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

[clang][Builtin][AVX512] Adding intrinsics for vmovshdup and vmovsldup instruction set 

Differential Revision: http://reviews.llvm.org/D19595

llvm-svn: 268196
This commit is contained in:
Michael Zuckerman 2016-05-01 14:43:43 +00:00
parent 110af565c7
commit ac1e519944
5 changed files with 210 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
clang/include/clang/Basic/BuiltinsX86.def
View File

@ -2224,6 +2224,12 @@ TARGET_BUILTIN(__builtin_ia32_compresssf512_mask, "V16fV16fV16fUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_compresssi512_mask, "V16iV16iV16iUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_cmpsd_mask, "UcV2dV2dIiUcIi","","avx512f")
TARGET_BUILTIN(__builtin_ia32_cmpss_mask, "UcV4fV4fIiUcIi","","avx512f")
TARGET_BUILTIN(__builtin_ia32_movshdup512_mask, "V16fV16fV16fUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_movsldup512_mask, "V16fV16fV16fUs","","avx512f")
TARGET_BUILTIN(__builtin_ia32_movshdup128_mask, "V4fV4fV4fUc","","avx512vl")
TARGET_BUILTIN(__builtin_ia32_movshdup256_mask, "V8fV8fV8fUc","","avx512vl")
TARGET_BUILTIN(__builtin_ia32_movsldup128_mask, "V4fV4fV4fUc","","avx512vl")
TARGET_BUILTIN(__builtin_ia32_movsldup256_mask, "V8fV8fV8fUc","","avx512vl")
#undef BUILTIN
#undef TARGET_BUILTIN

52
clang/lib/Headers/avx512fintrin.h
View File

@ -7681,6 +7681,58 @@ __builtin_ia32_cmpsd_mask ((__v2df)( __X),\
_MM_FROUND_CUR_DIRECTION);\
})
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_movehdup_ps (__m512 __A)
{
return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
(__v16sf)
_mm512_undefined_ps (),
(__mmask16) -1);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_movehdup_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
(__v16sf) __W,
(__mmask16) __U);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_movehdup_ps (__mmask16 __U, __m512 __A)
{
return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
(__v16sf)
_mm512_setzero_ps (),
(__mmask16) __U);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_moveldup_ps (__m512 __A)
{
return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
(__v16sf)
_mm512_undefined_ps (),
(__mmask16) -1);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_moveldup_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
(__v16sf) __W,
(__mmask16) __U);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_moveldup_ps (__mmask16 __U, __m512 __A)
{
return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
(__v16sf)
_mm512_setzero_ps (),
(__mmask16) __U);
}
#undef __DEFAULT_FN_ATTRS
#endif // __AVX512FINTRIN_H

68
clang/lib/Headers/avx512vlintrin.h
View File

@ -9293,6 +9293,74 @@ __builtin_ia32_alignq256_mask ((__v4di)( __A),\
(__mmask8)( __U));\
})
static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_mask_movehdup_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
return (__m128) __builtin_ia32_movshdup128_mask ((__v4sf) __A,
(__v4sf) __W,
(__mmask8) __U);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_maskz_movehdup_ps (__mmask8 __U, __m128 __A)
{
return (__m128) __builtin_ia32_movshdup128_mask ((__v4sf) __A,
(__v4sf)
_mm_setzero_ps (),
(__mmask8) __U);
}
static __inline__ __m256 __DEFAULT_FN_ATTRS
_mm256_mask_movehdup_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
return (__m256) __builtin_ia32_movshdup256_mask ((__v8sf) __A,
(__v8sf) __W,
(__mmask8) __U);
}
static __inline__ __m256 __DEFAULT_FN_ATTRS
_mm256_maskz_movehdup_ps (__mmask8 __U, __m256 __A)
{
return (__m256) __builtin_ia32_movshdup256_mask ((__v8sf) __A,
(__v8sf)
_mm256_setzero_ps (),
(__mmask8) __U);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_mask_moveldup_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
return (__m128) __builtin_ia32_movsldup128_mask ((__v4sf) __A,
(__v4sf) __W,
(__mmask8) __U);
}
static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_maskz_moveldup_ps (__mmask8 __U, __m128 __A)
{
return (__m128) __builtin_ia32_movsldup128_mask ((__v4sf) __A,
(__v4sf)
_mm_setzero_ps (),
(__mmask8) __U);
}
static __inline__ __m256 __DEFAULT_FN_ATTRS
_mm256_mask_moveldup_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
return (__m256) __builtin_ia32_movsldup256_mask ((__v8sf) __A,
(__v8sf) __W,
(__mmask8) __U);
}
static __inline__ __m256 __DEFAULT_FN_ATTRS
_mm256_maskz_moveldup_ps (__mmask8 __U, __m256 __A)
{
return (__m256) __builtin_ia32_movsldup256_mask ((__v8sf) __A,
(__v8sf)
_mm256_setzero_ps (),
(__mmask8) __U);
}
#undef __DEFAULT_FN_ATTRS
#undef __DEFAULT_FN_ATTRS_BOTH

