch64.c b/gas/config/tc-aarch64.c index fb1ec0bcc31..527d8590ca7 100644 --- a/gas/config/tc-aarch64.c +++ b/gas/config/tc-aarch64.c @@ -5130,6 +5130,10 @@ vectype_to_qualifier (const struct vector_type_el *vectype) if (vectype->type == NT_b && vectype->width == 4) return AARCH64_OPND_QLF_S_4B; + /* Special case S_2H. */ + if (vectype->type == NT_h && vectype->width == 2) + return AARCH64_OPND_QLF_S_2H; + /* Vector element register. */ return AARCH64_OPND_QLF_S_B + vectype->type; } @@ -9003,6 +9007,8 @@ static const struct aarch64_option_cpu_value_table aarch64_features[] = { | AARCH64_FEATURE_SHA3, 0)}, {"sve2-bitperm", AARCH64_FEATURE (AARCH64_FEATURE_SVE2_BITPERM, 0), AARCH64_FEATURE (AARCH64_FEATURE_SVE2, 0)}, + {"bf16", AARCH64_FEATURE (AARCH64_FEATURE_BFLOAT16, 0), + AARCH64_ARCH_NONE}, {NULL, AARCH64_ARCH_NONE, AARCH64_ARCH_NONE}, }; diff --git a/gas/doc/c-aarch64.texi b/gas/doc/c-aarch64.texi index a83a859e4d6..4cb94872daa 100644 --- a/gas/doc/c-aarch64.texi +++ b/gas/doc/c-aarch64.texi @@ -144,6 +144,8 @@ automatically cause those extensions to be disabled. @multitable @columnfractions .12 .17 .17 .54 @headitem Extension @tab Minimum Architecture @tab Enabled by default @tab Description +@item @code{bf16} @tab ARMv8.2-A @tab ARMv8.6-A or later + @tab Enable BFloat16 extension. @item @code{compnum} @tab ARMv8.2-A @tab ARMv8.3-A or later @tab Enable the complex number SIMD extensions. This implies @code{fp16} and @code{simd}. diff --git a/gas/testsuite/gas/aarch64/bfloat16.d b/gas/testsuite/gas/aarch64/bfloat16.d new file mode 100644 index 00000000000..78d10cb3cad --- /dev/null +++ b/gas/testsuite/gas/aarch64/bfloat16.d @@ -0,0 +1,56 @@ +#as: -march=armv8.6-a+bf16+sve +#objdump: -dr + +.*: file format .* + + +Disassembly of section \.text: + +0000000000000000 <\.text>: + *[0-9a-f]+: 647b82b1 bfdot z17\.s, z21\.h, z27\.h + *[0-9a-f]+: 64608000 bfdot z0\.s, z0\.h, z0\.h + *[0-9a-f]+: 647d42b1 bfdot z17\.s, z21\.h, z5\.h\[3\] + *[0-9a-f]+: 64784000 bfdot z0\.s, z0\.h, z0\.h\[3\] + *[0-9a-f]+: 64604000 bfdot z0\.s, z0\.h, z0\.h\[0\] + *[0-9a-f]+: 647be6b1 bfmmla z17\.s, z21\.h, z27\.h + *[0-9a-f]+: 6460e400 bfmmla z0\.s, z0\.h, z0\.h + *[0-9a-f]+: 658ab6b1 bfcvt z17\.h, p5/m, z21\.s + *[0-9a-f]+: 658aa000 bfcvt z0\.h, p0/m, z0\.s + *[0-9a-f]+: 648ab6b1 bfcvtnt z17\.h, p5/m, z21\.s + *[0-9a-f]+: 648aa000 bfcvtnt z0\.h, p0/m, z0\.s + *[0-9a-f]+: 64fb86b1 bfmlalt z17\.s, z21\.h, z27\.h + *[0-9a-f]+: 64e08400 bfmlalt z0\.s, z0\.h, z0\.h + *[0-9a-f]+: 64fb82b1 bfmlalb z17\.s, z21\.h, z27\.h + *[0-9a-f]+: 64e08000 bfmlalb z0\.s, z0\.h, z0\.h + *[0-9a-f]+: 64e546b1 bfmlalt z17\.s, z21\.h, z5\.h\[0\] + *[0-9a-f]+: 64f84c00 bfmlalt z0\.s, z0\.h, z0\.h\[7\] + *[0-9a-f]+: 64e542b1 bfmlalb z17\.s, z21\.h, z5\.h\[0\] + *[0-9a-f]+: 64f84800 bfmlalb z0\.s, z0\.h, z0\.h\[7\] + *[0-9a-f]+: 2e5bfeb1 bfdot v17\.2s, v21\.4h, v27\.4h + *[0-9a-f]+: 2e40fc00 bfdot v0\.2s, v0\.4h, v0\.4h + *[0-9a-f]+: 6e5bfeb1 bfdot v17\.4s, v21\.8h, v27\.8h + *[0-9a-f]+: 6e40fc00 bfdot v0\.4s, v0\.8h, v0\.8h + *[0-9a-f]+: 0f7bfab1 bfdot v17\.2s, v21\.4h, v27\.2h\[3\] + *[0-9a-f]+: 0f60f800 bfdot v0\.2s, v0\.4h, v0\.2h\[3\] + *[0-9a-f]+: 4f7bfab1 bfdot v17\.4s, v21\.8h, v27\.2h\[3\] + *[0-9a-f]+: 4f60f800 bfdot v0\.4s, v0\.8h, v0\.2h\[3\] + *[0-9a-f]+: 0f5bf2b1 bfdot v17\.2s, v21\.4h, v27\.2h\[0\] + *[0-9a-f]+: 0f40f000 bfdot v0\.2s, v0\.4h, v0\.2h\[0\] + *[0-9a-f]+: 4f5bf2b1 bfdot v17\.4s, v21\.8h, v27\.2h\[0\] + *[0-9a-f]+: 4f40f000 bfdot v0\.4s, v0\.8h, v0\.2h\[0\] + *[0-9a-f]+: 6e5beeb1 bfmmla v17\.4s, v21\.8h, v27\.8h + *[0-9a-f]+: 6e40ec00 bfmmla v0\.4s, v0\.8h, v0\.8h + *[0-9a-f]+: 2edbfeb1 bfmlalb v17\.4s, v21\.8h, v27\.8h + *[0-9a-f]+: 2ec0fc00 bfmlalb v0\.4s, v0\.8h, v0\.8h + *[0-9a-f]+: 6edbfeb1 bfmlalt v17\.4s, v21\.8h, v27\.8h + *[0-9a-f]+: 6ec0fc00 bfmlalt v0\.4s, v0\.8h, v0\.8h + *[0-9a-f]+: 0fcff2b1 bfmlalb v17\.4s, v21\.8h, v15\.h\[0\] + *[0-9a-f]+: 0ff0f800 bfmlalb v0\.4s, v0\.8h, v0\.h\[7\] + *[0-9a-f]+: 4fcff2b1 bfmlalt v17\.4s, v21\.8h, v15\.h\[0\] + *[0-9a-f]+: 4ff0f800 bfmlalt v0\.4s, v0\.8h, v0\.h\[7\] + *[0-9a-f]+: 0ea16ab1 bfcvtn v17\.4h, v21\.4s + *[0-9a-f]+: 0ea16800 bfcvtn v0\.4h, v0\.4s + *[0-9a-f]+: 4ea16ab1 bfcvtn2 v17\.8h, v21\.4s + *[0-9a-f]+: 4ea16800 bfcvtn2 v0\.8h, v0\.4s + *[0-9a-f]+: 1e6342b1 bfcvt h17, s21 + *[0-9a-f]+: 1e634000 bfcvt h0, s0 diff --git a/gas/testsuite/gas/aarch64/bfloat16.s b/gas/testsuite/gas/aarch64/bfloat16.s new file mode 100644 index 00000000000..5850a9c7d18 --- /dev/null +++ b/gas/testsuite/gas/aarch64/bfloat16.s @@ -0,0 +1,70 @@ +/* The instructions with non-zero register numbers are there to ensure we have + the correct argument positioning (i.e. check that the first argument is at + the end of the word etc). + The instructions with all-zero register numbers are to ensure the previous + encoding didn't just "happen" to fit -- so that if we change the registers + that changes the correct part of the word. + Each of the numbered patterns begin and end with a 1, so we can replace + them with all-zeros and see the entire range has changed. */ + +// SVE +bfdot z17.s, z21.h, z27.h +bfdot z0.s, z0.h, z0.h + +bfdot z17.s, z21.h, z5.h[3] +bfdot z0.s, z0.h, z0.h[3] +bfdot z0.s, z0.h, z0.h[0] + +bfmmla z17.s, z21.h, z27.h +bfmmla z0.s, z0.h, z0.h + +bfcvt z17.h, p5/m, z21.s +bfcvt z0.h, p0/m, z0.s +bfcvtnt z17.h, p5/m, z21.s +bfcvtnt z0.h, p0/m, z0.s + +bfmlalt z17.s, z21.h, z27.h +bfmlalt z0.s, z0.h, z0.h +bfmlalb z17.s, z21.h, z27.h +bfmlalb z0.s, z0.h, z0.h + +bfmlalt z17.s, z21.h, z5.h[0] +bfmlalt z0.s, z0.h, z0.h[7] +bfmlalb z17.s, z21.h, z5.h[0] +bfmlalb z0.s, z0.h, z0.h[7] + +// SIMD +bfdot v17.2s, v21.4h, v27.4h +bfdot v0.2s, v0.4h, v0.4h +bfdot v17.4s, v21.8h, v27.8h +bfdot v0.4s, v0.8h, v0.8h + +bfdot v17.2s, v21.4h, v27.2h[3] +bfdot v0.2s, v0.4h, v0.2h[3] +bfdot v17.4s, v21.8h, v27.2h[3] +bfdot v0.4s, v0.8h, v0.2h[3] +bfdot v17.2s, v21.4h, v27.2h[0] +bfdot v0.2s, v0.4h, v0.2h[0] +bfdot v17.4s, v21.8h, v27.2h[0] +bfdot v0.4s, v0.8h, v0.2h[0] + +bfmmla v17.4s, v21.8h, v27.8h +bfmmla v0.4s, v0.8h, v0.8h + +bfmlalb v17.4s, v21.8h, v27.8h +bfmlalb v0.4s, v0.8h, v0.8h +bfmlalt v17.4s, v21.8h, v27.8h +bfmlalt v0.4s, v0.8h, v0.8h + +bfmlalb v17.4s, v21.8h, v15.h[0] +bfmlalb v0.4s, v0.8h, v0.h[7] +bfmlalt v17.4s, v21.8h, v15.h[0] +bfmlalt v0.4s, v0.8h, v0.h[7] + +bfcvtn v17.4h, v21.4s +bfcvtn v0.4h, v0.4s +bfcvtn2 v17.8h, v21.4s +bfcvtn2 v0.8h, v0.4s + +bfcvt h17, s21 +bfcvt h0, s0 diff --git a/gas/testsuite/gas/aarch64/illegal-bfloat16.d b/gas/testsuite/gas/aarch64/illegal-bfloat16.d new file mode 100644 index 00000000000..275ac33af8f --- /dev/null +++ b/gas/testsuite/gas/aarch64/illegal-bfloat16.d @@ -0,0 +1,4 @@ +#name: Illegal Bfloat16 instructions +#as: -march=armv8.6-a+bf16+sve +#source: illegal-bfloat16.s +#error_output: illegal-bfloat16.l diff --git a/gas/testsuite/gas/aarch64/illegal-bfloat16.l b/gas/testsuite/gas/aarch64/illegal-bfloat16.l new file mode 100644 index 00000000000..c20f132de38 --- /dev/null +++ b/gas/testsuite/gas/aarch64/illegal-bfloat16.l @@ -0,0 +1,95 @@ +[^ :]+: Assembler messages: +[^ :]+:[0-9]+: Error: operand mismatch -- `bfdot z0\.s,z1\.h,z2\.s' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfdot z0\.s, z1\.h, z2\.h +[^ :]+:[0-9]+: Error: operand mismatch -- `bfdot z0\.s,z1\.h,z3\.s\[3\]' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfdot z0\.s, z1\.h, z3\.h\[3\] +[^ :]+:[0-9]+: Error: register element index out of range 0 to 3 at operand 3 -- `bfdot z0\.s,z1\.h,z3\.h\[4\]' +[^ :]+:[0-9]+: Error: z0-z7 expected at operand 3 -- `bfdot z0\.s,z1\.h,z8\.h\[3\]' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmmla z0\.s,z1\.h,z2\.s' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmmla z0\.s, z1\.h, z2\.h +[^ :]+:[0-9]+: Error: operand mismatch -- `bfcvt z0\.h,p1/z,z2\.s' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfcvt z0\.h, p1/m, z2\.s +[^ :]+:[0-9]+: Error: operand mismatch -- `bfcvt z0\.h,p1/m,z2\.h' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfcvt z0\.h, p1/m, z2\.s +[^ :]+:[0-9]+: Error: operand mismatch -- `bfcvtnt z0\.h,p1/z,z2\.s' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfcvtnt z0\.h, p1/m, z2\.s +[^ :]+:[0-9]+: Error: operand mismatch -- `bfcvtnt z0\.h,p1/m,z2\.h' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfcvtnt z0\.h, p1/m, z2\.s +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalt z0\.s,z0\.h,z0\.s' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalt z0\.s, z0\.h, z0\.h +[^ :]+:[0-9]+: Error: operand 1 must be an SVE vector register -- `bfmlalt z32\.s,z0\.h,z0\.h' +[^ :]+:[0-9]+: Error: operand 2 must be an SVE vector register -- `bfmlalt z0\.s,z32\.h,z0\.h' +[^ :]+:[0-9]+: Error: operand 3 must be an indexed SVE vector register -- `bfmlalt z0\.s,z0\.h,z32\.h' +[^ :]+:[0-9]+: Error: register element index out of range 0 to 7 at operand 3 -- `bfmlalt z0\.s,z0\.h,z0\.h\[8\]' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalt z0\.s,z0\.h,z0\.s\[0\]' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalt