r285505 - [AVX-512] Remove 128/256-bit masked pmulhrsw/pmulhuw/pmulhw builtins and use unmasked builtins and select instead.
Craig Topper via cfe-commits
cfe-commits at lists.llvm.org
Sat Oct 29 12:02:14 PDT 2016
Author: ctopper
Date: Sat Oct 29 14:02:14 2016
New Revision: 285505
URL: http://llvm.org/viewvc/llvm-project?rev=285505&view=rev
Log:
[AVX-512] Remove 128/256-bit masked pmulhrsw/pmulhuw/pmulhw builtins and use unmasked builtins and select instead.
Modified:
cfe/trunk/include/clang/Basic/BuiltinsX86.def
cfe/trunk/lib/Headers/avx512vlbwintrin.h
cfe/trunk/test/CodeGen/avx512vlbw-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=285505&r1=285504&r2=285505&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/BuiltinsX86.def (original)
+++ cfe/trunk/include/clang/Basic/BuiltinsX86.def Sat Oct 29 14:02:14 2016
@@ -1324,12 +1324,6 @@ TARGET_BUILTIN(__builtin_ia32_pmovuswb12
TARGET_BUILTIN(__builtin_ia32_pmovuswb256_mask, "V16cV16sV16cUs", "", "avx512vl,avx512bw")
TARGET_BUILTIN(__builtin_ia32_pmovwb128_mask, "V16cV8sV16cUc", "", "avx512vl,avx512bw")
TARGET_BUILTIN(__builtin_ia32_pmovwb256_mask, "V16cV16sV16cUs", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmulhrsw128_mask, "V8sV8sV8sV8sUc", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmulhrsw256_mask, "V16sV16sV16sV16sUs", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmulhuw128_mask, "V8sV8sV8sV8sUc", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmulhuw256_mask, "V16sV16sV16sV16sUs", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmulhw128_mask, "V8sV8sV8sV8sUc", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmulhw256_mask, "V16sV16sV16sV16sUs", "", "avx512vl,avx512bw")
TARGET_BUILTIN(__builtin_ia32_cvtpd2qq512_mask, "V8LLiV8dV8LLiUcIi", "", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvtpd2uqq512_mask, "V8LLiV8dV8LLiUcIi", "", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_cvtps2qq512_mask, "V8LLiV8fV8LLiUcIi", "", "avx512dq")
Modified: cfe/trunk/lib/Headers/avx512vlbwintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/avx512vlbwintrin.h?rev=285505&r1=285504&r2=285505&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/avx512vlbwintrin.h (original)
+++ cfe/trunk/lib/Headers/avx512vlbwintrin.h Sat Oct 29 14:02:14 2016
@@ -1892,104 +1892,89 @@ _mm256_mask_cvtusepi16_storeu_epi8 (void
{
__builtin_ia32_pmovuswb256mem_mask ((__v16qi*) __P, (__v16hi) __A, __M);
}
+
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_mulhrs_epi16 (__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
- return (__m128i) __builtin_ia32_pmulhrsw128_mask ((__v8hi) __X,
- (__v8hi) __Y,
- (__v8hi) __W,
- (__mmask8) __U);
+_mm_mask_mulhrs_epi16(__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
+ return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+ (__v8hi)_mm_mulhrs_epi16(__X, __Y),
+ (__v8hi)__W);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_mulhrs_epi16 (__mmask8 __U, __m128i __X, __m128i __Y) {
- return (__m128i) __builtin_ia32_pmulhrsw128_mask ((__v8hi) __X,
- (__v8hi) __Y,
- (__v8hi) _mm_setzero_si128(),
- (__mmask8) __U);
+_mm_maskz_mulhrs_epi16(__mmask8 __U, __m128i __X, __m128i __Y) {
+ return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+ (__v8hi)_mm_mulhrs_epi16(__X, __Y),
+ (__v8hi)_mm_setzero_si128());
