set_feature (FEATURE_AMX_AVX512);
if (eax & bit_AMX_TF32)
set_feature (FEATURE_AMX_TF32);
- if (eax & bit_AMX_TRANSPOSE)
- set_feature (FEATURE_AMX_TRANSPOSE);
if (eax & bit_AMX_FP8)
set_feature (FEATURE_AMX_FP8);
if (eax & bit_AMX_MOVRS)
(OPTION_MASK_ISA2_AMX_TILE_SET | OPTION_MASK_ISA2_AMX_AVX512)
#define OPTION_MASK_ISA2_AMX_TF32_SET \
(OPTION_MASK_ISA2_AMX_TILE_SET | OPTION_MASK_ISA2_AMX_TF32)
-#define OPTION_MASK_ISA2_AMX_TRANSPOSE_SET \
- (OPTION_MASK_ISA2_AMX_TILE_SET | OPTION_MASK_ISA2_AMX_TRANSPOSE)
#define OPTION_MASK_ISA2_AMX_FP8_SET \
(OPTION_MASK_ISA2_AMX_TILE_SET | OPTION_MASK_ISA2_AMX_FP8)
#define OPTION_MASK_ISA2_MOVRS_SET OPTION_MASK_ISA2_MOVRS
(OPTION_MASK_ISA2_AMX_TILE | OPTION_MASK_ISA2_AMX_INT8_UNSET \
| OPTION_MASK_ISA2_AMX_BF16_UNSET | OPTION_MASK_ISA2_AMX_FP16_UNSET \
| OPTION_MASK_ISA2_AMX_COMPLEX_UNSET | OPTION_MASK_ISA2_AMX_AVX512_UNSET \
- | OPTION_MASK_ISA2_AMX_TF32_UNSET | OPTION_MASK_ISA2_AMX_TRANSPOSE_UNSET \
- | OPTION_MASK_ISA2_AMX_FP8_UNSET | OPTION_MASK_ISA2_AMX_MOVRS_UNSET)
+ | OPTION_MASK_ISA2_AMX_TF32_UNSET | OPTION_MASK_ISA2_AMX_FP8_UNSET \
+ | OPTION_MASK_ISA2_AMX_MOVRS_UNSET)
#define OPTION_MASK_ISA2_AMX_INT8_UNSET OPTION_MASK_ISA2_AMX_INT8
#define OPTION_MASK_ISA2_AMX_BF16_UNSET OPTION_MASK_ISA2_AMX_BF16
#define OPTION_MASK_ISA2_UINTR_UNSET OPTION_MASK_ISA2_UINTR
#define OPTION_MASK_ISA2_AVX10_2_UNSET OPTION_MASK_ISA2_AVX10_2
#define OPTION_MASK_ISA2_AMX_AVX512_UNSET OPTION_MASK_ISA2_AMX_AVX512
#define OPTION_MASK_ISA2_AMX_TF32_UNSET OPTION_MASK_ISA2_AMX_TF32
-#define OPTION_MASK_ISA2_AMX_TRANSPOSE_UNSET OPTION_MASK_ISA2_AMX_TRANSPOSE
#define OPTION_MASK_ISA2_AMX_FP8_UNSET OPTION_MASK_ISA2_AMX_FP8
#define OPTION_MASK_ISA2_MOVRS_UNSET OPTION_MASK_ISA2_MOVRS
#define OPTION_MASK_ISA2_AMX_MOVRS_UNSET OPTION_MASK_ISA2_AMX_MOVRS
}
return true;
- case OPT_mamx_transpose:
- if (value)
- {
- opts->x_ix86_isa_flags2 |= OPTION_MASK_ISA2_AMX_TRANSPOSE_SET;
- opts->x_ix86_isa_flags2_explicit |= OPTION_MASK_ISA2_AMX_TRANSPOSE_SET;
- }
- else
- {
- opts->x_ix86_isa_flags2 &= ~OPTION_MASK_ISA2_AMX_TRANSPOSE_UNSET;
- opts->x_ix86_isa_flags2_explicit |=
- OPTION_MASK_ISA2_AMX_TRANSPOSE_UNSET;
- }
- return true;
-
case OPT_mamx_fp8:
if (value)
{
FEATURE_AVX10_2 = 116,
FEATURE_AMX_AVX512,
FEATURE_AMX_TF32,
- FEATURE_AMX_TRANSPOSE,
- FEATURE_AMX_FP8,
+ FEATURE_AMX_FP8 = 120,
FEATURE_MOVRS,
FEATURE_AMX_MOVRS,
CPU_FEATURE_MAX
ISA_NAMES_TABLE_ENTRY("amx-avx512", FEATURE_AMX_AVX512, P_NONE,
"-mamx-avx512")
ISA_NAMES_TABLE_ENTRY("amx-tf32", FEATURE_AMX_TF32, P_NONE, "-mamx-tf32")
- ISA_NAMES_TABLE_ENTRY("amx-transpose", FEATURE_AMX_TRANSPOSE,
- P_NONE, "-mamx-transpose")
ISA_NAMES_TABLE_ENTRY("amx-fp8", FEATURE_AMX_FP8, P_NONE, "-mamx-fp8")
ISA_NAMES_TABLE_ENTRY("movrs", FEATURE_MOVRS, P_NONE, "-mmovrs")
ISA_NAMES_TABLE_ENTRY("amx-movrs", FEATURE_AMX_MOVRS, P_NONE, "-mamx-movrs")
avx10_2mediaintrin.h avx10_2convertintrin.h
avx10_2bf16intrin.h avx10_2satcvtintrin.h
avx10_2minmaxintrin.h avx10_2copyintrin.h
- amxavx512intrin.h amxtf32intrin.h amxtransposeintrin.h
- amxfp8intrin.h movrsintrin.h amxmovrsintrin.h"
+ amxavx512intrin.h amxtf32intrin.h amxfp8intrin.h
+ movrsintrin.h amxmovrsintrin.h"
;;
ia64-*-*)
extra_headers=ia64intrin.h
#pragma GCC pop_options
#endif /* __DISABLE_AMX_MOVRS__ */
-#if !defined(__AMX_MOVRS__) || !defined (__AMX_TRANSPOSE__)
-#pragma GCC push_options
-#pragma GCC target("amx-movrs,amx-transpose")
-#define __DISABLE_AMX_MOVRS_TRANSPOSE__
-#endif /* __AMX_MOVRS_TRANSPOSE__ */
-
-#define _tile_2rpntlvwz0rs_internal(tdst, base, stride) \
- __asm__ volatile \
- ("{t2rpntlvwz0rs\t(%0,%1,1), %%tmm"#tdst \
- "|t2rpntlvwz0rs\t%%tmm"#tdst", [%0+%1*1]}" \
- :: "r" ((const void*) (base)), "r" ((long) (stride)))
-
-#define _tile_2rpntlvwz0rst1_internal(tdst, base, stride) \
- __asm__ volatile \
- ("{t2rpntlvwz0rst1\t(%0,%1,1), %%tmm"#tdst \
- "|t2rpntlvwz0rst1\t%%tmm"#tdst", [%0+%1*1]}" \
- :: "r" ((const void*) (base)), "r" ((long) (stride)))
-
-#define _tile_2rpntlvwz1rs_internal(tdst, base, stride) \
- __asm__ volatile \
- ("{t2rpntlvwz1rs\t(%0,%1,1), %%tmm"#tdst \
- "|t2rpntlvwz1rs\t%%tmm"#tdst", [%0+%1*1]}" \
- :: "r" ((const void*) (base)), "r" ((long) (stride)))
-
-#define _tile_2rpntlvwz1rst1_internal(tdst, base, stride) \
- __asm__ volatile \
- ("{t2rpntlvwz1rst1\t(%0,%1,1), %%tmm"#tdst \
- "|t2rpntlvwz1rst1\t%%tmm"#tdst", [%0+%1*1]}" \
- :: "r" ((const void*) (base)), "r" ((long) (stride)))
-
-#define _tile_2rpntlvwz0rs(tdst, base, stride) \
- _tile_2rpntlvwz0rs_internal(tdst, base, stride)
-
-#define _tile_2rpntlvwz0rst1(tdst, base, stride) \
- _tile_2rpntlvwz0rst1_internal(tdst, base, stride)
-
-#define _tile_2rpntlvwz1rs(tdst, base, stride) \
- _tile_2rpntlvwz1rs_internal(tdst, base, stride)
-
-#define _tile_2rpntlvwz1rst1(tdst, base, stride) \
- _tile_2rpntlvwz1rst1_internal(tdst, base, stride)
-
-#ifdef __DISABLE_AMX_MOVRS_TRANSPOSE__
-#undef __DISABLE_AMX_MOVRS_TRANSPOSE__
-#pragma GCC pop_options
-#endif /* __DISABLE_AMX_MOVRS_TRANSPOSE__ */
-
#endif /* __x86_64__ */
#endif /* _AMX_MOVRSINTRIN_H_INCLUDED */
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
<http://www.gnu.org/licenses/>. */
-#if !defined _IMMINTRIN_H_INCLUDED
-#error "Never use <amxtransposeintrin.h> directly; include <immintrin.h> instead."
