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

[Clang][Builtin][AVX512] Adding intrisnics for the vpconflict{q|d} instruction set 

Differential Revision: http://reviews.llvm.org/D19525

llvm-svn: 267728
This commit is contained in:
Michael Zuckerman 2016-04-27 15:35:13 +00:00
parent 3f61a183a8
commit 514f05543f
3 changed files with 174 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -1245,6 +1245,10 @@ TARGET_BUILTIN(__builtin_ia32_vpermi2varhi512_mask, "V32sV32sV32sV32sUi", "", "a
TARGET_BUILTIN(__builtin_ia32_vpermt2varhi512_mask, "V32sV32sV32sV32sUi", "", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_vpermt2varhi512_maskz, "V32sV32sV32sV32sUi", "", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_vpconflictdi_128_mask, "V2LLiV2LLiV2LLiUc","","avx512cd,avx512vl")
TARGET_BUILTIN(__builtin_ia32_vpconflictdi_256_mask, "V4LLiV4LLiV4LLiUc","","avx512cd,avx512vl")
TARGET_BUILTIN(__builtin_ia32_vpconflictsi_128_mask, "V4iV4iV4iUc","","avx512cd,avx512vl")
TARGET_BUILTIN(__builtin_ia32_vpconflictsi_256_mask, "V8iV8iV8iUc","","avx512cd,avx512vl")
TARGET_BUILTIN(__builtin_ia32_vpconflictdi_512_mask, "V8LLiV8LLiV8LLiUc", "", "avx512cd")
TARGET_BUILTIN(__builtin_ia32_vpconflictsi_512_mask, "V16iV16iV16iUs", "", "avx512cd")
TARGET_BUILTIN(__builtin_ia32_vplzcntd_512_mask, "V16iV16iV16iUs", "", "avx512cd")

98
clang/lib/Headers/avx512vlcdintrin.h
View File

@ -56,6 +56,104 @@ _mm256_broadcastmw_epi32 (__mmask16 __A)
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_conflict_epi64 (__m128i __A)
{
return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
(__v2di) _mm_undefined_si128 (),
(__mmask8) -1);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_mask_conflict_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
(__v2di) __W,
(__mmask8) __U);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_maskz_conflict_epi64 (__mmask8 __U, __m128i __A)
{
return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
(__v2di)
_mm_setzero_di (),
(__mmask8) __U);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm256_conflict_epi64 (__m256i __A)
{
return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
(__v4di) _mm256_undefined_si256 (),
(__mmask8) -1);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm256_mask_conflict_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
{
return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
(__v4di) __W,
(__mmask8) __U);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm256_maskz_conflict_epi64 (__mmask8 __U, __m256i __A)
{
return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
(__v4di) _mm256_setzero_si256 (),
(__mmask8) __U);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_conflict_epi32 (__m128i __A)
{
return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A,
(__v4si) _mm_undefined_si128 (),
(__mmask8) -1);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_mask_conflict_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A,
(__v4si) __W,
(__mmask8) __U);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
_mm_maskz_conflict_epi32 (__mmask8 __U, __m128i __A)
{
return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A,
(__v4si) _mm_setzero_si128 (),
(__mmask8) __U);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm256_conflict_epi32 (__m256i __A)
{
return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
(__v8si) _mm256_undefined_si256 (),
(__mmask8) -1);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm256_mask_conflict_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
{
return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
(__v8si) __W,
(__mmask8) __U);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm256_maskz_conflict_epi32 (__mmask8 __U, __m256i __A)
{
return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
(__v8si)
_mm256_setzero_si256 (),
(__mmask8) __U);
}
#undef __DEFAULT_FN_ATTRS
#endif /* __AVX512VLCDINTRIN_H */

72
clang/test/CodeGen/avx512vlcd-builtins.c
View File

@ -28,3 +28,75 @@ __m256i test_mm256_broadcastmw_epi32(__mmask16 __A) {
// CHECK: @llvm.x86.avx512.broadcastmw.256
return _mm256_broadcastmw_epi32(__A);
}
__m128i test_mm_conflict_epi64(__m128i __A) {
// CHECK-LABEL: @test_mm_conflict_epi64
// CHECK: @llvm.x86.avx512.mask.conflict.q.128
return _mm_conflict_epi64(__A);
}
__m128i test_mm_mask_conflict_epi64(__m128i __W, __mmask8 __U, __m128i __A) {
// CHECK-LABEL: @test_mm_mask_conflict_epi64
// CHECK: @llvm.x86.avx512.mask.conflict.q.128
return _mm_mask_conflict_epi64(__W, __U, __A);
}
__m128i test_mm_maskz_conflict_epi64(__mmask8 __U, __m128i __A) {
// CHECK-LABEL: @test_mm_maskz_conflict_epi64
// CHECK: @llvm.x86.avx512.mask.conflict.q.128
return _mm_maskz_conflict_epi64(__U, __A);
}
__m256i test_mm256_conflict_epi64(__m256i __A) {
// CHECK-LABEL: @test_mm256_conflict_epi64
// CHECK: @llvm.x86.avx512.mask.conflict.q.256
return _mm256_conflict_epi64(__A);
}
__m256i test_mm256_mask_conflict_epi64(__m256i __W, __mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_mask_conflict_epi64
// CHECK: @llvm.x86.avx512.mask.conflict.q.256
return _mm256_mask_conflict_epi64(__W, __U, __A);
}
__m256i test_mm256_maskz_conflict_epi64(__mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_maskz_conflict_epi64
// CHECK: @llvm.x86.avx512.mask.conflict.q.256
return _mm256_maskz_conflict_epi64(__U, __A);
}
__m128i test_mm_conflict_epi32(__m128i __A) {
// CHECK-LABEL: @test_mm_conflict_epi32
// CHECK: @llvm.x86.avx512.mask.conflict.d.128
return _mm_conflict_epi32(__A);
}
__m128i test_mm_mask_conflict_epi32(__m128i __W, __mmask8 __U, __m128i __A) {
// CHECK-LABEL: @test_mm_mask_conflict_epi32
// CHECK: @llvm.x86.avx512.mask.conflict.d.128
return _mm_mask_conflict_epi32(__W, __U, __A);
}
__m128i test_mm_maskz_conflict_epi32(__mmask8 __U, __m128i __A) {
// CHECK-LABEL: @test_mm_maskz_conflict_epi32
// CHECK: @llvm.x86.avx512.mask.conflict.d.128
return _mm_maskz_conflict_epi32(__U, __A);
}
__m256i test_mm256_conflict_epi32(__m256i __A) {
// CHECK-LABEL: @test_mm256_conflict_epi32
// CHECK: @llvm.x86.avx512.mask.conflict.d.256
return _mm256_conflict_epi32(__A);
}
__m256i test_mm256_mask_conflict_epi32(__m256i __W, __mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_mask_conflict_epi32
// CHECK: @llvm.x86.avx512.mask.conflict.d.256
return _mm256_mask_conflict_epi32(__W, __U, __A);
}
__m256i test_mm256_maskz_conflict_epi32(__mmask8 __U, __m256i __A) {
// CHECK-LABEL: @test_mm256_maskz_conflict_epi32
// CHECK: @llvm.x86.avx512.mask.conflict.d.256
return _mm256_maskz_conflict_epi32(__U, __A);
}