z0\.s, z0\.h, z0\.h\[0\] +[^ :]+:[0-9]+: Error: operand 1 must be an SVE vector register -- `bfmlalt z32\.s,z0\.h,z0\.h\[0\]' +[^ :]+:[0-9]+: Error: operand 2 must be an SVE vector register -- `bfmlalt z0\.s,z32\.h,z0\.h\[0\]' +[^ :]+:[0-9]+: Error: z0-z7 expected at operand 3 -- `bfmlalt z0\.s,z0\.h,z8\.h\[0\]' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalb z0\.s,z0\.h,z0\.s' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalb z0\.s, z0\.h, z0\.h +[^ :]+:[0-9]+: Error: operand 1 must be an SVE vector register -- `bfmlalb z32\.s,z0\.h,z0\.h' +[^ :]+:[0-9]+: Error: operand 2 must be an SVE vector register -- `bfmlalb z0\.s,z32\.h,z0\.h' +[^ :]+:[0-9]+: Error: operand 3 must be an indexed SVE vector register -- `bfmlalb z0\.s,z0\.h,z32\.h' +[^ :]+:[0-9]+: Error: register element index out of range 0 to 7 at operand 3 -- `bfmlalb z0\.s,z0\.h,z0\.h\[8\]' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalb z0\.s,z0\.h,z0\.s\[0\]' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalb z0\.s, z0\.h, z0\.h\[0\] +[^ :]+:[0-9]+: Error: operand 1 must be an SVE vector register -- `bfmlalb z32\.s,z0\.h,z0\.h\[0\]' +[^ :]+:[0-9]+: Error: operand 2 must be an SVE vector register -- `bfmlalb z0\.s,z32\.h,z0\.h\[0\]' +[^ :]+:[0-9]+: Error: z0-z7 expected at operand 3 -- `bfmlalb z0\.s,z0\.h,z8\.h\[0\]' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfdot v0\.2s,v1\.4h,v2\.2s\[3\]' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfdot v0\.2s, v1\.4h, v2\.2h\[3\] +[^ :]+:[0-9]+: Info: other valid variant\(s\): +[^ :]+:[0-9]+: Info: bfdot v0\.4s, v1\.8h, v2\.2h\[3\] +[^ :]+:[0-9]+: Error: register element index out of range 0 to 3 at operand 3 -- `bfdot v0\.4s,v1\.8h,v2\.2h\[4\]' +[^ :]+:[0-9]+: Error: invalid element size 8 and vector size combination s at operand 3 -- `bfmmla v0\.4s,v1\.8h,v2\.8s' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmmla v0\.4s,v1\.4h,v2\.8h' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmmla v0\.4s, v1\.8h, v2\.8h +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalb v0\.4s,v0\.4h,v0\.8h' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalb v0\.4s, v0\.8h, v0\.8h +[^ :]+:[0-9]+: Error: operand 1 must be an SVE vector register -- `bfmlalb v32\.4s,v0\.8h,v0\.8h' +[^ :]+:[0-9]+: Error: operand 2 must be a SIMD vector register -- `bfmlalb v0\.4s,v32\.8h,v0\.8h' +[^ :]+:[0-9]+: Error: operand 3 must be a SIMD vector register -- `bfmlalb v0\.4s,v0\.8h,v32\.8h' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalt v0\.4s,v0\.8h,v0\.4h' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalt v0\.4s, v0\.8h, v0\.8h +[^ :]+:[0-9]+: Error: operand 1 must be an SVE vector register -- `bfmlalt v32\.4s,v0\.8h,v0\.8h' +[^ :]+:[0-9]+: Error: operand 2 must be a SIMD vector register -- `bfmlalt v0\.4s,v32\.8h,v0\.8h' +[^ :]+:[0-9]+: Error: operand 3 must be a SIMD vector register -- `bfmlalt v0\.4s,v0\.8h,v32\.8h' +[^ :]+:[0-9]+: Error: register element index out of range 0 to 7 at operand 3 -- `bfmlalb v0\.4s,v0\.8h,v0\.h\[8\]' +[^ :]+:[0-9]+: Error: operand 1 must be an SVE vector register -- `bfmlalb v32\.4s,v0\.8h,v0\.h\[0\]' +[^ :]+:[0-9]+: Error: operand 2 must be a SIMD vector register -- `bfmlalb v0\.4s,v32\.8h,v0\.h\[0\]' +[^ :]+:[0-9]+: Error: register number out of range 0 to 15 at operand 3 -- `bfmlalb v0\.4s,v0\.8h,v16\.h\[0\]' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalb v0\.4s,v0\.4h,v0\.h\[0\]' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalb v0\.4s, v0\.8h, v0\.h\[0\] +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalb v0\.4s,v0\.8h,v0\.s\[0\]' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalb v0\.4s, v0\.8h, v0\.h\[0\] +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalt v0\.4s,v0\.8h,v0\.s\[0\]' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalt v0\.4s, v0\.8h, v0\.h\[0\] +[^ :]+:[0-9]+: Error: operand mismatch -- `bfmlalt v0\.4s,v0\.4h,v0\.h\[0\]' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfmlalt v0\.4s, v0\.8h, v0\.h\[0\] +[^ :]+:[0-9]+: Error: register element index out of range 0 to 7 at operand 3 -- `bfmlalt v0\.4s,v0\.8h,v0\.h\[8\]' +[^ :]+:[0-9]+: Error: operand 1 must be an SVE vector register -- `bfmlalt v32\.4s,v0\.8h,v0\.h\[0\]' +[^ :]+:[0-9]+: Error: operand 2 must be a SIMD vector register -- `bfmlalt v0\.4s,v32\.8h,v0\.h\[0\]' +[^ :]+:[0-9]+: Error: register number out of range 0 to 15 at operand 3 -- `bfmlalt v0\.4s,v0\.8h,v16\.h\[0\]' +[^ :]+:[0-9]+: Error: operand mismatch -- `bfcvt