-#endif
-
-#ifndef _AMXTRANSPOSEINTRIN_H_INCLUDED
-#define _AMXTRANSPOSEINTRIN_H_INCLUDED
-
-#if !defined(__AMX_TRANSPOSE__)
-#pragma GCC push_options
-#pragma GCC target("amx-transpose")
-#define __DISABLE_AMX_TRANSPOSE__
-#endif /* __AMX_TRANSPOSE__ */
-
-#if defined(__x86_64__)
-#define _tile_transposed_internal(dst,src) \
- __asm__ volatile\
- ("{ttransposed\t%%tmm"#src", %%tmm"#dst"|ttransposed\t%%tmm"#dst", %%tmm"#src"}" ::)
-
-#define _tile_2rpntlvwz0_internal(dst,base,stride) \
- __asm__ volatile\
- ("{t2rpntlvwz0\t(%0,%1,1), %%tmm"#dst"|t2rpntlvwz0\t%%tmm"#dst", [%0+%1*1]}" \
- :: "r" ((const void*) (base)), "r" ((long) (stride)))
-
-#define _tile_2rpntlvwz0t1_internal(dst,base,stride) \
- __asm__ volatile\
- ("{t2rpntlvwz0t1\t(%0,%1,1), %%tmm"#dst"|t2rpntlvwz0t1\t%%tmm"#dst", [%0+%1*1]}" \
- :: "r" ((const void*)(base)), "r" ((long)(stride)))
-
-#define _tile_2rpntlvwz1_internal(dst,base,stride) \
- __asm__ volatile\
- ("{t2rpntlvwz1\t(%0,%1,1), %%tmm"#dst"|t2rpntlvwz1\t%%tmm"#dst", [%0+%1*1]}" \
- :: "r" ((const void*)(base)), "r" ((long)(stride)))
-
-#define _tile_2rpntlvwz1t1_internal(dst,base,stride) \
- __asm__ volatile\
- ("{t2rpntlvwz1t1\t(%0,%1,1), %%tmm"#dst"|t2rpntlvwz1t1\t%%tmm"#dst", [%0+%1*1]}" \
- :: "r" ((const void*)(base)), "r" ((long)(stride)))
-
-#define _tile_transposed(dst,src) \
- _tile_transposed_internal (dst, src)
-
-#define _tile_2rpntlvwz0(dst,base,stride) \
- _tile_2rpntlvwz0_internal (dst, base, stride)
-
-#define _tile_2rpntlvwz0t1(dst,base,stride) \
- _tile_2rpntlvwz0t1_internal (dst, base, stride)
-
-#define _tile_2rpntlvwz1(dst,base,stride) \
- _tile_2rpntlvwz1_internal (dst, base, stride)
-
-#define _tile_2rpntlvwz1t1(dst,base,stride) \
- _tile_2rpntlvwz1t1_internal (dst, base, stride)
-
-#if !defined(__AMX_BF16__)
-#pragma GCC push_options
-#pragma GCC target("amx-bf16")
-#define __DISABLE_AMX_BF16__
-#endif /* __AMX_BF16__ */
-
-#define _tile_tdpbf16ps_internal(src1_dst,src2,src3) \
- __asm__ volatile\
- ("{ttdpbf16ps\t%%tmm"#src3", %%tmm"#src2", %%tmm"#src1_dst"|ttdpbf16ps\t%%tmm"#src1_dst", %%tmm"#src2", %%tmm"#src3"}" ::)
-
-#define _tile_tdpbf16ps(src1_dst,src2,src3) \
- _tile_tdpbf16ps_internal (src1_dst, src2, src3)
-
-#ifdef __DISABLE_AMX_BF16__
-#undef __DISABLE_AMX_BF16__
-#pragma GCC pop_options
-#endif /* __DISABLE_AMX_BF16__ */
-
-#if !defined(__AMX_FP16__)
-#pragma GCC push_options
-#pragma GCC target("amx-fp16")
-#define __DISABLE_AMX_FP16__
-#endif /* __AMX_FP16__ */
-
-#define _tile_tdpfp16ps_internal(src1_dst,src2,src3) \
- __asm__ volatile\
- ("{ttdpfp16ps\t%%tmm"#src3", %%tmm"#src2", %%tmm"#src1_dst"|ttdpfp16ps\t%%tmm"#src1_dst", %%tmm"#src2", %%tmm"#src3"}" ::)
-
-#define _tile_tdpfp16ps(src1_dst,src2,src3) \
- _tile_tdpfp16ps_internal (src1_dst, src2, src3)
-
-#ifdef __DISABLE_AMX_FP16__
-#undef __DISABLE_AMX_FP16__
-#pragma GCC pop_options
-#endif /* __DISABLE_AMX_FP16__ */
-
-#if !defined(__AMX_COMPLEX__)
-#pragma GCC push_options
-#pragma GCC target("amx-complex")
-#define __DISABLE_AMX_COMPLEX__
-#endif /* __AMX_COMPLEX__ */
-
-#define _tile_conjtcmmimfp16ps_internal(src1_dst,src2,src3) \
- __asm__ volatile\
- ("{tconjtcmmimfp16ps\t%%tmm"#src3", %%tmm"#src2", %%tmm"#src1_dst"|tconjtcmmimfp16ps\t%%tmm"#src1_dst", %%tmm"#src2", %%tmm"#src3"}" ::)
-
-#define _tile_conjtfp16_internal(dst,src) \
- __asm__ volatile\
- ("{tconjtfp16\t%%tmm"#src", %%tmm"#dst"|tconjtfp16\t%%tmm"#dst", %%tmm"#src"}" ::)
-
-#define _tile_tcmmimfp16ps_internal(src1_dst,src2,src3) \
- __asm__ volatile\
- ("{ttcmmimfp16ps\t%%tmm"#src3", %%tmm"#src2", %%tmm"#src1_dst"|ttcmmimfp16ps\t%%tmm"#src1_dst", %%tmm"#src2", %%tmm"#src3"}" ::)
-
-#define _tile_tcmmrlfp16ps_internal(src1_dst,src2,src3) \
- __asm__ volatile\
- ("{ttcmmrlfp16ps\t%%tmm"#src3", %%tmm"#src2", %%tmm"#src1_dst"|ttcmmrlfp16ps\t%%tmm"#src1_dst", %%tmm"#src2", %%tmm"#src3"}" ::)
-
-#define _tile_conjtcmmimfp16ps(src1_dst,src2,src3) \
- _tile_conjtcmmimfp16ps_internal (src1_dst, src2, src3)
-
-#define _tile_conjtfp16(dst,src) \
- _tile_conjtfp16_internal (dst, src)
-
-#define _tile_tcmmimfp16ps(src1_dst,src2,src3) \
- _tile_tcmmimfp16ps_internal (src1_dst, src2, src3)
-
-#define _tile_tcmmrlfp16ps(src1_dst,src2,src3) \
- _tile_tcmmrlfp16ps_internal (src1_dst, src2, src3)
-
-#ifdef __DISABLE_AMX_COMPLEX__
-#undef __DISABLE_AMX_COMPLEX__
-#pragma GCC pop_options
-#endif /* __DISABLE_AMX_COMPLEX__ */
-
-#if !defined(__AMX_TF32__)
-#pragma GCC push_options
-#pragma GCC target("amx-tf32")
-#define __DISABLE_AMX_TF32__
-#endif /* __AMX_TF32__ */
-
-#define _tile_tmmultf32ps_internal(src1_dst,src2,src3) \
- __asm__ volatile\
- ("{ttmmultf32ps\t%%tmm"#src3", %%tmm"#src2", %%tmm"#src1_dst"|ttmmultf32ps\t%%tmm"#src1_dst", %%tmm"#src2", %%tmm"#src3"}" ::)
-
-#define _tile_tmmultf32ps(src1_dst,src2,src3) \
- _tile_tmmultf32ps_internal (src1_dst, src2, src3)
-
-#ifdef __DISABLE_AMX_TF32__
-#undef __DISABLE_AMX_TF32__
-#pragma GCC pop_options
-#endif /* __DISABLE_AMX_TF32__ */
-
-#endif /* __x86_64__ */
-
-#ifdef __DISABLE_AMX_TRANSPOSE__
-#undef __DISABLE_AMX_TRANSPOSE__
-#pragma GCC pop_options
-#endif /* __DISABLE_AMX_TRANSPOSE__ */
-
-#endif /* _AMXTRANSPOSEINTRIN_H_INCLUDED */
+#error "AMX-TRANSPOSE support has been removed since GCC 16."
