Skip to content

Commit

Permalink
[X86][AVX512BW] add clang intrinsics for pmulhrsw / pmulhuw / pmulhw
Browse files Browse the repository at this point in the history
also made minor fix in "test_mm512_maskz_permutex2var_epi16"

Differential Revision: http://reviews.llvm.org/D11336

llvm-svn: 242635
  • Loading branch information
Asaf Badouh committed Jul 19, 2015
1 parent ca345e0 commit d4419ca
Show file tree
Hide file tree
Showing 3 changed files with 141 additions and 2 deletions.
4 changes: 3 additions & 1 deletion clang/include/clang/Basic/BuiltinsX86.def
Expand Up @@ -1217,5 +1217,7 @@ BUILTIN(__builtin_ia32_vpermt2varhi128_mask, "V8sV8sV8sV8sUc", "")
BUILTIN(__builtin_ia32_vpermt2varhi128_maskz, "V8sV8sV8sV8sUc", "")
BUILTIN(__builtin_ia32_vpermt2varhi256_mask, "V16sV16sV16sV16sUs", "")
BUILTIN(__builtin_ia32_vpermt2varhi256_maskz, "V16sV16sV16sV16sUs", "")

BUILTIN(__builtin_ia32_pmulhrsw512_mask, "V32sV32sV32sV32sUi", "")
BUILTIN(__builtin_ia32_pmulhuw512_mask, "V32sV32sV32sV32sUi", "")
BUILTIN(__builtin_ia32_pmulhw512_mask, "V32sV32sV32sV32sUi", "")
#undef BUILTIN
90 changes: 90 additions & 0 deletions clang/lib/Headers/avx512bwintrin.h
Expand Up @@ -1204,6 +1204,96 @@ _mm512_maskz_permutex2var_epi16 (__mmask32 __U, __m512i __A,
(__mmask32) __U);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mulhrs_epi16 (__m512i __A, __m512i __B)
{
return (__m512i) __builtin_ia32_pmulhrsw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi)
_mm512_setzero_hi (),
(__mmask32) -1);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_mulhrs_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
__m512i __B)
{
return (__m512i) __builtin_ia32_pmulhrsw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) __W,
(__mmask32) __U);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_mulhrs_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
return (__m512i) __builtin_ia32_pmulhrsw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi)
_mm512_setzero_hi (),
(__mmask32) __U);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mulhi_epi16 (__m512i __A, __m512i __B)
{
return (__m512i) __builtin_ia32_pmulhw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi)
_mm512_setzero_hi (),
(__mmask32) -1);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_mulhi_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
__m512i __B)
{
return (__m512i) __builtin_ia32_pmulhw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) __W,
(__mmask32) __U);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_mulhi_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
return (__m512i) __builtin_ia32_pmulhw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi)
_mm512_setzero_hi (),
(__mmask32) __U);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mulhi_epu16 (__m512i __A, __m512i __B)
{
return (__m512i) __builtin_ia32_pmulhuw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi)
_mm512_setzero_hi (),
(__mmask32) -1);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_mulhi_epu16 (__m512i __W, __mmask32 __U, __m512i __A,
__m512i __B)
{
return (__m512i) __builtin_ia32_pmulhuw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi) __W,
(__mmask32) __U);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_mulhi_epu16 (__mmask32 __U, __m512i __A, __m512i __B)
{
return (__m512i) __builtin_ia32_pmulhuw512_mask ((__v32hi) __A,
(__v32hi) __B,
(__v32hi)
_mm512_setzero_hi (),
(__mmask32) __U);
}

#define _mm512_cmp_epi8_mask(a, b, p) __extension__ ({ \
(__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \
(__v64qi)(__m512i)(b), \
Expand Down
49 changes: 48 additions & 1 deletion clang/test/CodeGen/avx512bw-builtins.c
Expand Up @@ -830,6 +830,53 @@ __m512i test_mm512_mask_permutex2var_epi16(__m512i __A, __mmask32 __U, __m512i _
}
__m512i test_mm512_maskz_permutex2var_epi16(__mmask32 __U, __m512i __A, __m512i __I, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_permutex2var_epi16
// CHECK: @llvm.x86.avx512.mask.vpermt2var.hi.512
// CHECK: @llvm.x86.avx512.maskz.vpermt2var.hi.512
return _mm512_maskz_permutex2var_epi16(__U,__A,__I,__B);
}

__m512i test_mm512_mulhrs_epi16(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mulhrs_epi16
// CHECK: @llvm.x86.avx512.mask.pmul.hr.sw.512
return _mm512_mulhrs_epi16(__A,__B);
}
__m512i test_mm512_mask_mulhrs_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mask_mulhrs_epi16
// CHECK: @llvm.x86.avx512.mask.pmul.hr.sw.512
return _mm512_mask_mulhrs_epi16(__W,__U,__A,__B);
}
__m512i test_mm512_maskz_mulhrs_epi16(__mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_mulhrs_epi16
// CHECK: @llvm.x86.avx512.mask.pmul.hr.sw.512
return _mm512_maskz_mulhrs_epi16(__U,__A,__B);
}
__m512i test_mm512_mulhi_epi16(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mulhi_epi16
// CHECK: @llvm.x86.avx512.mask.pmulh.w.512
return _mm512_mulhi_epi16(__A,__B);
}
__m512i test_mm512_mask_mulhi_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mask_mulhi_epi16
// CHECK: @llvm.x86.avx512.mask.pmulh.w.512
return _mm512_mask_mulhi_epi16(__W,__U,__A,__B);
}
__m512i test_mm512_maskz_mulhi_epi16(__mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_mulhi_epi16
// CHECK: @llvm.x86.avx512.mask.pmulh.w.512
return _mm512_maskz_mulhi_epi16(__U,__A,__B);
}
__m512i test_mm512_mulhi_epu16(__m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mulhi_epu16
// CHECK: @llvm.x86.avx512.mask.pmulhu.w.512
return _mm512_mulhi_epu16(__A,__B);
}
__m512i test_mm512_mask_mulhi_epu16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_mask_mulhi_epu16
// CHECK: @llvm.x86.avx512.mask.pmulhu.w.512
return _mm512_mask_mulhi_epu16(__W,__U,__A,__B);
}
__m512i test_mm512_maskz_mulhi_epu16(__mmask32 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: @test_mm512_maskz_mulhi_epu16
// CHECK: @llvm.x86.avx512.mask.pmulhu.w.512
return _mm512_maskz_mulhi_epu16(__U,__A,__B);
}

0 comments on commit d4419ca

Please sign in to comment.