h0,h1' +[^ :]+:[0-9]+: Info: did you mean this\? +[^ :]+:[0-9]+: Info: bfcvt h0, s1 diff --git a/gas/testsuite/gas/aarch64/illegal-bfloat16.s b/gas/testsuite/gas/aarch64/illegal-bfloat16.s new file mode 100644 index 00000000000..6b5333435ee --- /dev/null +++ b/gas/testsuite/gas/aarch64/illegal-bfloat16.s @@ -0,0 +1,67 @@ +// SVE +bfdot z0.s, z1.h, z2.s // Fails from size types + +bfdot z0.s, z1.h, z3.s[3] // Fails from size types +bfdot z0.s, z1.h, z3.h[4] // Fails from index size +bfdot z0.s, z1.h, z8.h[3] // Fails from vector number + +bfmmla z0.s, z1.h, z2.s // Fails from size types + +bfcvt z0.h, p1/z, z2.s // Fails from merge type +bfcvt z0.h, p1/m, z2.h // Fails from size type + +bfcvtnt z0.h, p1/z, z2.s // Fails from merge type +bfcvtnt z0.h, p1/m, z2.h // Fails from size type + +bfmlalt z0.s, z0.h, z0.s // Fails from size type +bfmlalt z32.s, z0.h, z0.h +bfmlalt z0.s, z32.h, z0.h +bfmlalt z0.s, z0.h, z32.h + +bfmlalt z0.s, z0.h, z0.h[8] // Fails from index size +bfmlalt z0.s, z0.h, z0.s[0] // Fails from size type +bfmlalt z32.s, z0.h, z0.h[0] +bfmlalt z0.s, z32.h, z0.h[0] +bfmlalt z0.s, z0.h, z8.h[0] // Fails from vector index + +bfmlalb z0.s, z0.h, z0.s // Fails from size type +bfmlalb z32.s, z0.h, z0.h +bfmlalb z0.s, z32.h, z0.h +bfmlalb z0.s, z0.h, z32.h + +bfmlalb z0.s, z0.h, z0.h[8] // Fails from index size +bfmlalb z0.s, z0.h, z0.s[0] // Fails from size type +bfmlalb z32.s, z0.h, z0.h[0] +bfmlalb z0.s, z32.h, z0.h[0] +bfmlalb z0.s, z0.h, z8.h[0] // Fails from vector index + +// SIMD +bfdot v0.2s, v1.4h, v2.2s[3] // Fails from size types +bfdot v0.4s, v1.8h, v2.2h[4] // Fails from index size + +bfmmla v0.4s, v1.8h, v2.8s // Fails from size types +bfmmla v0.4s, v1.4h, v2.8h // Fails from size types + +bfmlalb v0.4s, v0.4h, v0.8h +bfmlalb v32.4s, v0.8h, v0.8h +bfmlalb v0.4s, v32.8h, v0.8h +bfmlalb v0.4s, v0.8h, v32.8h +bfmlalt v0.4s, v0.8h, v0.4h +bfmlalt v32.4s, v0.8h, v0.8h +bfmlalt v0.4s, v32.8h, v0.8h +bfmlalt v0.4s, v0.8h, v32.8h + +bfmlalb v0.4s, v0.8h, v0.h[8] +bfmlalb v32.4s, v0.8h, v0.h[0] +bfmlalb v0.4s, v32.8h, v0.h[0] +bfmlalb v0.4s, v0.8h, v16.h[0] +bfmlalb v0.4s, v0.4h, v0.h[0] +bfmlalb v0.4s, v0.8h, v0.s[0] +bfmlalt v0.4s, v0.8h, v0.s[0] +bfmlalt v0.4s, v0.4h, v0.h[0] +bfmlalt v0.4s, v0.8h, v0.h[8] +bfmlalt v32.4s, v0.8h, v0.h[0] +bfmlalt v0.4s, v32.8h, v0.h[0] +bfmlalt v0.4s, v0.8h, v16.h[0] + +bfcvt h0, h1 // Fails from size types diff --git a/gas/testsuite/gas/aarch64/sve-bfloat-movprfx.d b/gas/testsuite/gas/aarch64/sve-bfloat-movprfx.d new file mode 100644 index 00000000000..a7971586f74 --- /dev/null +++ b/gas/testsuite/gas/aarch64/sve-bfloat-movprfx.d @@ -0,0 +1,27 @@ +#as: -march=armv8.6-a+bf16+sve +#objdump: -dr + +.* file format .* + + +Disassembly of section \.text: + +0000000000000000 <\.text>: + *[0-9a-f]+: 0420bc20 movprfx z0, z1 + *[0-9a-f]+: 64638040 bfdot z0\.s, z2\.h, z3\.h + *[0-9a-f]+: 0420bc20 movprfx z0, z1 + *[0-9a-f]+: 64634040 bfdot z0\.s, z2\.h, z3\.h\[0\] + *[0-9a-f]+: 0420bc20 movprfx z0, z1 + *[0-9a-f]+: 6463e440 bfmmla z0\.s, z2\.h, z3\.h + *[0-9a-f]+: 0420bc20 movprfx z0, z1 + *[0-9a-f]+: 64e38040 bfmlalb z0\.s, z2\.h, z3\.h + *[0-9a-f]+: 0420bc20 movprfx z0, z1 + *[0-9a-f]+: 64e38440 bfmlalt z0\.s, z2\.h, z3\.h + *[0-9a-f]+: 0420bc20 movprfx z0, z1 + *[0-9a-f]+: 64e34040 bfmlalb z0\.s, z2\.h, z3\.h\[0\] + *[0-9a-f]+: 0420bc20 movprfx z0, z1 + *[0-9a-f]+: 64e34440 bfmlalt z0\.s, z2\.h, z3\.h\[0\] + *[0-9a-f]+: 0420bc20 movprfx z0, z1 + *[0-9a-f]+: 658aa040 bfcvt z0\.h, p0/m, z2\.s + *[0-9a-f]+: 04512020 movprfx z0\.h, p0/m, z1\.h + *[0-9a-f]+: 658aa040 bfcvt z0\.h, p0/m, z2\.s diff --git a/gas/testsuite/gas/aarch64/sve-bfloat-movprfx.s b/gas/testsuite/gas/aarch64/sve-bfloat-movprfx.s new file mode 100644 index 00000000000..c322532a807 --- /dev/null +++ b/gas/testsuite/gas/aarch64/sve-bfloat-movprfx.s @@ -0,0 +1,31 @@ +.text +.arch armv8.2-a+bf16+sve + +movprfx z0, z1 +bfdot z0.s, z2.h, z3.h + +movprfx z0, z1 +bfdot z0.s, z2.h, z3.h[0] + +movprfx z0, z1 +bfmmla z0.s, z2.h, z3.h + +movprfx z0, z1 +bfmlalb z0.s, z2.h, z3.h + +movprfx z0, z1 +bfmlalt z0.s, z2.h, z3.h + +movprfx z0, z1 +bfmlalb z0.s, z2.h, z3.h[0] + +movprfx z0, z1 +bfmlalt z0.s, z2.h, z3.h[0] + +# Unpredicated movprfx + bfcvt +movprfx z0, z1 +bfcvt z0.h, p0/m, z2.s + +# Predicated movprfx + bfcvt +movprfx z0.h, p0/m, z1.h +bfcvt z0.h, p0/m, z2.s diff --git a/include/ChangeLog b/include/ChangeLog index 246dc49031b..d7c891d9ce1 100644 --- a/include/ChangeLog +++ b/include/ChangeLog @@ -1,3 +1,14 @@ +2019-11-07 Mihail Ionescu +2019-11-07 Matthew Malcomson + + * opcode/aarch64.h (AARCH64_FEATURE_BFLOAT16): New feature macros. + (AARCH64_ARCH_V8_6): Include BFloat16 feature macros. + (enum aarch64_opnd_qualifier): Introduce new operand qualifier + AARCH64_OPND_QLF_S_2H. + (enum aarch64_insn_class): Introduce new class "bfloat16". + (BFLOAT16_SVE_INSNC): New feature set for bfloat16 + instructions to support the movprfx constraint. + 2019-11-07 Mihail Ionescu 2019-11-07 Matthew Malcomson diff --git a/include/opcode/aarch64.h b/include/opcode/aarch64.h index 493e8f8655b..4cda1e079e0 100644 --- a/include/opcode/aarch64.h +++ b/include/opcode/aarch64.h @@ -64,6 +64,7 @@ typedef uint32_t aarch64_insn; #define AARCH64_FEATURE_F16_FML 0x1000000000ULL /* v8.2 FP16FML ins. */ #define AARCH64_FEATURE_V8_5 0x2000000000ULL /* ARMv8.5 processors. */ #define AARCH64_FEATURE_V8_6 0x00000002 /* ARMv8.6 processors. */ +#define AARCH64_FEATURE_BFLOAT16 0x00000004 /* Bfloat16 insns. */ /* Flag Manipulation insns. */ #define AARCH64_FEATURE_FLAGMANIP 0x4000000000ULL @@ -131,7 +132,8 @@ typedef uint32_t aarch64_insn; | AARCH64_FEATURE_ID_PFR2 \ | AARCH64_FEATURE_SSBS) #define AARCH64_ARCH_V8_6 AARCH64_FEATURE (AARCH64_ARCH_V8_5, \ - AARCH64_FEATURE_V8_6) + AARCH64_FEATURE_V8_6 \ + | AARCH64_FEATURE_BFLOAT16) #define AARCH64_ARCH_NONE AARCH64_FEATURE (0, 0) #define AARCH64_ANY AARCH64_FEATURE (-1, 0) /* Any basic core. */ @@ -462,11 +464,13 @@ enum aarch64_opnd_qualifier AARCH64_OPND_QLF_S_S, AARCH64_OPND_QLF_S_D, AARCH64_OPND_QLF_S_Q, - /* This type qualifier has a special meaning in that it means that 4 x 1 byte - are selected by the instruction. Other than that it has no difference - with AARCH64_OPND_QLF_S_B in encoding. It is here purely for syntactical - reasons and is an exception from normal AArch64 disassembly scheme. */ + /* These type qualifiers have a special meaning in that they mean 4 x 1 byte + or 2 x 2 byte are selected by the instruction. Other than that they have + no difference with AARCH64_OPND_QLF_S_B in encoding. They are here purely + for syntactical reasons and is an exception from normal AArch64 + disassembly scheme. */ AARCH64_OPND_QLF_S_4B, + AARCH64_OPND_QLF_S_2H, /* Qualifying an operand which is a SIMD vector register or a SIMD vector register list; indicating register shape. @@ -609,6 +613,7 @@ enum aarch64_insn_class cryptosm3, cryptosm4, dotproduct, + bfloat16, }; /* Opcode enumerators. */ diff --git a/opcodes/ChangeLog b/opcodes/ChangeLog index e357230e31c..ef6ecf5e0bb 100644 --- a/opcodes/ChangeLog +++ b/opcodes/ChangeLog @@ -1,3 +1,25 @@ +2019-11-07 Mihail Ionescu +2019-11-07 Matthew Malcomson + + * aarch64-asm.c (aarch64_ins_reglane): Use AARCH64_OPND_QLF_S_2H + in reglane special case. + * aarch64-dis-2.c (aarch64_opcode_lookup_1, + aarch64_find_next_opcode): Account for new instructions. + * aarch64-dis.c (aarch64_ext_reglane): Use AARCH64_OPND_QLF_S_2H + in reglane special case. + * aarch64-opc.c (struct operand_qualifier_data): Add data for + new AARCH64_OPND_QLF_S_2H qualifier. + * aarch64-tbl.h (QL_BFDOT QL_BFDOT64, QL_BFDOT64I, QL_BFMMLA2, + QL_BFCVT64, QL_BFCVTN64, QL_BFCVTN2_64): New qualifiers. + (aarch64_feature_bfloat16, aarch64_feature_bfloat16_sve): New feature + sets. + (BFLOAT_SVE, BFLOAT): New feature set macros. + (BFLOAT_SVE_INSN, BFLOAT_INSN): New macros to define BFloat16 + instructions. + (aarch64_opcode_table): Define new instructions bfdot, + bfmmla, bfcvt, bfcvtnt, bfdot, bfdot, bfcvtn, bfmlal[b/t] + bfcvtn2, bfcvt. + 2019-11-07 Mihail Ionescu 2019-11-07 Matthew Malcomson diff --git a/opcodes/aarch64-asm.c b/opcodes/aarch64-asm.c index 67ebad687cc..3dcd84a5cb7 100644 --- a/opcodes/aarch64-asm.c +++ b/opcodes/aarch64-asm.c @@ -130,6 +130,7 @@ aarch64_ins_reglane (const aarch64_operand *self, const aarch64_opnd_info *info, switch (info->qualifier) { case AARCH64_OPND_QLF_S_4B: + case AARCH64_OPND_QLF_S_2H: /* L:H */ assert (reglane_index < 4); insert_fields (code, reglane_index, 0, 2, FLD_L, FLD_H); diff --git a/opcodes/aarch64-dis-2.c b/opcodes/aarch64-dis-2.c index 17edc76dcd0..cd1f6b91a93 100644 --- a/opcodes/aarch64-dis-2.c +++ b/opcodes/aarch64-dis-2.c @@ -8660,11 +8660,22 @@ aarch64_opcode_lookup_1 (uint32_t word) { if (((word >> 16) & 0x1) == 0) { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - 011001x0100xxxx0101xxxxxxxxxxxxx - fcvtnt. */ - return 2068; + if (((word >> 17) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x0100xxx00101xxxxxxxxxxxxx + fcvtnt. */ + return 2068; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x0100xxx10101xxxxxxxxxxxxx + bfcvtnt. */ + return 2394; + } } else { @@ -9118,19 +9129,52 @@ aarch64_opcode_lookup_1 (uint32_t word) { if (((word >> 23) & 0x1) == 0) { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - x11001x0011xxxxx010xxxxxxxxxxxxx - st1b. */ - return 1868; + if (((word >> 31) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x0011xxxxx010xxxxxxxxxxxxx + bfdot. */ + return 2391; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 111001x0011xxxxx010xxxxxxxxxxxxx + st1b. */ + return 1868; + } } else { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - x11001x0111xxxxx010xxxxxxxxxxxxx - st1h. */ - return 1889; + if (((word >> 31) & 0x1) == 0) + { + if (((word >> 10) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x0111xxxxx010xx0xxxxxxxxxx + bfmlalb. */ + return 2398; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x0111xxxxx010xx1xxxxxxxxxx + bfmlalt. */ + return 2397; + } + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 111001x0111xxxxx010xxxxxxxxxxxxx + st1h. */ + return 1889; + } } } } @@ -9169,11 +9213,44 @@ aarch64_opcode_lookup_1 (uint32_t word) } else { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - x11001x0x11xxxxx1x0xxxxxxxxxxxxx - st1h. */ - return 1890; + if (((word >> 23) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + x11001x0011xxxxx1x0xxxxxxxxxxxxx + bfdot. */ + return 2390; + } + else + { + if (((word >> 31) & 0x1) == 0) + { + if (((word >> 10) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x0111xxxxx1x0xx0xxxxxxxxxx + bfmlalb. */ + return 2396; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x0111xxxxx1x0xx1xxxxxxxxxx + bfmlalt. */ + return 2395; + } + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 111001x0111xxxxx1x0xxxxxxxxxxxxx + st1h. */ + return 1890; + } + } } } } @@ -9529,9 +9606,9 @@ aarch64_opcode_lookup_1 (uint32_t word) } else { - if (((word >> 20) & 0x1) == 0) + if (((word >> 22) & 0x1) == 0) { - if (((word >> 22) & 0x1) == 0) + if (((word >> 20) & 0x1) == 0) { if (((word >> 23) & 0x1) == 0) { @@ -9556,50 +9633,61 @@ aarch64_opcode_lookup_1 (uint32_t word) { /* 33222222222211111111110000000000 10987654321098765432109876543210 - x11001x00110xxxx111xxxxxxxxxxxxx - st1b. */ - return 1874; + x11001x00011xxxx111xxxxxxxxxxxxx + st2b. */ + return 1909; } else { /* 33222222222211111111110000000000 10987654321098765432109876543210 - x11001x01110xxxx111xxxxxxxxxxxxx - st1h. */ - return 1895; + x11001x01011xxxx111xxxxxxxxxxxxx + st2h. */ + return 1913; } } } else { - if (((word >> 22) & 0x1) == 0) + if (((word >> 23) & 0x1) == 0) { - if (((word >> 23) & 0x1) == 0) + if (((word >> 31) & 0x1) == 0) { /* 33222222222211111111110000000000 10987654321098765432109876543210 - x11001x00011xxxx111xxxxxxxxxxxxx - st2b. */ - return 1909; + 011001x0011xxxxx111xxxxxxxxxxxxx + bfmmla. */ + return 2392; } else { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - x11001x01011xxxx111xxxxxxxxxxxxx - st2h. */ - return 1913; + if (((word >> 20) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 111001x00110xxxx111xxxxxxxxxxxxx + st1b. */ + return 1874; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 111001x00111xxxx111xxxxxxxxxxxxx + st4b. */ + return 1925; + } } } else { - if (((word >> 23) & 0x1) == 0) + if (((word >> 20) & 0x1) == 0) { /* 33222222222211111111110000000000 10987654321098765432109876543210 - x11001x00111xxxx111xxxxxxxxxxxxx - st4b. */ - return 1925; + x11001x01110xxxx111xxxxxxxxxxxxx + st1h. */ + return 1895; } else { @@ -13993,11 +14081,22 @@ aarch64_opcode_lookup_1 (uint32_t word) { if (((word >> 22) & 0x1) == 0) { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - 011001x1x0001x10101xxxxxxxxxxxxx - fcvtx. */ - return 2070; + if (((word >> 23) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x100001x10101xxxxxxxxxxxxx + fcvtx. */ + return 2070; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 011001x110001x10101xxxxxxxxxxxxx + bfcvt. */ + return 2393; + } } else { @@ -16503,11 +16602,55 @@ aarch64_opcode_lookup_1 (uint32_t word) } else { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - xx101110xx0xxxxx1x1xx1xxxxxxxxxx - fcadd. */ - return 373; + if (((word >> 11) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + xx101110xx0xxxxx1x1x01xxxxxxxxxx + fcadd. */ + return 373; + } + else + { + if (((word >> 12) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + xx101110xx0xxxxx1x1011xxxxxxxxxx + bfmmla. */ + return 2401; + } + else + { + if (((word >> 23) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + xx1011100x0xxxxx1x1111xxxxxxxxxx + bfdot. */ + return 2399; + } + else + { + if (((word >> 30) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + x01011101x0xxxxx1x1111xxxxxxxxxx + bfmlalb. */ + return 2406; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + x11011101x0xxxxx1x1111xxxxxxxxxx + bfmlalt. */ + return 2405; + } + } + } + } } } } @@ -17060,21 +17203,43 @@ aarch64_opcode_lookup_1 (uint32_t word) } else { - if (((word >> 30) & 0x1) == 0) + if (((word >> 23) & 0x1) == 0) { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - 00001110xx1xxxx1011010xxxxxxxxxx - fcvtn. */ - return 178; + if (((word >> 30) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 000011100x1xxxx1011010xxxxxxxxxx + fcvtn. */ + return 178; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 010011100x1xxxx1011010xxxxxxxxxx + fcvtn2. */ + return 179; + } } else { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - 01001110xx1xxxx1011010xxxxxxxxxx - fcvtn2. */ - return 179; + if (((word >> 30) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 000011101x1xxxx1011010xxxxxxxxxx + bfcvtn. */ + return 2402; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + 010011101x1xxxx1011010xxxxxxxxxx + bfcvtn2. */ + return 2403; + } } } } @@ -22165,11 +22330,44 @@ aarch64_opcode_lookup_1 (uint32_t word) } else { - /* 33222222222211111111110000000000 - 10987654321098765432109876543210 - xxx01111xxxxxxxx1111x0xxxxxxxxxx - sqrdmlsh. */ - return 130; + if (((word >> 29) & 0x1) == 0) + { + if (((word >> 23) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + xx0011110xxxxxxx1111x0xxxxxxxxxx + bfdot. */ + return 2400; + } + else + { + if (((word >> 30) & 0x1) == 0) + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + x00011111xxxxxxx1111x0xxxxxxxxxx + bfmlalb. */ + return 2408; + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + x10011111xxxxxxx1111x0xxxxxxxxxx + bfmlalt. */ + return 2407; + } + } + } + else + { + /* 33222222222211111111110000000000 + 10987654321098765432109876543210 + xx101111xxxxxxxx1111x0xxxxxxxxxx + sqrdmlsh. */ + return 130; + } } } } @@ -22787,6 +22985,8 @@ aarch64_find_next_opcode (const aarch64_opcode *opcode) case 823: return NULL; /* fsqrt --> NULL. */ case 831: value = 832; break; /* frintz --> frintz. */ case 832: return NULL; /* frintz --> NULL. */ + case 824: value = 2404; break; /* fcvt --> bfcvt. */ + case 2404: return NULL; /* bfcvt --> NULL. */ case 833: value = 834; break; /* frinta --> frinta. */ case 834: return NULL; /* frinta --> NULL. */ case 835: value = 836; break; /* frintx --> frintx. */ diff --git a/opcodes/aarch64-dis.c b/opcodes/aarch64-dis.c index 7ae844a601c..93005698138 100644 --- a/opcodes/aarch64-dis.c +++ b/opcodes/aarch64-dis.c @@ -348,6 +348,7 @@ aarch64_ext_reglane (const aarch64_operand *self, aarch64_opnd_info *info, switch (info->qualifier) { case AARCH64_OPND_QLF_S_4B: + case AARCH64_OPND_QLF_S_2H: /* L:H */ info->reglane.index = extract_fields (code, 0, 2, FLD_H, FLD_L); info->reglane.regno &= 0x1f; diff --git a/opcodes/aarch64-opc.c b/opcodes/aarch64-opc.c index 992a2af1b3a..456bac02326 100644 --- a/opcodes/aarch64-opc.c +++ b/opcodes/aarch64-opc.c @@ -712,6 +712,7 @@ struct operand_qualifier_data aarch64_opnd_qualifiers[] = {8, 1, 0x3, "d", OQK_OPD_VARIANT}, {16, 1, 0x4, "q", OQK_OPD_VARIANT}, {4, 1, 0x0, "4b", OQK_OPD_VARIANT}, + {4, 1, 0x0, "2h", OQK_OPD_VARIANT}, {1, 4, 0x0, "4b", OQK_OPD_VARIANT}, {1, 8, 0x0, "8b", OQK_OPD_VARIANT}, diff --git a/opcodes/aarch64-tbl.h b/opcodes/aarch64-tbl.h index cdebac3f10a..3425336d2fc 100644 --- a/opcodes/aarch64-tbl.h +++ b/opcodes/aarch64-tbl.h @@ -2257,6 +2257,50 @@ { \ QLF2(X, NIL), \ } + +/* e.g. BFDOT ., ., . */ +#define QL_BFDOT64 \ +{ \ + QLF3(V_2S, V_4H, V_4H),\ + QLF3(V_4S, V_8H, V_8H),\ +} + +/* e.g. BFDOT ., ., .2H[] */ +#define QL_BFDOT64I \ +{ \ + QLF3(V_2S, V_4H, S_2H),\ + QLF3(V_4S, V_8H, S_2H),\ +} + +/* e.g. BFMMLA .4s, .8h, .8h */ +#define QL_BFMMLA \ +{ \ + QLF3(V_4S, V_8H, V_8H),\ +} + +/* e.g. BFCVT , */ +#define QL_BFCVT64 \ +{ \ + QLF2(S_H,S_S), \ +} + +/* e.g. BFCVT , */ +#define QL_BFCVTN64 \ +{ \ + QLF2(V_4H,V_4S), \ +} + +/* e.g. BFCVT , */ +#define QL_BFCVTN2_64 \ +{ \ + QLF2(V_8H,V_4S), \ +} + +/* e.g. BFMLAL2 .4s, .8h, .H[] */ +#define QL_V3BFML4S \ +{ \ + QLF3(V_4S, V_8H, S_H), \ +} /* Opcode table. */ @@ -2331,6 +2375,10 @@ static const aarch64_feature_set aarch64_feature_bti = AARCH64_FEATURE (AARCH64_FEATURE_BTI, 