/* AMX sub leaf (%eax == 0x1e, %ecx == 1) */
/* %eax */
#define bit_AMX_FP8 (1 << 4)
-#define bit_AMX_TRANSPOSE (1 << 5)
#define bit_AMX_TF32 (1 << 6)
#define bit_AMX_AVX512 (1 << 7)
#define bit_AMX_MOVRS (1 << 8)
def_or_undef (parse_in, "__AMX_AVX512__");
if (isa_flag2 & OPTION_MASK_ISA2_AMX_TF32)
def_or_undef (parse_in, "__AMX_TF32__");
- if (isa_flag2 & OPTION_MASK_ISA2_AMX_TRANSPOSE)
- def_or_undef (parse_in, "__AMX_TRANSPOSE__");
if (isa_flag2 & OPTION_MASK_ISA2_AMX_FP8)
def_or_undef (parse_in, "__AMX_FP8__");
if (isa_flag2 & OPTION_MASK_ISA2_MOVRS)
DEF_PTA(AVX10_2)
DEF_PTA(AMX_AVX512)
DEF_PTA(AMX_TF32)
-DEF_PTA(AMX_TRANSPOSE)
DEF_PTA(AMX_FP8)
DEF_PTA(MOVRS)
DEF_PTA(AMX_MOVRS)
{ "-mavx10.2", OPTION_MASK_ISA2_AVX10_2 },
{ "-mamx-avx512", OPTION_MASK_ISA2_AMX_AVX512 },
{ "-mamx-tf32", OPTION_MASK_ISA2_AMX_TF32 },
- { "-mamx-transpose", OPTION_MASK_ISA2_AMX_TRANSPOSE },
{ "-mamx-fp8", OPTION_MASK_ISA2_AMX_FP8 },
{ "-mmovrs", OPTION_MASK_ISA2_MOVRS },
{ "-mamx-movrs", OPTION_MASK_ISA2_AMX_MOVRS }
IX86_ATTR_ISA ("avx10.2", OPT_mavx10_2),
IX86_ATTR_ISA ("amx-avx512", OPT_mamx_avx512),
IX86_ATTR_ISA ("amx-tf32", OPT_mamx_tf32),
- IX86_ATTR_ISA ("amx-transpose", OPT_mamx_transpose),
IX86_ATTR_ISA ("amx-fp8", OPT_mamx_fp8),
IX86_ATTR_ISA ("movrs", OPT_mmovrs),
IX86_ATTR_ISA ("amx-movrs", OPT_mamx_movrs),
constexpr wide_int_bitmask PTA_DIAMONDRAPIDS = PTA_GRANITERAPIDS_D
| PTA_AVXIFMA | PTA_AVXNECONVERT | PTA_AVXVNNIINT16 | PTA_AVXVNNIINT8
| PTA_CMPCCXADD | PTA_SHA512 | PTA_SM3 | PTA_SM4 | PTA_AVX10_2
- | PTA_APX_F | PTA_AMX_AVX512 | PTA_AMX_FP8 | PTA_AMX_TF32 | PTA_AMX_TRANSPOSE
- | PTA_MOVRS | PTA_AMX_MOVRS | PTA_USER_MSR;
+ | PTA_APX_F | PTA_AMX_AVX512 | PTA_AMX_FP8 | PTA_AMX_TF32 | PTA_MOVRS
+ | PTA_AMX_MOVRS | PTA_USER_MSR;
constexpr wide_int_bitmask PTA_BDVER1 = PTA_64BIT | PTA_MMX | PTA_SSE
| PTA_SSE2 | PTA_SSE3 | PTA_SSE4A | PTA_CX16 | PTA_POPCNT | PTA_LZCNT
Target Mask(ISA2_AMX_TF32) Var(ix86_isa_flags2) Save
Support AMX-TF32 built-in functions and code generation.
-mamx-transpose
-Target Mask(ISA2_AMX_TRANSPOSE) Var(ix86_isa_flags2) Save
-Support AMX-TRANSPOSE built-in functions and code generation.
-
mamx-fp8
Target Mask(ISA2_AMX_FP8) Var(ix86_isa_flags2) Save
Support AMX-FP8 built-in functions and code generation.
mamx-tf32
UrlSuffix(gcc/x86-Options.html#index-mamx-tf32)
-mamx-transpose
-UrlSuffix(gcc/x86-Options.html#index-mamx-transpose)
-
mamx-fp8
UrlSuffix(gcc/x86-Options.html#index-mamx-fp8)
#include <amxtf32intrin.h>
-#include <amxtransposeintrin.h>
-
#include <amxfp8intrin.h>
#include <prfchwintrin.h>
@itemx no-amx-tf32
Enable/disable the generation of the AMX-TF32 instructions.
-@cindex @code{target("amx-transpose")} function attribute, x86
-@item amx-transpose
-@itemx no-amx-transpose
-Enable/disable the generation of the AMX-TRANSPOSE instructions.
-
@cindex @code{target("amx-fp8")} function attribute, x86
@item amx-fp8
@itemx no-amx-fp8
-mamx-tile -mamx-int8 -mamx-bf16 -muintr -mhreset -mavxvnni -mamx-fp8
-mavx512fp16 -mavxifma -mavxvnniint8 -mavxneconvert -mcmpccxadd -mamx-fp16
-mprefetchi -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mapxf
--musermsr -mavx10.1 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-transpose -mmovrs
--mamx-movrs
+-musermsr -mavx10.1 -mavx10.2 -mamx-avx512 -mamx-tf32 -mmovrs -mamx-movrs
-mcldemote -mms-bitfields -mno-align-stringops -minline-all-stringops
-minline-stringops-dynamically -mstringop-strategy=@var{alg}
-mkl -mwidekl
@opindex mamx-tf32
@itemx -mamx-tf32
@need 200
-@opindex mamx-transpose
-@itemx -mamx-transpose
-@need 200
@itemx -mamx-fp8
@opindex mamx-fp8
@need 200
AVX512VPOPCNTDQ, AVX512VNNI, SERIALIZE, UINTR, HRESET, AMXTILE, AMXINT8,
AMXBF16, KL, WIDEKL, AVXVNNI, AVX512-FP16, AVXIFMA, AVXVNNIINT8, AVXNECONVERT,
CMPCCXADD, AMX-FP16, PREFETCHI, RAOINT, AMX-COMPLEX, AVXVNNIINT16, SM3, SHA512,
-SM4, APX_F, USER_MSR, AVX10.1, AVX10.2, AMX-AVX512, AMX-TF32, AMX-TRANSPOSE,
-AMX-FP8, MOVRS, AMX-MOVRS or CLDEMOTE extended instruction sets. Each has a
-corresponding @option{-mno-} option to disable use of these instructions.
