replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[X86][AVX512] Converted the VPERMILPD/VPERMILPS intrinsics to generic IR 

llvm-svn: 274492
This commit is contained in:
Simon Pilgrim 2016-07-04 11:06:15 +00:00
parent 7cf00d191a
commit 17388f2569
2 changed files with 56 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

58
clang/lib/Headers/avx512fintrin.h
View File

@ -6540,34 +6540,56 @@ _mm512_mask2_permutex2var_epi64 (__m512i __A, __m512i __I,
}
#define _mm512_permute_pd(X, C) __extension__ ({ \
(__m512d)__builtin_ia32_vpermilpd512_mask((__v8df)(__m512d)(X), (int)(C), \
(__v8df)_mm512_undefined_pd(), \
(__mmask8)-1); })
(__m512d)__builtin_shufflevector((__v8df)(__m512d)(X), \
(__v8df)_mm512_setzero_pd(), \
0 + (((C) & 0x01) >> 0), \
0 + (((C) & 0x02) >> 1), \
2 + (((C) & 0x04) >> 2), \
2 + (((C) & 0x08) >> 3), \
4 + (((C) & 0x10) >> 4), \
4 + (((C) & 0x20) >> 5), \
6 + (((C) & 0x40) >> 6), \
6 + (((C) & 0x80) >> 7)); })
#define _mm512_mask_permute_pd(W, U, X, C) __extension__ ({ \
(__m512d)__builtin_ia32_vpermilpd512_mask((__v8df)(__m512d)(X), (int)(C), \
(__v8df)(__m512d)(W), \
(__mmask8)(U)); })
(__m512d)__builtin_ia32_selectpd_512((__mmask8)(U), \
(__v8df)_mm512_permute_pd((X), (C)), \
(__v8df)(__m512d)(W)); })
#define _mm512_maskz_permute_pd(U, X, C) __extension__ ({ \
(__m512d)__builtin_ia32_vpermilpd512_mask((__v8df)(__m512d)(X), (int)(C), \
(__v8df)_mm512_setzero_pd(), \
(__mmask8)(U)); })
(__m512d)__builtin_ia32_selectpd_512((__mmask8)(U), \
(__v8df)_mm512_permute_pd((X), (C)), \
(__v8df)_mm512_setzero_pd()); })
#define _mm512_permute_ps(X, C) __extension__ ({ \
(__m512)__builtin_ia32_vpermilps512_mask((__v16sf)(__m512)(X), (int)(C), \
(__v16sf)_mm512_undefined_ps(), \
(__mmask16)-1); })
(__m512)__builtin_shufflevector((__v16sf)(__m512)(X), \
(__v16sf)_mm512_setzero_ps(), \
0 + (((C) & 0x03) >> 0), \
0 + (((C) & 0x0c) >> 2), \
0 + (((C) & 0x30) >> 4), \
0 + (((C) & 0xc0) >> 6), \
4 + (((C) & 0x03) >> 0), \
4 + (((C) & 0x0c) >> 2), \
4 + (((C) & 0x30) >> 4), \
4 + (((C) & 0xc0) >> 6), \
8 + (((C) & 0x03) >> 0), \
8 + (((C) & 0x0c) >> 2), \
8 + (((C) & 0x30) >> 4), \
8 + (((C) & 0xc0) >> 6), \
12 + (((C) & 0x03) >> 0), \
12 + (((C) & 0x0c) >> 2), \
12 + (((C) & 0x30) >> 4), \
12 + (((C) & 0xc0) >> 6)); })
#define _mm512_mask_permute_ps(W, U, X, C) __extension__ ({ \
(__m512)__builtin_ia32_vpermilps512_mask((__v16sf)(__m512)(X), (int)(C), \
(__v16sf)(__m512)(W), \
(__mmask16)(U)); })
(__m512)__builtin_ia32_selectps_512((__mmask16)(U), \
(__v16sf)_mm512_permute_ps((X), (C)), \
(__v16sf)(__m512)(W)); })
#define _mm512_maskz_permute_ps(U, X, C) __extension__ ({ \
(__m512)__builtin_ia32_vpermilps512_mask((__v16sf)(__m512)(X), (int)(C), \
(__v16sf)_mm512_setzero_ps(), \
(__mmask16)(U)); })
(__m512)__builtin_ia32_selectps_512((__mmask16)(U), \
(__v16sf)_mm512_permute_ps((X), (C)), \
(__v16sf)_mm512_setzero_ps()); })
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_permutevar_pd (__m512d __A, __m512i __C)