0); static const aarch64_feature_set aarch64_feature_memtag = AARCH64_FEATURE (AARCH64_FEATURE_V8_5 | AARCH64_FEATURE_MEMTAG, 0); +static const aarch64_feature_set aarch64_feature_bfloat16 = + AARCH64_FEATURE (AARCH64_FEATURE_BFLOAT16, 0); +static const aarch64_feature_set aarch64_feature_bfloat16_sve = + AARCH64_FEATURE (AARCH64_FEATURE_BFLOAT16 | AARCH64_FEATURE_SVE, 0); static const aarch64_feature_set aarch64_feature_tme = AARCH64_FEATURE (AARCH64_FEATURE_TME, 0); static const aarch64_feature_set aarch64_feature_sve2 = @@ -2387,6 +2435,8 @@ static const aarch64_feature_set aarch64_feature_v8_6 = #define SVE2_SM4 &aarch64_feature_sve2sm4 #define SVE2_BITPERM &aarch64_feature_sve2bitperm #define ARMV8_6 &aarch64_feature_v8_6 +#define BFLOAT16_SVE &aarch64_feature_bfloat16_sve +#define BFLOAT16 &aarch64_feature_bfloat16 #define CORE_INSN(NAME,OPCODE,MASK,CLASS,OP,OPS,QUALS,FLAGS) \ { NAME, OPCODE, MASK, CLASS, OP, CORE, OPS, QUALS, FLAGS, 0, 0, NULL } @@ -2477,6 +2527,13 @@ static const aarch64_feature_set aarch64_feature_v8_6 = #define SVE2BITPERM_INSN(NAME,OPCODE,MASK,CLASS,OP,OPS,QUALS,FLAGS,TIED) \ { NAME, OPCODE, MASK, CLASS, OP, SVE2_BITPERM, OPS, QUALS, \ FLAGS | F_STRICT, 0, TIED, NULL } +#define BFLOAT16_SVE_INSN(NAME,OPCODE,MASK,CLASS,OPS,QUALS,FLAGS) \ + { NAME, OPCODE, MASK, CLASS, 0, BFLOAT16_SVE, OPS, QUALS, FLAGS, 0, 0, NULL } +#define BFLOAT16_SVE_INSNC(NAME,OPCODE,MASK,CLASS,OPS,QUALS,FLAGS, CONSTRAINTS, TIED) \ + { NAME, OPCODE, MASK, CLASS, 0, BFLOAT16_SVE, OPS, QUALS, FLAGS | F_STRICT, \ + CONSTRAINTS, TIED, NULL } +#define BFLOAT16_INSN(NAME,OPCODE,MASK,CLASS,OPS,QUALS,FLAGS) \ + { NAME, OPCODE, MASK, CLASS, 0, BFLOAT16, OPS, QUALS, FLAGS, 0, 0, NULL } struct aarch64_opcode aarch64_opcode_table[] = { @@ -4974,6 +5031,29 @@ struct aarch64_opcode aarch64_opcode_table[] = V8_4_INSN ("ldapursw", 0x99800000, 0xffe00c00, ldst_unscaled, OP2 (Rt, ADDR_OFFSET), QL_STLX, 0), V8_4_INSN ("stlur", 0xd9000000, 0xffe00c00, ldst_unscaled, OP2 (Rt, ADDR_OFFSET), QL_STLX, 0), V8_4_INSN ("ldapur", 0xd9400000, 0xffe00c00, ldst_unscaled, OP2 (Rt, ADDR_OFFSET), QL_STLX, 0), + + /* BFloat instructions. */ + BFLOAT16_SVE_INSNC ("bfdot", 0x64608000, 0xffe0fc00, sve_misc, OP3 (SVE_Zd, SVE_Zn, SVE_Zm_16), OP_SVE_SHH, 0, C_SCAN_MOVPRFX, 0), + BFLOAT16_SVE_INSNC ("bfdot", 0x64604000, 0xffe0fc00, sve_misc, OP3 (SVE_Zd, SVE_Zn, SVE_Zm3_INDEX), OP_SVE_SHH, 0, C_SCAN_MOVPRFX, 0), + BFLOAT16_SVE_INSNC ("bfmmla", 0x6460e400, 0xffe0fc00, sve_misc, OP3 (SVE_Zd, SVE_Zn, SVE_Zm_16), OP_SVE_SHH, 0, C_SCAN_MOVPRFX, 0), + BFLOAT16_SVE_INSNC ("bfcvt", 0x658aa000, 0xffffe000, sve_misc, OP3 (SVE_Zd, SVE_Pg3, SVE_Zn), OP_SVE_HMS, 0, C_SCAN_MOVPRFX, 0), + BFLOAT16_SVE_INSNC ("bfcvtnt", 0x648aa000, 0xffffe000, sve_misc, OP3 (SVE_Zd, SVE_Pg3, SVE_Zn), OP_SVE_HMS, 0, C_SCAN_MOVPRFX, 0), + BFLOAT16_SVE_INSNC ("bfmlalt", 0x64e08400, 0xffe0fc00, sve_misc, OP3 (SVE_Zd, SVE_Zn, SVE_Zm_16), OP_SVE_SHH, 0, C_SCAN_MOVPRFX, 0), + BFLOAT16_SVE_INSNC ("bfmlalb", 0x64e08000, 0xffe0fc00, sve_misc, OP3 (SVE_Zd, SVE_Zn, SVE_Zm_16), OP_SVE_SHH, 0, C_SCAN_MOVPRFX, 0), + BFLOAT16_SVE_INSNC ("bfmlalt", 0x64e04400, 0xffe0f400, sve_misc, OP3 (SVE_Zd, SVE_Zn, SVE_Zm3_11_INDEX), OP_SVE_SHH, 0, C_SCAN_MOVPRFX, 0), + BFLOAT16_SVE_INSNC ("bfmlalb", 0x64e04000, 0xffe0f400, sve_misc, OP3 (SVE_Zd, SVE_Zn, SVE_Zm3_11_INDEX), OP_SVE_SHH, 0, C_SCAN_MOVPRFX, 0), + /* BFloat Advanced SIMD instructions. */ + BFLOAT16_INSN ("bfdot", 0x2e40fc00, 0xbfe0fc00, bfloat16, OP3 (Vd, Vn, Vm), QL_BFDOT64, F_SIZEQ), + /* Using dotproduct as iclass to treat instruction similar to udot. */ + BFLOAT16_INSN ("bfdot", 0x0f40f000, 0xbfc0f400, dotproduct, OP3 (Vd, Vn, Em), QL_BFDOT64I, F_SIZEQ), + BFLOAT16_INSN ("bfmmla", 0x6e40ec00, 0xffe0fc00, bfloat16, OP3 (Vd, Vn, Vm), QL_BFMMLA, F_SIZEQ), + BFLOAT16_INSN ("bfcvtn", 0x0ea16800, 0xfffffc00, bfloat16, OP2 (Vd, Vn), QL_BFCVTN64, 0), + BFLOAT16_INSN ("bfcvtn2", 0x4ea16800, 0xfffffc00, bfloat16, OP2 (Vd, Vn), QL_BFCVTN2_64, 0), + BFLOAT16_INSN ("bfcvt", 0x1e634000, 0xfffffc00, bfloat16, OP2 (Fd, Fn), QL_BFCVT64, 0), + BFLOAT16_INSN ("bfmlalt", 0x6ec0fc00, 0xffe0fc00, bfloat16, OP3 (Vd, Vn, Vm), QL_BFMMLA, 0), + BFLOAT16_INSN ("bfmlalb", 0x2ec0fc00, 0xffe0fc00, bfloat16, OP3 (Vd, Vn, Vm), QL_BFMMLA, 0), + BFLOAT16_INSN ("bfmlalt", 0x4fc0f000, 0xffc0f400, bfloat16, OP3 (Vd, Vn, Em16), QL_V3BFML4S, 0), + BFLOAT16_INSN ("bfmlalb", 0x0fc0f000, 0xffc0f400, bfloat16, OP3 (Vd, Vn, Em16), QL_V3BFML4S, 0), {0, 0, 0, 0, 0, 0, {}, {}, 0, 0, 0, NULL}, };