+SM4, APX_F, USER_MSR, AVX10.1, AVX10.2, AMX-AVX512, AMX-TF32, AMX-FP8, MOVRS,
+AMX-MOVRS or CLDEMOTE extended instruction sets. Each has a corresponding
+@option{-mno-} option to disable use of these instructions.
These extensions are also available as built-in functions: see
@ref{x86 Built-in Functions}, for details of the functions enabled and
@item amx_tf32
Target supports the execution of @code{amx-tf32} instructions.
-@item amx_transpose
-Target supports the execution of @code{amx-transpose} instructions.
-
@item amx_fp8
Target supports the execution of @code{amx-fp8} instructions.
/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
-/* { dg-options "-O -pedantic-errors -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mcmpccxadd -mamx-fp16 -mprefetchi -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-transpose -mamx-fp8 -mmovrs -mamx-movrs" } */
+/* { dg-options "-O -pedantic-errors -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mcmpccxadd -mamx-fp16 -mprefetchi -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-fp8 -mmovrs -mamx-movrs" } */
/* { dg-skip-if "requires hosted libstdc++ for cstdlib malloc" { ! hostedlib } } */
/* Test that {,x,e,p,t,s,w,a,b,i}mmintrin.h, mm3dnow.h, fma4intrin.h,
/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
-/* { dg-options "-O -fkeep-inline-functions -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mcmpccxadd -mamx-fp16 -mprefetchi -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-transpose -mamx-fp8 -mmovrs -mamx-movrs" } */
+/* { dg-options "-O -fkeep-inline-functions -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mcmpccxadd -mamx-fp16 -mprefetchi -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-fp8 -mmovrs -mamx-movrs" } */
/* { dg-skip-if "requires hosted libstdc++ for cstdlib malloc" { ! hostedlib } } */
/* Test that {,x,e,p,t,s,w,a,b,i}mmintrin.h, mm3dnow.h, fma4intrin.h,
#ifdef AMX_TF32
&& __builtin_cpu_supports ("amx-tf32")
#endif
-#ifdef AMX_TRANSPOSE
- && __builtin_cpu_supports ("amx-transpose")
-#endif
#ifdef AMX_FP8
&& __builtin_cpu_supports ("amx-fp8")
#endif
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_movrs } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-require-effective-target avx512fp16 } */
-/* { dg-options "-O2 -mamx-movrs -mamx-transpose -mavx512fp16 -mavx512bf16" } */
-#define AMX_MOVRS
-#define AMX_TRANSPOSE
-#define DO_TEST test_amx_movrs_t2rpntlvwrs
-void test_amx_movrs_t2rpntlvwrs ();
-#include "amx-helper.h"
-
-#define init_pair_tile_reg_and_src_z_t(tmm_num, src, buffer, ztype, wtype) \
-{ \
- init_pair_tile_src (tmm_num, &src, buffer, ztype); \
- _tile_2rpntlvwz##ztype##rs##wtype (tmm_num, buffer, _STRIDE); \
-}
-
-void test_amx_movrs_t2rpntlvwrs ()
-{
- __tilecfg_u cfg;
- __tilepair src;
- __tile ref_0, ref_1;
- uint8_t buffer[2048];
- int i;
-
- init_tile_config (&cfg);
-
- for (i = 0; i < 2048; i++)
- buffer[i] = i % 256;
-
- /* Check t2rpntlvwz0rs. */
- init_pair_tile_reg_and_src_z_t (0, src, buffer, 0,);
- _tile_stored (0, ref_0.buf, _STRIDE);
- _tile_stored (1, ref_1.buf, _STRIDE);
- if (!check_pair_tile_register (&ref_0, &ref_1, &src))
- abort ();
-
- /* Check t2rpntlvwz1rs. */
- init_pair_tile_reg_and_src_z_t (0, src, buffer, 1,);
- _tile_stored (0, ref_0.buf, _STRIDE);
- _tile_stored (1, ref_1.buf, _STRIDE);
- if (!check_pair_tile_register (&ref_0, &ref_1, &src))
- abort ();
-
- /* Check t2rpntlvwz0t1rs. */
- init_pair_tile_reg_and_src_z_t (0, src, buffer, 0, t1);
- _tile_stored (0, ref_0.buf, _STRIDE);
- _tile_stored (1, ref_1.buf, _STRIDE);
- if (!check_pair_tile_register (&ref_0, &ref_1, &src))
- abort ();
-
- /* Check t2rpntlvwz1t1rs. */
- init_pair_tile_reg_and_src_z_t (0, src, buffer, 1, t1);
- _tile_stored (0, ref_0.buf, _STRIDE);
- _tile_stored (1, ref_1.buf, _STRIDE);
- if (!check_pair_tile_register (&ref_0, &ref_1, &src))
- abort ();
-}
/* { dg-do compile { target { ! ia32 } } } */
-/* { dg-options "-O2 -mamx-movrs -mamx-transpose" } */
+/* { dg-options "-O2 -mamx-movrs" } */
/* { dg-final { scan-assembler "tileloaddrs\[ \\t]+\[^\n\]*\(%\[a-z0-9\]*\,%\[a-z0-9\]*\,\[124\]\)+\[^\n\]*%tmm\[0-9\]" } } */
/* { dg-final { scan-assembler "tileloaddrst1\[ \\t]+\[^\n\]*\(%\[a-z0-9]*\,%\[a-z0-9\]*\,\[124\]\)+\[^\n\]*%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz0rs\[ \\t]+\[^\n\]*\(%\[a-z0-9\]*\,%\[a-z0-9\]*\,\[124\]\)+\[^\n\]*%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz0rst1\[ \\t]+\[^\n\]*\(%\[a-z0-9\]*\,%\[a-z0-9\]*\,\[124\]\)+\[^\n\]*%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz1rs\[ \\t]+\[^\n\]*\(%\[a-z0-9\]*\,%\[a-z0-9\]*\,\[124\]\)+\[^\n\]*%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz1rst1\[ \\t]+\[^\n\]*\(%\[a-z0-9\]*\,%\[a-z0-9\]*\,\[124\]\)+\[^\n\]*%tmm\[0-9\]" } } */
#include <immintrin.h>
extern const void* base;
{
_tile_loaddrs (TMM1, base, stride);
_tile_loaddrst1 (TMM1, base, stride);
- _tile_2rpntlvwz0rs (TMM0, base, stride);
- _tile_2rpntlvwz0rst1 (TMM1, base, stride);
- _tile_2rpntlvwz1rs (TMM2, base, stride);
- _tile_2rpntlvwz1rst1 (TMM3, base, stride);
}
/* { dg-do compile { target { ! ia32 } } } */
/* { dg-require-effective-target masm_intel } */
-/* { dg-options "-O2 -mamx-movrs -mamx-transpose -masm=intel" } */
+/* { dg-options "-O2 -mamx-movrs -masm=intel" } */
/* { dg-final { scan-assembler-times "tileloaddrs\[ \\t]%tmm\[0-9\]" 1 } } */
/* { dg-final { scan-assembler-times "tileloaddrst1\[ \\t]%tmm\[0-9\]" 1 } } */
-/* { dg-final { scan-assembler "t2rpntlvwz0rs\[ \\t]%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz0rst1\[ \\t]%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz1rs\[ \\t]%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz1rst1\[ \\t]%tmm\[0-9\]" } } */
#include <immintrin.h>
extern const void* base;
{
_tile_loaddrs (TMM1, base, stride);
_tile_loaddrst1 (TMM1, base, stride);
- _tile_2rpntlvwz0rs (TMM0, base, stride);
- _tile_2rpntlvwz0rst1 (TMM1, base, stride);