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_mulhrs_epi16 (__m256i __W, __mmask16 __U, __m256i __X, __m256i __Y) {
- return (__m256i) __builtin_ia32_pmulhrsw256_mask ((__v16hi) __X,
- (__v16hi) __Y,
- (__v16hi) __W,
- (__mmask16) __U);
+_mm256_mask_mulhrs_epi16(__m256i __W, __mmask16 __U, __m256i __X, __m256i __Y) {
+ return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+ (__v16hi)_mm256_mulhrs_epi16(__X, __Y),
+ (__v16hi)__W);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_mulhrs_epi16 (__mmask16 __U, __m256i __X, __m256i __Y) {
- return (__m256i) __builtin_ia32_pmulhrsw256_mask ((__v16hi) __X,
- (__v16hi) __Y,
- (__v16hi) _mm256_setzero_si256(),
- (__mmask16) __U);
+_mm256_maskz_mulhrs_epi16(__mmask16 __U, __m256i __X, __m256i __Y) {
+ return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+ (__v16hi)_mm256_mulhrs_epi16(__X, __Y),
+ (__v16hi)_mm256_setzero_si256());
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_mulhi_epu16 (__m128i __W, __mmask8 __U, __m128i __A,
- __m128i __B) {
- return (__m128i) __builtin_ia32_pmulhuw128_mask ((__v8hi) __A,
- (__v8hi) __B,
- (__v8hi) __W,
- (__mmask8) __U);
+_mm_mask_mulhi_epu16(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+ return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+ (__v8hi)_mm_mulhi_epu16(__A, __B),
+ (__v8hi)__W);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_mulhi_epu16 (__mmask8 __U, __m128i __A, __m128i __B) {
- return (__m128i) __builtin_ia32_pmulhuw128_mask ((__v8hi) __A,
- (__v8hi) __B,
- (__v8hi) _mm_setzero_si128(),
- (__mmask8) __U);
+_mm_maskz_mulhi_epu16(__mmask8 __U, __m128i __A, __m128i __B) {
+ return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+ (__v8hi)_mm_mulhi_epu16(__A, __B),
+ (__v8hi)_mm_setzero_si128());
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_mulhi_epu16 (__m256i __W, __mmask16 __U, __m256i __A,
- __m256i __B) {
- return (__m256i) __builtin_ia32_pmulhuw256_mask ((__v16hi) __A,
- (__v16hi) __B,
- (__v16hi) __W,
- (__mmask16) __U);
+_mm256_mask_mulhi_epu16(__m256i __W, __mmask16 __U, __m256i __A, __m256i __B) {
+ return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+ (__v16hi)_mm256_mulhi_epu16(__A, __B),
+ (__v16hi)__W);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_mulhi_epu16 (__mmask16 __U, __m256i __A, __m256i __B) {
- return (__m256i) __builtin_ia32_pmulhuw256_mask ((__v16hi) __A,
- (__v16hi) __B,
- (__v16hi) _mm256_setzero_si256(),
- (__mmask16) __U);
+_mm256_maskz_mulhi_epu16(__mmask16 __U, __m256i __A, __m256i __B) {
+ return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+ (__v16hi)_mm256_mulhi_epu16(__A, __B),
+ (__v16hi)_mm256_setzero_si256());
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_mulhi_epi16 (__m128i __W, __mmask8 __U, __m128i __A,
- __m128i __B) {
- return (__m128i) __builtin_ia32_pmulhw128_mask ((__v8hi) __A,
- (__v8hi) __B,
- (__v8hi) __W,
- (__mmask8) __U);
+_mm_mask_mulhi_epi16(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+ return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+ (__v8hi)_mm_mulhi_epi16(__A, __B),
+ (__v8hi)__W);
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_mulhi_epi16 (__mmask8 __U, __m128i __A, __m128i __B) {
- return (__m128i) __builtin_ia32_pmulhw128_mask ((__v8hi) __A,
- (__v8hi) __B,
- (__v8hi) _mm_setzero_si128(),
- (__mmask8) __U);
