maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[AVX-512] Remove masking from 512-bit vpermil builtins. The backend now has versions without masking so wrap it with select. 

This will allow the backend to constant fold these to generic shuffle vectors like 128-bit and 256-bit without having to working about handling masking.

llvm-svn: 289351
This commit is contained in:
Craig Topper 2016-12-11 01:26:52 +00:00
parent 1f1b441267
commit 678b07fe3c
3 changed files with 32 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
clang/include/clang/Basic/BuiltinsX86.def
View File

@ -1489,8 +1489,8 @@ TARGET_BUILTIN(__builtin_ia32_vpermi2vard512_mask, "V16iV16iV16iV16iUs","","avx5
TARGET_BUILTIN(__builtin_ia32_vpermi2varpd512_mask, "V8dV8dV8LLiV8dUc","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermi2varps512_mask, "V16fV16fV16iV16fUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermi2varq512_mask, "V8LLiV8LLiV8LLiV8LLiUc","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermilvarpd512_mask, "V8dV8dV8LLiV8dUc","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermilvarps512_mask, "V16fV16fV16iV16fUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermilvarpd512, "V8dV8dV8LLi","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermilvarps512, "V16fV16fV16i","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermt2vard512_maskz, "V16iV16iV16iV16iUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermt2varpd512_maskz, "V8dV8LLiV8dV8dUc","","avx512f")
TARGET_BUILTIN(__builtin_ia32_vpermt2varps512_maskz, "V16fV16iV16fV16fUs","","avx512f")

54
clang/lib/Headers/avx512fintrin.h
View File

@ -6588,61 +6588,47 @@ _mm512_mask2_permutex2var_epi64 (__m512i __A, __m512i __I,
(__v16sf)_mm512_setzero_ps()); })
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_permutevar_pd (__m512d __A, __m512i __C)
_mm512_permutevar_pd(__m512d __A, __m512i __C)
{
return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A,
(__v8di) __C,
(__v8df)
_mm512_undefined_pd (),
(__mmask8) -1);
return (__m512d)__builtin_ia32_vpermilvarpd512((__v8df)__A, (__v8di)__C);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_permutevar_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512i __C)
_mm512_mask_permutevar_pd(__m512d __W, __mmask8 __U, __m512d __A, __m512i __C)
{
return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A,
(__v8di) __C,
(__v8df) __W,
(__mmask8) __U);
return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U,
(__v8df)_mm512_permutevar_pd(__A, __C),
(__v8df)__W);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_permutevar_pd (__mmask8 __U, __m512d __A, __m512i __C)
_mm512_maskz_permutevar_pd(__mmask8 __U, __m512d __A, __m512i __C)
{
return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A,
(__v8di) __C,
(__v8df)
_mm512_setzero_pd (),
(__mmask8) __U);
return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U,
(__v8df)_mm512_permutevar_pd(__A, __C),
(__v8df)_mm512_setzero_pd());
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_permutevar_ps (__m512 __A, __m512i __C)
_mm512_permutevar_ps(__m512 __A, __m512i __C)
{
return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A,
(__v16si) __C,
(__v16sf)
_mm512_undefined_ps (),
(__mmask16) -1);
return (__m512)__builtin_ia32_vpermilvarps512((__v16sf)__A, (__v16si)__C);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_permutevar_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512i __C)
_mm512_mask_permutevar_ps(__m512 __W, __mmask16 __U, __m512 __A, __m512i __C)
{
return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A,
(__v16si) __C,
(__v16sf) __W,
(__mmask16) __U);
return (__m512)__builtin_ia32_selectps_512((__mmask16)__U,
(__v16sf)_mm512_permutevar_ps(__A, __C),
(__v16sf)__W);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_permutevar_ps (__mmask16 __U, __m512 __A, __m512i __C)
_mm512_maskz_permutevar_ps(__mmask16 __U, __m512 __A, __m512i __C)
{
return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A,
(__v16si) __C,
(__v16sf)
_mm512_setzero_ps (),
(__mmask16) __U);
return (__m512)__builtin_ia32_selectps_512((__mmask16)__U,
(__v16sf)_mm512_permutevar_ps(__A, __C),
(__v16sf)_mm512_setzero_ps());
}
static __inline __m512d __DEFAULT_FN_ATTRS

16
clang/test/CodeGen/avx512f-builtins.c
View File

@ -3738,37 +3738,41 @@ __m512 test_mm512_maskz_permute_ps(__mmask16 __U, __m512 __X) {
__m512d test_mm512_permutevar_pd(__m512d __A, __m512i __C) {
// CHECK-LABEL: @test_mm512_permutevar_pd
// CHECK: @llvm.x86.avx512.mask.vpermilvar.pd.512
// CHECK: @llvm.x86.avx512.vpermilvar.pd.512
return _mm512_permutevar_pd(__A, __C);
}
__m512d test_mm512_mask_permutevar_pd(__m512d __W, __mmask8 __U, __m512d __A, __m512i __C) {
// CHECK-LABEL: @test_mm512_mask_permutevar_pd
// CHECK: @llvm.x86.avx512.mask.vpermilvar.pd.512
// CHECK: @llvm.x86.avx512.vpermilvar.pd.512
// CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
return _mm512_mask_permutevar_pd(__W, __U, __A, __C);
}
__m512d test_mm512_maskz_permutevar_pd(__mmask8 __U, __m512d __A, __m512i __C) {
// CHECK-LABEL: @test_mm512_maskz_permutevar_pd
// CHECK: @llvm.x86.avx512.mask.vpermilvar.pd.512
// CHECK: @llvm.x86.avx512.vpermilvar.pd.512
// CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
return _mm512_maskz_permutevar_pd(__U, __A, __C);
}
__m512 test_mm512_permutevar_ps(__m512 __A, __m512i __C) {
// CHECK-LABEL: @test_mm512_permutevar_ps
// CHECK: @llvm.x86.avx512.mask.vpermilvar.ps.512
// CHECK: @llvm.x86.avx512.vpermilvar.ps.512
return _mm512_permutevar_ps(__A, __C);
}
__m512 test_mm512_mask_permutevar_ps(__m512 __W, __mmask16 __U, __m512 __A, __m512i __C) {
// CHECK-LABEL: @test_mm512_mask_permutevar_ps
// CHECK: @llvm.x86.avx512.mask.vpermilvar.ps.512
// CHECK: @llvm.x86.avx512.vpermilvar.ps.512
// CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
return _mm512_mask_permutevar_ps(__W, __U, __A, __C);
}
__m512 test_mm512_maskz_permutevar_ps(__mmask16 __U, __m512 __A, __m512i __C) {
// CHECK-LABEL: @test_mm512_maskz_permutevar_ps
// CHECK: @llvm.x86.avx512.mask.vpermilvar.ps.512
// CHECK: @llvm.x86.avx512.vpermilvar.ps.512
// CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
return _mm512_maskz_permutevar_ps(__U, __A, __C);
}