- _tile_2rpntlvwz1rs (TMM2, base, stride);
- _tile_2rpntlvwz1rst1 (TMM3, base, stride);
}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-options "-O2 -mamx-transpose" } */
-#define AMX_TRANSPOSE
-#define DO_TEST test_amx_transpose_t2rpntlvw
-void test_amx_transpose_t2rpntlvw ();
-#include "amx-helper.h"
-#define init_pair_tile_reg_and_src_z_t(tmm_num, src, buffer, ztype, wtype) \
-{ \
- init_pair_tile_src (tmm_num, &src, buffer, ztype); \
- _tile_2rpntlvwz##ztype##wtype (tmm_num, buffer, _STRIDE); \
-}
-
-void test_amx_transpose_t2rpntlvw ()
-{
- __tilecfg_u cfg;
- __tilepair src;
- __tile ref_0, ref_1;
- uint8_t buffer[2048];
- int i;
-
- init_tile_config (&cfg);
-
- for (i = 0; i < 2048; i++)
- buffer[i] = i % 256;
-
- /* Check t2rpntlvwz0. */
- init_pair_tile_reg_and_src_z_t (0, src, buffer, 0,);
- _tile_stored (0, ref_0.buf, _STRIDE);
- _tile_stored (1, ref_1.buf, _STRIDE);
- if (!check_pair_tile_register (&ref_0, &ref_1, &src))
- abort ();
-
- /* Check t2rpntlvwz1. */
- init_pair_tile_reg_and_src_z_t (0, src, buffer, 1,);
- _tile_stored (0, ref_0.buf, _STRIDE);
- _tile_stored (1, ref_1.buf, _STRIDE);
- if (!check_pair_tile_register (&ref_0, &ref_1, &src))
- abort ();
-
- /* Check t2rpntlvwz0t1. */
- init_pair_tile_reg_and_src_z_t (0, src, buffer, 0, t1);
- _tile_stored (0, ref_0.buf, _STRIDE);
- _tile_stored (1, ref_1.buf, _STRIDE);
- if (!check_pair_tile_register (&ref_0, &ref_1, &src))
- abort ();
-
- /* Check t2rpntlvwz1t1. */
- init_pair_tile_reg_and_src_z_t (0, src, buffer, 1, t1);
- _tile_stored (0, ref_0.buf, _STRIDE);
- _tile_stored (1, ref_1.buf, _STRIDE);
- if (!check_pair_tile_register (&ref_0, &ref_1, &src))
- abort ();
-}
+++ /dev/null
-/* { dg-do compile { target { ! ia32 } } } */
-/* { dg-options "-O2 -mamx-transpose -mamx-bf16 -mamx-complex -mamx-fp16 -mamx-tf32" } */
-/* { dg-final { scan-assembler "ttdpbf16ps\[ \\t]+\[^\n\]*%tmm3+\[^\n\]*%tmm2+\[^\n\]*%tmm1" } } */
-/* { dg-final { scan-assembler "ttdpfp16ps\[ \\t]+\[^\n\]*%tmm3+\[^\n\]*%tmm2+\[^\n\]*%tmm1" } } */
-/* { dg-final { scan-assembler "ttransposed\[ \\t]+\[^\n\]*%tmm2+\[^\n\]*%tmm1" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz0\[ \\t]+\[^\n\]*\\(%\[a-z0-9]*\,%\[a-z0-9\]*\,\[124\]\\)+\[^\n\]*%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz0t1\[ \\t]+\[^\n\]*\\(%\[a-z0-9]*\,%\[a-z0-9\]*\,\[124\]\\)+\[^\n\]*%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz1\[ \\t]+\[^\n\]*\\(%\[a-z0-9]*\,%\[a-z0-9\]*\,\[124\]\\)+\[^\n\]*%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz1t1\[ \\t]+\[^\n\]*\\(%\[a-z0-9]*\,%\[a-z0-9\]*\,\[124\]\\)+\[^\n\]*%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "tconjtcmmimfp16ps\[ \\t]+\[^\n\]*%tmm3+\[^\n\]*%tmm2+\[^\n\]*%tmm1" } } */
-/* { dg-final { scan-assembler "tconjtfp16\[ \\t]+\[^\n\]*%tmm2+\[^\n\]*%tmm1" } } */
-/* { dg-final { scan-assembler "ttcmmimfp16ps\[ \\t]+\[^\n\]*%tmm3+\[^\n\]*%tmm2+\[^\n\]*%tmm1" } } */
-/* { dg-final { scan-assembler "ttcmmrlfp16ps\[ \\t]+\[^\n\]*%tmm3+\[^\n\]*%tmm2+\[^\n\]*%tmm1" } } */
-/* { dg-final { scan-assembler "ttmmultf32ps\[ \\t]+\[^\n\]*%tmm3+\[^\n\]*%tmm2+\[^\n\]*%tmm1" } } */
-#include <immintrin.h>
-
-extern const void* base;
-extern const int stride;
-
-#define TMM0 0
-#define TMM1 1
-#define TMM2 2
-#define TMM3 3
-
-void TEST()
-{
- _tile_tdpbf16ps (TMM1, TMM2, TMM3);
- _tile_tdpfp16ps (TMM1, TMM2, TMM3);
- _tile_transposed (TMM1, TMM2);
- _tile_2rpntlvwz0 (TMM0, base, stride);
- _tile_2rpntlvwz0t1 (TMM1, base, stride);
- _tile_2rpntlvwz1 (TMM2, base, stride);
- _tile_2rpntlvwz1t1 (TMM3, base, stride);
- _tile_conjtcmmimfp16ps (TMM1, TMM2, TMM3);
- _tile_conjtfp16 (TMM1, TMM2);
- _tile_tcmmimfp16ps (TMM1, TMM2, TMM3);
- _tile_tcmmrlfp16ps (TMM1, TMM2, TMM3);
- _tile_tmmultf32ps (TMM1, TMM2, TMM3);
-}
+++ /dev/null
-/* { dg-do compile { target { ! ia32 } } } */
-/* { dg-require-effective-target masm_intel } */
-/* { dg-options "-O2 -mamx-transpose -mamx-bf16 -mamx-complex -mamx-fp16 -mamx-tf32 -masm=intel" } */
-/* { dg-final { scan-assembler "ttdpbf16ps\[ \\t]+\[^\n\]*%tmm1+\[^\n\]*%tmm2+\[^\n\]*%tmm3" } } */
-/* { dg-final { scan-assembler "ttdpfp16ps\[ \\t]+\[^\n\]*%tmm1+\[^\n\]*%tmm2+\[^\n\]*%tmm3" } } */
-/* { dg-final { scan-assembler "ttransposed\[ \\t]+\[^\n\]*%tmm1+\[^\n\]*%tmm2" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz0\[ \\t]%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz0t1\[ \\t]%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz1\[ \\t]%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "t2rpntlvwz1t1\[ \\t]%tmm\[0-9\]" } } */
-/* { dg-final { scan-assembler "tconjtcmmimfp16ps\[ \\t]+\[^\n\]*%tmm1+\[^\n\]*%tmm2+\[^\n\]*%tmm3" } } */
-/* { dg-final { scan-assembler "tconjtfp16\[ \\t]+\[^\n\]*%tmm1+\[^\n\]*%tmm2" } } */
-/* { dg-final { scan-assembler "ttcmmimfp16ps\[ \\t]+\[^\n\]*%tmm1+\[^\n\]*%tmm2+\[^\n\]*%tmm3" } } */
-/* { dg-final { scan-assembler "ttcmmrlfp16ps\[ \\t]+\[^\n\]*%tmm1+\[^\n\]*%tmm2+\[^\n\]*%tmm3" } } */
-/* { dg-final { scan-assembler "ttmmultf32ps\[ \\t]+\[^\n\]*%tmm1+\[^\n\]*%tmm2+\[^\n\]*%tmm3" } } */
-#include <immintrin.h>
-
-extern const void* base;
-extern const int stride;
-
-void TEST()
-{
- _tile_tdpbf16ps (1, 2, 3);
- _tile_tdpfp16ps (1, 2, 3);
- _tile_transposed (1, 2);
- _tile_2rpntlvwz0 (5, base, stride);
- _tile_2rpntlvwz0t1 (4, base, stride);
- _tile_2rpntlvwz1 (3, base, stride);
- _tile_2rpntlvwz1t1 (2, base, stride);
- _tile_conjtcmmimfp16ps (1, 2, 3);
- _tile_conjtfp16 (1, 2);
- _tile_tcmmimfp16ps (1, 2, 3);
- _tile_tcmmrlfp16ps (1, 2, 3);
- _tile_tmmultf32ps (1, 2, 3);
-}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-require-effective-target amx_complex } */
-/* { dg-require-effective-target avx512fp16 } */
-/* { dg-options "-O2 -mamx-transpose -mamx-complex -mavx512fp16" } */
-#define AMX_TRANSPOSE
-#define AMX_COMPLEX
-#define DO_TEST test_amx_transpose_conjtcmmimfp16ps
-void test_amx_transpose_conjtcmmimfp16ps ();
-#include "amx-helper.h"
-
