r285503 - [AVX-512] Remove masked 128/256-bit pmuludq/pmuldq builtins and replace them with unmasked builtins and a select.
Craig Topper via cfe-commits
cfe-commits at lists.llvm.org
Sat Oct 29 12:02:07 PDT 2016
Author: ctopper
Date: Sat Oct 29 14:02:07 2016
New Revision: 285503
URL: http://llvm.org/viewvc/llvm-project?rev=285503&view=rev
Log:
[AVX-512] Remove masked 128/256-bit pmuludq/pmuldq builtins and replace them with unmasked builtins and a select.
Modified:
cfe/trunk/include/clang/Basic/BuiltinsX86.def
cfe/trunk/lib/Headers/avx512vlintrin.h
cfe/trunk/test/CodeGen/avx512vl-builtins.c
Modified: cfe/trunk/include/clang/Basic/BuiltinsX86.def
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/BuiltinsX86.def?rev=285503&r1=285502&r2=285503&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/BuiltinsX86.def (original)
+++ cfe/trunk/include/clang/Basic/BuiltinsX86.def Sat Oct 29 14:02:07 2016
@@ -1076,11 +1076,6 @@ TARGET_BUILTIN(__builtin_ia32_ucmpd512_m
TARGET_BUILTIN(__builtin_ia32_ucmpq512_mask, "UcV8LLiV8LLiIiUc", "", "avx512f")
TARGET_BUILTIN(__builtin_ia32_ucmpw512_mask, "UiV32sV32sIiUi", "", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmuldq256_mask, "V4LLiV8iV8iV4LLiUc", "", "avx512vl")
-TARGET_BUILTIN(__builtin_ia32_pmuldq128_mask, "V2LLiV4iV4iV2LLiUc", "", "avx512vl")
-TARGET_BUILTIN(__builtin_ia32_pmuludq256_mask, "V4LLiV8iV8iV4LLiUc", "", "avx512vl")
-TARGET_BUILTIN(__builtin_ia32_pmuludq128_mask, "V2LLiV4iV4iV2LLiUc", "", "avx512vl")
-
TARGET_BUILTIN(__builtin_ia32_pabsb512_mask, "V64cV64cV64cULLi", "", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_pabsw512_mask, "V32sV32sV32sUi", "", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_packssdw512_mask, "V32sV16iV16iV32sUi", "", "avx512bw")
Modified: cfe/trunk/lib/Headers/avx512vlintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/avx512vlintrin.h?rev=285503&r1=285502&r2=285503&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/avx512vlintrin.h (original)
+++ cfe/trunk/lib/Headers/avx512vlintrin.h Sat Oct 29 14:02:07 2016
@@ -744,79 +744,67 @@ _mm_maskz_sub_epi64(__mmask8 __U, __m128
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_mul_epi32 (__m256i __W, __mmask8 __M, __m256i __X,
- __m256i __Y)
+_mm256_mask_mul_epi32(__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y)
{
- return (__m256i) __builtin_ia32_pmuldq256_mask ((__v8si) __X,
- (__v8si) __Y,
- (__v4di) __W, __M);
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__M,
+ (__v4di)_mm256_mul_epi32(__X, __Y),
+ (__v4di)__W);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_mul_epi32 (__mmask8 __M, __m256i __X, __m256i __Y)
+_mm256_maskz_mul_epi32(__mmask8 __M, __m256i __X, __m256i __Y)
{
- return (__m256i) __builtin_ia32_pmuldq256_mask ((__v8si) __X,
- (__v8si) __Y,
- (__v4di)
- _mm256_setzero_si256 (),
- __M);
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__M,
+ (__v4di)_mm256_mul_epi32(__X, __Y),
+ (__v4di)_mm256_setzero_si256());
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_mul_epi32 (__m128i __W, __mmask8 __M, __m128i __X,
- __m128i __Y)
+_mm_mask_mul_epi32(__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y)
{
- return (__m128i) __builtin_ia32_pmuldq128_mask ((__v4si) __X,
- (__v4si) __Y,
- (__v2di) __W, __M);
+ return (__m128i)__builtin_ia32_selectq_128((__mmask8)__M,
+ (__v2di)_mm_mul_epi32(__X, __Y),
+ (__v2di)__W);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_mul_epi32 (__mmask8 __M, __m128i __X, __m128i __Y)
+_mm_maskz_mul_epi32(__mmask8 __M, __m128i __X, __m128i __Y)
{
- return (__m128i) __builtin_ia32_pmuldq128_mask ((__v4si) __X,
- (__v4si) __Y,
- (__v2di)
- _mm_setzero_si128 (),
- __M);
+ return (__m128i)__builtin_ia32_selectq_128((__mmask8)__M,
+ (__v2di)_mm_mul_epi32(__X, __Y),
+ (__v2di)_mm_setzero_si128());
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_mul_epu32 (__m256i __W, __mmask8 __M, __m256i __X,
- __m256i __Y)
+_mm256_mask_mul_epu32(__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y)
{
- return (__m256i) __builtin_ia32_pmuludq256_mask ((__v8si) __X,
- (__v8si) __Y,
- (__v4di) __W, __M);
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__M,
+ (__v4di)_mm256_mul_epu32(__X, __Y),