36
clang/test/CodeGen/avx512f-builtins.c
View File

@ -5333,3 +5333,39 @@ __mmask8 test_mm_mask_cmp_sd_mask(__mmask8 __M, __m128d __X, __m128d __Y) {
// CHECK: @llvm.x86.avx512.mask.cmp
return _mm_mask_cmp_sd_mask(__M, __X, __Y, 5);
}
__m512 test_mm512_movehdup_ps(__m512 __A) {
// CHECK-LABEL: @test_mm512_movehdup_ps
// CHECK: @llvm.x86.avx512.mask.movshdup.512
return _mm512_movehdup_ps(__A);
}
__m512 test_mm512_mask_movehdup_ps(__m512 __W, __mmask16 __U, __m512 __A) {
// CHECK-LABEL: @test_mm512_mask_movehdup_ps
// CHECK: @llvm.x86.avx512.mask.movshdup.512
return _mm512_mask_movehdup_ps(__W, __U, __A);
}
__m512 test_mm512_maskz_movehdup_ps(__mmask16 __U, __m512 __A) {
// CHECK-LABEL: @test_mm512_maskz_movehdup_ps
// CHECK: @llvm.x86.avx512.mask.movshdup.512
return _mm512_maskz_movehdup_ps(__U, __A);
}
__m512 test_mm512_moveldup_ps(__m512 __A) {
// CHECK-LABEL: @test_mm512_moveldup_ps
// CHECK: @llvm.x86.avx512.mask.movsldup.512
return _mm512_moveldup_ps(__A);
}
__m512 test_mm512_mask_moveldup_ps(__m512 __W, __mmask16 __U, __m512 __A) {
// CHECK-LABEL: @test_mm512_mask_moveldup_ps
// CHECK: @llvm.x86.avx512.mask.movsldup.512
return _mm512_mask_moveldup_ps(__W, __U, __A);
}
__m512 test_mm512_maskz_moveldup_ps(__mmask16 __U, __m512 __A) {
// CHECK-LABEL: @test_mm512_maskz_moveldup_ps
// CHECK: @llvm.x86.avx512.mask.movsldup.512
return _mm512_maskz_moveldup_ps(__U, __A);
}

48
clang/test/CodeGen/avx512vl-builtins.c
View File

@ -6533,3 +6533,51 @@ __m256i test_mm256_maskz_alignr_epi64(__mmask8 __U, __m256i __A, __m256i __B) {
// CHECK: @llvm.x86.avx512.mask.valign.q.256
return _mm256_maskz_alignr_epi64(__U, __A, __B, 1);
}
__m128 test_mm_mask_movehdup_ps(__m128 __W, __mmask8 __U, __m128 __A) {
// CHECK-LABEL: @test_mm_mask_movehdup_ps
// CHECK: @llvm.x86.avx512.mask.movshdup.128
return _mm_mask_movehdup_ps(__W, __U, __A);
}
__m128 test_mm_maskz_movehdup_ps(__mmask8 __U, __m128 __A) {
// CHECK-LABEL: @test_mm_maskz_movehdup_ps
// CHECK: @llvm.x86.avx512.mask.movshdup.128
return _mm_maskz_movehdup_ps(__U, __A);
}
__m256 test_mm256_mask_movehdup_ps(__m256 __W, __mmask8 __U, __m256 __A) {
// CHECK-LABEL: @test_mm256_mask_movehdup_ps
// CHECK: @llvm.x86.avx512.mask.movshdup.256
return _mm256_mask_movehdup_ps(__W, __U, __A);
}
__m256 test_mm256_maskz_movehdup_ps(__mmask8 __U, __m256 __A) {
// CHECK-LABEL: @test_mm256_maskz_movehdup_ps
// CHECK: @llvm.x86.avx512.mask.movshdup.256
return _mm256_maskz_movehdup_ps(__U, __A);
}
__m128 test_mm_mask_moveldup_ps(__m128 __W, __mmask8 __U, __m128 __A) {
// CHECK-LABEL: @test_mm_mask_moveldup_ps
// CHECK: @llvm.x86.avx512.mask.movsldup.128
return _mm_mask_moveldup_ps(__W, __U, __A);
}
__m128 test_mm_maskz_moveldup_ps(__mmask8 __U, __m128 __A) {
// CHECK-LABEL: @test_mm_maskz_moveldup_ps
// CHECK: @llvm.x86.avx512.mask.movsldup.128
return _mm_maskz_moveldup_ps(__U, __A);
}
__m256 test_mm256_mask_moveldup_ps(__m256 __W, __mmask8 __U, __m256 __A) {
// CHECK-LABEL: @test_mm256_mask_moveldup_ps
// CHECK: @llvm.x86.avx512.mask.movsldup.256
return _mm256_mask_moveldup_ps(__W, __U, __A);
}
__m256 test_mm256_maskz_moveldup_ps(__mmask8 __U, __m256 __A) {
// CHECK-LABEL: @test_mm256_maskz_moveldup_ps
// CHECK: @llvm.x86.avx512.mask.movsldup.256
return _mm256_maskz_moveldup_ps(__U, __A);
}