-void calc_matrix_conjtcmmimfp16ps (__tile *dst, __tile *src1, __tile *src2)
-{
- uint16_t *src1_buf = (uint16_t *) src1->buf;
- uint16_t *src2_buf = (uint16_t *) src2->buf;
- float *dst_buf = (float *) dst->buf;
-
- int K = src1->rows;
- int M = src1->colsb / 4;
- int N = src2->colsb / 4;
- int m, k, n, t;
-
- for (m = 0; m < M; m++)
- for (k = 0; k < K; k++)
- for (n = 0; n < N; n++)
- for (t = 0; t < 2; t+=2)
- dst_buf[m * N + n] +=
- (make_fp16_f32(src1_buf[k * 2 * M + 2 * m + t]) *
- make_fp16_f32(src2_buf[k * 2 * N + 2 * n + t + 1])) -
- (make_fp16_f32(src1_buf[k * 2 * M + 2 * m + t + 1]) *
- make_fp16_f32(src2_buf[k * 2 * N + 2 * n + t]));
-}
-
-void test_amx_transpose_conjtcmmimfp16ps ()
-{
- __tilecfg_u cfg;
- __tile dst, dst_ref, src1, src2;
- uint8_t tmp_dst_buf[1024], tmp_dst_zero_buf[1024];
-
- init_fp16_max_tile_buffer (tmp_dst_buf);
- init_fp16_max_tile_zero_buffer (tmp_dst_zero_buf);
-
- init_tile_config (&cfg);
- init_tile_reg_and_src_with_buffer (1, dst, tmp_dst_zero_buf);
- init_tile_reg_and_src_with_buffer (2, src1, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (3, src2, tmp_dst_buf);
-
- calc_matrix_conjtcmmimfp16ps (&dst, &src1, &src2);
-
- _tile_conjtcmmimfp16ps (1, 2, 3);
- _tile_stored (1, dst_ref.buf, _STRIDE);
-
- if (!check_tile_register (&dst_ref, &dst))
- abort ();
-}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-require-effective-target amx_complex } */
-/* { dg-require-effective-target avx512fp16 } */
-/* { dg-options "-O2 -mamx-transpose -mamx-complex -mavx512fp16" } */
-#define AMX_TRANSPOSE
-#define AMX_COMPLEX
-#define DO_TEST test_amx_transpose_conjtfp16
-void test_amx_transpose_conjtfp16 ();
-#include "amx-helper.h"
-
-void calc_matrix_conjtfp16 (__tile *dst, __tile *src)
-{
- uint16_t *src_buf = (uint16_t *) src->buf;
- float *dst_buf = (float *) dst->buf;
-
- int M = dst->rows;
- int N = dst->colsb / 4;
- int i, j, t;
-
- for (i = 0; i < M; i++)
- for (j = 0; j < N; j++)
- for (t = 0; t < 2; t+=2)
- {
- dst_buf[i * 2 * N + 2 * j + t] = src_buf[j * 2 * M + 2 * i + t];
- dst_buf[i * 2 * N + 2 * j + t + 1] = -src_buf[j * 2 * M + 2 * i + t + 1];
- }
-}
-
-void test_amx_transpose_conjtfp16 ()
-{
- __tilecfg_u cfg;
- __tile src, dst, ref;
- uint8_t tmp_dst_buf[1024];
-
- init_fp16_max_tile_buffer (tmp_dst_buf);
- init_tile_config (&cfg);
- init_tile_reg_and_src_with_buffer (1, dst, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (2, src, tmp_dst_buf);
-
- /* Check tconjtfp16. */
- calc_matrix_conjtfp16 (&dst, &src);
- _tile_conjtfp16 (1, 2);
- _tile_stored (1, ref.buf, _STRIDE);
-
- if (!check_tile_register (&ref, &dst))
- abort ();
-}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-require-effective-target amx_complex } */
-/* { dg-require-effective-target avx512fp16 } */
-/* { dg-options "-O2 -mamx-transpose -mamx-complex -mavx512fp16" } */
-#define AMX_TRANSPOSE
-#define AMX_COMPLEX
-#define DO_TEST test_amx_transpose_tcmmimfp16ps
-void test_amx_transpose_tcmmimfp16ps ();
-#include "amx-helper.h"
-
-void calc_matrix_tcmmimfp16ps (__tile *dst, __tile *src1, __tile *src2)
-{
- uint16_t *src1_buf = (uint16_t *) src1->buf;
- uint16_t *src2_buf = (uint16_t *) src2->buf;
- float *dst_buf = (float *) dst->buf;
-
- int K = src1->rows;
- int M = src1->colsb / 4;
- int N = src2->colsb / 4;
- int m, k, n, t;
-
- for (m = 0; m < M; m++)
- for (k = 0; k < K; k++)
- for (n = 0; n < N; n++)
- for (t = 0; t < 2; t+=2)
- dst_buf[m * N + n] +=
- (make_fp16_f32(src1_buf[k * 2 * M + 2 * m + t]) *
- make_fp16_f32(src2_buf[k * 2 * N + 2 * n + t + 1])) +
- (make_fp16_f32(src1_buf[k * 2 * M + 2 * m + t + 1]) *
- make_fp16_f32(src2_buf[k * 2 * N + 2 * n + t]));
-}
-
-void test_amx_transpose_tcmmimfp16ps ()
-{
- __tilecfg_u cfg;
- __tile dst, dst_ref, src1, src2;
- uint8_t tmp_dst_buf[1024], tmp_dst_zero_buf[1024];
-
- init_fp16_max_tile_buffer (tmp_dst_buf);
- init_fp16_max_tile_zero_buffer (tmp_dst_zero_buf);
-
- init_tile_config (&cfg);
- init_tile_reg_and_src_with_buffer (1, dst, tmp_dst_zero_buf);
- init_tile_reg_and_src_with_buffer (2, src1, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (3, src2, tmp_dst_buf);
-
- calc_matrix_tcmmimfp16ps (&dst, &src1, &src2);
-
- _tile_tcmmimfp16ps (1, 2, 3);
- _tile_stored (1, dst_ref.buf, _STRIDE);
-
- if (!check_tile_register (&dst_ref, &dst))
- abort ();
-}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-require-effective-target amx_complex } */
-/* { dg-require-effective-target avx512fp16 } */
-/* { dg-options "-O2 -mamx-transpose -mamx-complex -mavx512fp16" } */
-#define AMX_TRANSPOSE
-#define AMX_COMPLEX
-#define DO_TEST test_amx_transpose_tcmmrlfp16ps
-void test_amx_transpose_tcmmrlfp16ps ();
-#include "amx-helper.h"
-
-void calc_matrix_tcmmrlfp16ps (__tile *dst, __tile *src1, __tile *src2)
-{
- uint16_t *src1_buf = (uint16_t *) src1->buf;
- uint16_t *src2_buf = (uint16_t *) src2->buf;
- float *dst_buf = (float *) dst->buf;
-
- int K = src1->rows;
- int M = src1->colsb / 4;
- int N = src2->colsb / 4;
- int m, k, n, t;
-
- for (m = 0; m < M; m++)
- for (k = 0; k < K; k++)
- for (n = 0; n < N; n++)
- for (t = 0; t < 2; t+=2)
- dst_buf[m * N + n] +=
- (make_fp16_f32(src1_buf[k * 2 * M + 2 * m + t]) *
- make_fp16_f32(src2_buf[k * 2 * N + 2 * n + t])) -
- (make_fp16_f32(src1_buf[k * 2 * M + 2 * m + t + 1]) *
- make_fp16_f32(src2_buf[k * 2 * N + 2 * n + t + 1]));
-}
-
-void test_amx_transpose_tcmmrlfp16ps ()
-{
- __tilecfg_u cfg;
- __tile dst, dst_ref, src1, src2;
- uint8_t tmp_dst_buf[1024], tmp_dst_zero_buf[1024];
-
- init_fp16_max_tile_buffer (tmp_dst_buf);
- init_fp16_max_tile_zero_buffer (tmp_dst_zero_buf);
-
- init_tile_config (&cfg);
- init_tile_reg_and_src_with_buffer (1, dst, tmp_dst_zero_buf);
- init_tile_reg_and_src_with_buffer (2, src1, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (3, src2, tmp_dst_buf);
-
- calc_matrix_tcmmrlfp16ps (&dst, &src1, &src2);
-
- _tile_tcmmrlfp16ps (1, 2, 3);
- _tile_stored (1, dst_ref.buf, _STRIDE);