+ (__v4di)__W);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_mul_epu32 (__mmask8 __M, __m256i __X, __m256i __Y)
+_mm256_maskz_mul_epu32(__mmask8 __M, __m256i __X, __m256i __Y)
{
- return (__m256i) __builtin_ia32_pmuludq256_mask ((__v8si) __X,
- (__v8si) __Y,
- (__v4di)
- _mm256_setzero_si256 (),
- __M);
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__M,
+ (__v4di)_mm256_mul_epu32(__X, __Y),
+ (__v4di)_mm256_setzero_si256());
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_mul_epu32 (__m128i __W, __mmask8 __M, __m128i __X,
- __m128i __Y)
+_mm_mask_mul_epu32(__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y)
{
- return (__m128i) __builtin_ia32_pmuludq128_mask ((__v4si) __X,
- (__v4si) __Y,
- (__v2di) __W, __M);
+ return (__m128i)__builtin_ia32_selectq_128((__mmask8)__M,
+ (__v2di)_mm_mul_epu32(__X, __Y),
+ (__v2di)__W);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_mul_epu32 (__mmask8 __M, __m128i __X, __m128i __Y)
+_mm_maskz_mul_epu32(__mmask8 __M, __m128i __X, __m128i __Y)
{
- return (__m128i) __builtin_ia32_pmuludq128_mask ((__v4si) __X,
- (__v4si) __Y,
- (__v2di)
- _mm_setzero_si128 (),
- __M);
+ return (__m128i)__builtin_ia32_selectq_128((__mmask8)__M,
+ (__v2di)_mm_mul_epu32(__X, __Y),
+ (__v2di)_mm_setzero_si128());
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
Modified: cfe/trunk/test/CodeGen/avx512vl-builtins.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/avx512vl-builtins.c?rev=285503&r1=285502&r2=285503&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/avx512vl-builtins.c (original)
+++ cfe/trunk/test/CodeGen/avx512vl-builtins.c Sat Oct 29 14:02:07 2016
@@ -727,13 +727,15 @@ __m128i test_mm_maskz_sub_epi64 (__mmask
__m256i test_mm256_mask_mul_epi32 (__m256i __W, __mmask8 __M, __m256i __X,
__m256i __Y) {
//CHECK-LABEL: @test_mm256_mask_mul_epi32
- //CHECK: @llvm.x86.avx512.mask.pmul.dq.256
+ //CHECK: @llvm.x86.avx2.pmul.dq
+ //CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_mask_mul_epi32(__W, __M, __X, __Y);
}
__m256i test_mm256_maskz_mul_epi32 (__mmask8 __M, __m256i __X, __m256i __Y) {
//CHECK-LABEL: @test_mm256_maskz_mul_epi32
- //CHECK: @llvm.x86.avx512.mask.pmul.dq.256
+ //CHECK: @llvm.x86.avx2.pmul.dq
+ //CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_maskz_mul_epi32(__M, __X, __Y);
}
@@ -741,39 +743,45 @@ __m256i test_mm256_maskz_mul_epi32 (__mm
__m128i test_mm_mask_mul_epi32 (__m128i __W, __mmask8 __M, __m128i __X,
__m128i __Y) {
//CHECK-LABEL: @test_mm_mask_mul_epi32
- //CHECK: @llvm.x86.avx512.mask.pmul.dq.128
+ //CHECK: @llvm.x86.sse41.pmuldq
+ //CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_mask_mul_epi32(__W, __M, __X, __Y);
}
__m128i test_mm_maskz_mul_epi32 (__mmask8 __M, __m128i __X, __m128i __Y) {
//CHECK-LABEL: @test_mm_maskz_mul_epi32
- //CHECK: @llvm.x86.avx512.mask.pmul.dq.128
+ //CHECK: @llvm.x86.sse41.pmuldq
+ //CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_maskz_mul_epi32(__M, __X, __Y);
}
__m256i test_mm256_mask_mul_epu32 (__m256i __W, __mmask8 __M, __m256i __X,
__m256i __Y) {
//CHECK-LABEL: @test_mm256_mask_mul_epu32
- //CHECK: @llvm.x86.avx512.mask.pmulu.dq.256
+ //CHECK: @llvm.x86.avx2.pmulu.dq
+ //CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_mask_mul_epu32(__W, __M, __X, __Y);
}
__m256i test_mm256_maskz_mul_epu32 (__mmask8 __M, __m256i __X, __m256i __Y) {
//CHECK-LABEL: @test_mm256_maskz_mul_epu32
- //CHECK: @llvm.x86.avx512.mask.pmulu.dq.256
+ //CHECK: @llvm.x86.avx2.pmulu.dq
+ //CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
return _mm256_maskz_mul_epu32(__M, __X, __Y);
}
__m128i test_mm_mask_mul_epu32 (__m128i __W, __mmask8 __M, __m128i __X,
__m128i __Y) {
//CHECK-LABEL: @test_mm_mask_mul_epu32
- //CHECK: @llvm.x86.avx512.mask.pmulu.dq.128
+ //CHECK: @llvm.x86.sse2.pmulu.dq
+ //CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_mask_mul_epu32(__W, __M, __X, __Y);
}
__m128i test_mm_maskz_mul_epu32 (__mmask8 __M, __m128i __X, __m128i __Y) {
//CHECK-LABEL: @test_mm_maskz_mul_epu32
- //CHECK: @llvm.x86.avx512.mask.pmulu.dq.128
+ //CHECK: @llvm.x86.sse2.pmulu.dq
+ //CHECK: select <2 x i1> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}
return _mm_maskz_mul_epu32(__M, __X, __Y);
}
More information about the cfe-commits
mailing list