28
clang/test/CodeGen/avx512f-builtins.c
View File

@ -3409,38 +3409,42 @@ __m512i test_mm512_mask2_permutex2var_epi64(__m512i __A, __m512i __I, __mmask8 _
__m512d test_mm512_permute_pd(__m512d __X) {
// CHECK-LABEL: @test_mm512_permute_pd
// CHECK: @llvm.x86.avx512.mask.vpermil.pd.512
return _mm512_permute_pd(__X, 2);
// CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
return _mm512_permute_pd(__X, 2);
}
__m512d test_mm512_mask_permute_pd(__m512d __W, __mmask8 __U, __m512d __X) {
// CHECK-LABEL: @test_mm512_mask_permute_pd
// CHECK: @llvm.x86.avx512.mask.vpermil.pd.512
return _mm512_mask_permute_pd(__W, __U, __X, 2);
// CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
// CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
return _mm512_mask_permute_pd(__W, __U, __X, 2);
}
__m512d test_mm512_maskz_permute_pd(__mmask8 __U, __m512d __X) {
// CHECK-LABEL: @test_mm512_maskz_permute_pd
// CHECK: @llvm.x86.avx512.mask.vpermil.pd.512
return _mm512_maskz_permute_pd(__U, __X, 2);
// CHECK: shufflevector <8 x double> %{{.*}}, <8 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
// CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
return _mm512_maskz_permute_pd(__U, __X, 2);
}
__m512 test_mm512_permute_ps(__m512 __X) {
// CHECK-LABEL: @test_mm512_permute_ps
// CHECK: @llvm.x86.avx512.mask.vpermil.ps.512
return _mm512_permute_ps(__X, 2);
// CHECK: shufflevector <16 x float> %{{.*}}, <16 x float> %{{.*}}, <16 x i32> <i32 2, i32 0, i32 0, i32 0, i32 6, i32 4, i32 4, i32 4, i32 10, i32 8, i32 8, i32 8, i32 14, i32 12, i32 12, i32 12>
return _mm512_permute_ps(__X, 2);
}
__m512 test_mm512_mask_permute_ps(__m512 __W, __mmask16 __U, __m512 __X) {
// CHECK-LABEL: @test_mm512_mask_permute_ps
// CHECK: @llvm.x86.avx512.mask.vpermil.ps.512
return _mm512_mask_permute_ps(__W, __U, __X, 2);
// CHECK: shufflevector <16 x float> %{{.*}}, <16 x float> %{{.*}}, <16 x i32> <i32 2, i32 0, i32 0, i32 0, i32 6, i32 4, i32 4, i32 4, i32 10, i32 8, i32 8, i32 8, i32 14, i32 12, i32 12, i32 12>
// CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
return _mm512_mask_permute_ps(__W, __U, __X, 2);
}
__m512 test_mm512_maskz_permute_ps(__mmask16 __U, __m512 __X) {
// CHECK-LABEL: @test_mm512_maskz_permute_ps
// CHECK: @llvm.x86.avx512.mask.vpermil.ps.512
return _mm512_maskz_permute_ps(__U, __X, 2);
// CHECK: shufflevector <16 x float> %{{.*}}, <16 x float> %{{.*}}, <16 x i32> <i32 2, i32 0, i32 0, i32 0, i32 6, i32 4, i32 4, i32 4, i32 10, i32 8, i32 8, i32 8, i32 14, i32 12, i32 12, i32 12>
// CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
return _mm512_maskz_permute_ps(__U, __X, 2);
}
__m512d test_mm512_permutevar_pd(__m512d __A, __m512i __C) {