-
- if (!check_tile_register (&dst_ref, &dst))
- abort ();
-}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-require-effective-target amx_bf16 } */
-/* { dg-options "-O2 -mamx-transpose -mamx-bf16 -mavx512bf16" } */
-#define AMX_TRANSPOSE
-#define AMX_BF16
-#define DO_TEST test_amx_transpose_tdpbf16ps
-void test_amx_transpose_tdpbf16ps ();
-#include "amx-helper.h"
-
-void calc_matrix_tdpbf16ps(__tile *dst, __tile *src1, __tile *src2)
-{
- uint16_t *src1_buf = (uint16_t *) src1->buf;
- uint16_t *src2_buf = (uint16_t *) src2->buf;
- float *dst_buf = (float *) dst->buf;
-
- int K = src1->rows;
- int M = src1->colsb / 4;
- int N = src2->colsb / 4;
- int m, k, n, t;
-
- for (m = 0; m < M; m++)
- for (k = 0; k < K; k++)
- for (n = 0; n < N; n++)
- for (t = 0; t < 2; t+=2)
- dst_buf[m * N + n] +=
- (make_bf16_f32 (src1_buf[k * 2 * M + 2 * m + t]) *
- make_bf16_f32 (src2_buf[k * 2 * N + 2 * n + t])) +
- (make_bf16_f32 (src1_buf[k * 2 * M + 2 * m + t + 1]) *
- make_bf16_f32 (src2_buf[k * 2 * N + 2 * n + t + 1]));
-}
-
-void test_amx_transpose_tdpbf16ps ()
-{
- __tilecfg_u cfg;
- __tile dst, dst_ref, src1, src2;
- uint8_t tmp_dst_buf[1024];
-
- init_bf16_max_tile_buffer (tmp_dst_buf);
-
- init_tile_config (&cfg);
- init_tile_reg_and_src_with_buffer (1, dst, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (2, src1, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (3, src2, tmp_dst_buf);
-
- calc_matrix_tdpbf16ps (&dst, &src1, &src2);
-
- _tile_tdpbf16ps (1, 2, 3);
- _tile_stored (1, dst_ref.buf, _STRIDE);
-
- if (!check_float_tile_register (&dst_ref, &dst))
- abort ();
-}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-require-effective-target amx_fp16 } */
-/* { dg-require-effective-target avx512fp16 } */
-/* { dg-options "-O2 -mamx-transpose -mamx-fp16 -mavx512fp16" } */
-#define AMX_TRANSPOSE
-#define AMX_FP16
-#define DO_TEST test_amx_transpose_tdpfp16ps
-void test_amx_transpose_tdpfp16ps ();
-#include "amx-helper.h"
-
-void calc_matrix_tdpfp16ps(__tile *dst, __tile *src1, __tile *src2)
-{
- uint16_t *src1_buf = (uint16_t *) src1->buf;
- uint16_t *src2_buf = (uint16_t *) src2->buf;
- float *dst_buf = (float *) dst->buf;
-
- int K = src1->rows;
- int M = src1->colsb / 4;
- int N = src2->colsb / 4;
- int m, k, n, t;
-
- for (m = 0; m < M; m++)
- for (k = 0; k < K; k++)
- for (n = 0; n < N; n++)
- for (t = 0; t < 2; t+=2)
- dst_buf[m * N + n] +=
- (make_fp16_f32 (src1_buf[k * 2 * M + 2 * m + t]) *
- make_fp16_f32 (src2_buf[k * 2 * N + 2 * n + t])) +
- (make_fp16_f32 (src1_buf[k * 2 * M + 2 * m + t + 1]) *
- make_fp16_f32 (src2_buf[k * 2 * N + 2 * n + t + 1]));
-}
-
-void test_amx_transpose_tdpfp16ps ()
-{
- __tilecfg_u cfg;
- __tile dst, dst_ref, src1, src2;
- uint8_t tmp_dst_buf[1024], tmp_dst_zero_buf[1024];
-
- init_fp16_max_tile_buffer (tmp_dst_buf);
- init_fp16_max_tile_zero_buffer(tmp_dst_zero_buf);
-
- init_tile_config (&cfg);
- init_tile_reg_and_src_with_buffer (1, dst, tmp_dst_zero_buf);
- init_tile_reg_and_src_with_buffer (2, src1, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (3, src2, tmp_dst_buf);
-
- calc_matrix_tdpfp16ps (&dst, &src1, &src2);
-
- _tile_tdpfp16ps (1, 2, 3);
- _tile_stored (1, dst_ref.buf, _STRIDE);
-
- if (!check_float_tile_register (&dst_ref, &dst))
- abort ();
-}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-require-effective-target amx_tf32 } */
-/* { dg-options "-O2 -mamx-transpose -mamx-tf32" } */
-#define AMX_TRANSPOSE
-#define AMX_TF32
-#define DO_TEST test_amx_transpose_tmmultf32ps
-void test_amx_transpose_tmmultf32ps();
-#include "amx-helper.h"
-
-void calc_matrix_tmmultf32ps(__tile *dst, __tile *src1, __tile *src2)
-{
- float *src1_buf = (float *) src1->buf;
- float *src2_buf = (float *) src2->buf;
- float *dst_buf = (float *) dst->buf;
-
- int K = src1->rows;
- int M = src1->colsb / 4;
- int N = src2->colsb / 4;
- int m, n, k;
-
- for (m = 0; m < M; m++)
- for (k = 0; k < K; k++)
- for (n = 0; n < N; n++)
- dst_buf[m * N + n] +=
- zero_lower_mantissa_bits_fp32 (silence_snan_fp32 (src1_buf[k * M + m])) *
- zero_lower_mantissa_bits_fp32 (silence_snan_fp32 (src2_buf[k * N + n]));
-
-}
-
-void test_amx_transpose_tmmultf32ps ()
-{
- __tilecfg_u cfg;
- __tile dst, dst_ref, src1, src2;
- uint8_t tmp_dst_buf[1024];
-
- init_fp32_max_tile_buffer (tmp_dst_buf);
-
- init_tile_config (&cfg);
- init_tile_reg_and_src_with_buffer (1, dst, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (2, src1, tmp_dst_buf);
- init_tile_reg_and_src_with_buffer (3, src2, tmp_dst_buf);
-
- calc_matrix_tmmultf32ps (&dst, &src1, &src2);
-
- _tile_tmmultf32ps (1, 2, 3);
- _tile_stored (1, dst_ref.buf, _STRIDE);
-
- if (!check_tile_register (&dst_ref, &dst))
- abort ();
-}
+++ /dev/null
-/* { dg-do run { target { ! ia32 } } } */
-/* { dg-require-effective-target amx_transpose } */
-/* { dg-options "-O2 -mamx-transpose" } */
-#define AMX_TRANSPOSE
-#define DO_TEST test_amx_transpose_transposed
-void test_amx_transpose_transposed ();
-#include "amx-helper.h"
-
-void calc_matrix_ttransposed (__tile *dst, __tile *src)
-{
- uint32_t *src_buf = (uint32_t *) src->buf;
- uint32_t *dst_buf = (uint32_t *) dst->buf;
-
- int M = src->rows;
- int N = src->colsb / 4;
- int i, j;
-
- for (i = 0; i < M; i++)
- for (j = 0; j < N; j++)
- dst_buf[j * M + i] = (uint32_t) src_buf[i * N + j];
-}
-
-void test_amx_transpose_transposed ()
-{
- __tilecfg_u cfg;
- __tile src, dst, ref;
-
- init_tile_config (&cfg);
- init_tile_reg_and_src (1, dst);
- init_tile_reg_and_src (2, src);
-
- /* Check ttransposed. */
- calc_matrix_ttransposed (&dst, &src);
- _tile_transposed (1, 2);
- _tile_stored (1, ref.buf, _STRIDE);
-
- if (!check_tile_register (&ref, &dst))
- abort ();
-}
extern void test_avx10_2 (void) __attribute__((__target__("avx10.2")));
extern void test_amx_avx512 (void) __attribute__((__target__("amx-avx512")));
extern void test_amx_tf32 (void) __attribute__((__target__("amx-tf32")));
-extern void test_amx_transpose (void) __attribute__((__target__("amx-transpose")));
extern void test_amx_fp8 (void) __attribute__((__target__("amx-fp8")));
extern void test_movrs (void) __attribute__((__target__("movrs")));
extern void test_amx_movrs (void) __attribute__((__target__("amx-movrs")));
extern void test_no_avx10_2 (void) __attribute__((__target__("no-avx10.2")));
extern void test_no_amx_avx512 (void) __attribute__((__target__("no-amx-avx512")));
extern void test_no_amx_tf32 (void) __attribute__((__target__("no-amx-tf32")));
-extern void test_no_amx_transpose (void) __attribute__((__target__("no-amx-transpose")));
extern void test_no_amx_fp8 (void) __attribute__((__target__("no-amx-fp8")));
extern void test_no_movrs (void) __attribute__((__target__("no-movrs")));
extern void test_no_amx_movrs (void) __attribute__((__target__("no-amx-movrs")));
popcntintrin.h gfniintrin.h and mm_malloc.h are usable
with -O -std=c89 -pedantic-errors. */
/* { dg-do compile } */
-/* { dg-options "-O -std=c89 -pedantic-errors -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mamx-fp16 -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-transpose -mamx-fp8 -mmovrs -mamx-movrs" } */
+/* { dg-options "-O -std=c89 -pedantic-errors -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mamx-fp16 -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-fp8 -mmovrs -mamx-movrs" } */
#include <x86intrin.h>
/* { dg-do compile } */
-/* { dg-options "-O2 -Werror-implicit-function-declaration -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mavx512vp2intersect -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mcmpccxadd -mamx-fp16 -mprefetchi -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-transpose -mamx-fp8 -mmovrs -mamx-movrs" } */
+/* { dg-options "-O2 -Werror-implicit-function-declaration -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mavx512vp2intersect -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mcmpccxadd -mamx-fp16 -mprefetchi -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-fp8 -mmovrs -mamx-movrs" } */
/* { dg-add-options bind_pic_locally } */
#include <mm_malloc.h>
/* { dg-do compile } */
-/* { dg-options "-O0 -Werror-implicit-function-declaration -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mamx-fp16 -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-transpose -mamx-fp8 -mmovrs -mamx-movrs" } */
+/* { dg-options "-O0 -Werror-implicit-function-declaration -march=k8 -msse4a -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr -mxsaveopt -msha -mxsavec -mxsaves -mclflushopt -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mpconfig -mwbnoinvd -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavxifma -mavxvnniint8 -mavxneconvert -mamx-fp16 -mraoint -mamx-complex -mavxvnniint16 -msm3 -msha512 -msm4 -mavx10.2 -mamx-avx512 -mamx-tf32 -mamx-fp8 -mmovrs -mamx-movrs" } */
/* { dg-add-options bind_pic_locally } */
#include <mm_malloc.h>
#ifndef DIFFERENT_PRAGMAS
-#pragma GCC target ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,sha,gfni,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avxifma,avxvnniint8,avxneconvert,amx-fp16,raoint,amx-complex,avxvnniint16,sm3,sha512,sm4,avx10.2,amx-avx512,amx-tf32,amx-transpose,amx-fp8,movrs,amx-movrs")
+#pragma GCC target ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,sha,gfni,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avxifma,avxvnniint8,avxneconvert,amx-fp16,raoint,amx-complex,avxvnniint16,sm3,sha512,sm4,avx10.2,amx-avx512,amx-tf32,amx-fp8,movrs,amx-movrs")
#endif
/* Following intrinsics require immediate arguments. They
/* immintrin.h (AVX/AVX2/RDRND/FSGSBASE/F16C/RTM/AVX512F/SHA) */
#ifdef DIFFERENT_PRAGMAS
-#pragma GCC target ("avx,avx2,rdrnd,fsgsbase,f16c,rtm,sha,gfni,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avxifma,avxvnniint8,avxneconvert,amx-fp16,raoint,amx-complex,avxvnniint16,sm3,sha512,sm4,avx10.2,amx-avx512,amx-tf32,amx-transpose,amx-fp8,movrs,amx-movrs")
+#pragma GCC target ("avx,avx2,rdrnd,fsgsbase,f16c,rtm,sha,gfni,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avxifma,avxvnniint8,avxneconvert,amx-fp16,raoint,amx-complex,avxvnniint16,sm3,sha512,sm4,avx10.2,amx-avx512,amx-tf32,amx-fp8,movrs,amx-movrs")
#endif
#include <immintrin.h>
test_1 (_cvtss_sh, unsigned short, float, 1)
#define __builtin_ia32_minmaxsh_mask_round(A, B, C, D, E, F) __builtin_ia32_minmaxsh_mask_round (A, B, 100, D, E, 4)
#define __builtin_ia32_minmaxss_mask_round(A, B, C, D, E, F) __builtin_ia32_minmaxss_mask_round (A, B, 100, D, E, 4)
-#pragma GCC target ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,fma,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,sha,xsavec,xsaves,clflushopt,clwb,mwaitx,clzero,pku,sgx,rdpid,gfni,vpclmulqdq,pconfig,wbnoinvd,enqcmd,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avxifma,avxvnniint8,avxneconvert,cmpccxadd,amx-fp16,prefetchi,raoint,amx-complex,avxvnniint16,sm3,sha512,sm4,avx10.2,amx-avx512,amx-tf32,amx-transpose,amx-fp8,movrs,amx-movrs")
+#pragma GCC target ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,fma,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,sha,xsavec,xsaves,clflushopt,clwb,mwaitx,clzero,pku,sgx,rdpid,gfni,vpclmulqdq,pconfig,wbnoinvd,enqcmd,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avxifma,avxvnniint8,avxneconvert,cmpccxadd,amx-fp16,prefetchi,raoint,amx-complex,avxvnniint16,sm3,sha512,sm4,avx10.2,amx-avx512,amx-tf32,amx-fp8,movrs,amx-movrs")
#include <x86intrin.h>
} "-mamx-tf32" ]
}
-# Return 1 if amx-transpose instructions can be compiled.
-proc check_effective_target_amx_transpose { } {
- return [check_no_compiler_messages amx_transpose object {
- void
- foo ()
- {
- __asm__ volatile ("ttransposed\t%%tmm1, %%tmm2" ::);
- }
- } "-mamx-transpose" ]
-}
-
# Return 1 if amx-fp8 instructions can be compiled.
proc check_effective_target_amx_fp8 { } {
return [check_no_compiler_messages amx_fp8 object {
projects / thirdparty / gcc.git / commitdiff