+_mm_maskz_mulhi_epi16(__mmask8 __U, __m128i __A, __m128i __B) {
+ return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+ (__v8hi)_mm_mulhi_epi16(__A, __B),
+ (__v8hi)_mm_setzero_si128());
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_mulhi_epi16 (__m256i __W, __mmask16 __U, __m256i __A,
- __m256i __B) {
- return (__m256i) __builtin_ia32_pmulhw256_mask ((__v16hi) __A,
- (__v16hi) __B,
- (__v16hi) __W,
- (__mmask16) __U);
+_mm256_mask_mulhi_epi16(__m256i __W, __mmask16 __U, __m256i __A, __m256i __B) {
+ return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+ (__v16hi)_mm256_mulhi_epi16(__A, __B),
+ (__v16hi)__W);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_mulhi_epi16 (__mmask16 __U, __m256i __A, __m256i __B) {
- return (__m256i) __builtin_ia32_pmulhw256_mask ((__v16hi) __A,
- (__v16hi) __B,
- (__v16hi) _mm256_setzero_si256(),
- (__mmask16) __U);
+_mm256_maskz_mulhi_epi16(__mmask16 __U, __m256i __A, __m256i __B) {
+ return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+ (__v16hi)_mm256_mulhi_epi16(__A, __B),
+ (__v16hi)_mm256_setzero_si256());
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
Modified: cfe/trunk/test/CodeGen/avx512vlbw-builtins.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/avx512vlbw-builtins.c?rev=285505&r1=285504&r2=285505&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/avx512vlbw-builtins.c (original)
+++ cfe/trunk/test/CodeGen/avx512vlbw-builtins.c Sat Oct 29 14:02:14 2016
@@ -1746,73 +1746,85 @@ __m128i test_mm256_maskz_cvtepi16_epi8(_
__m128i test_mm_mask_mulhrs_epi16(__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
// CHECK-LABEL: @test_mm_mask_mulhrs_epi16
- // CHECK: @llvm.x86.avx512.mask.pmul.hr.sw.128
+ // CHECK: @llvm.x86.ssse3.pmul.hr.sw
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
return _mm_mask_mulhrs_epi16(__W, __U, __X, __Y);
}
__m128i test_mm_maskz_mulhrs_epi16(__mmask8 __U, __m128i __X, __m128i __Y) {
// CHECK-LABEL: @test_mm_maskz_mulhrs_epi16
- // CHECK: @llvm.x86.avx512.mask.pmul.hr.sw.128
+ // CHECK: @llvm.x86.ssse3.pmul.hr.sw
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
return _mm_maskz_mulhrs_epi16(__U, __X, __Y);
}
__m256i test_mm256_mask_mulhrs_epi16(__m256i __W, __mmask16 __U, __m256i __X, __m256i __Y) {
// CHECK-LABEL: @test_mm256_mask_mulhrs_epi16
- // CHECK: @llvm.x86.avx512.mask.pmul.hr.sw.256
+ // CHECK: @llvm.x86.avx2.pmul.hr.sw
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
return _mm256_mask_mulhrs_epi16(__W, __U, __X, __Y);
}
__m256i test_mm256_maskz_mulhrs_epi16(__mmask16 __U, __m256i __X, __m256i __Y) {
// CHECK-LABEL: @test_mm256_maskz_mulhrs_epi16
- // CHECK: @llvm.x86.avx512.mask.pmul.hr.sw.256
+ // CHECK: @llvm.x86.avx2.pmul.hr.sw
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
return _mm256_maskz_mulhrs_epi16(__U, __X, __Y);
}
__m128i test_mm_mask_mulhi_epu16(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
// CHECK-LABEL: @test_mm_mask_mulhi_epu16
- // CHECK: @llvm.x86.avx512.mask.pmulhu.w.128
+ // CHECK: @llvm.x86.sse2.pmulhu.w
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
return _mm_mask_mulhi_epu16(__W, __U, __A, __B);
}
__m128i test_mm_maskz_mulhi_epu16(__mmask8 __U, __m128i __A, __m128i __B) {
// CHECK-LABEL: @test_mm_maskz_mulhi_epu16
- // CHECK: @llvm.x86.avx512.mask.pmulhu.w.128
+ // CHECK: @llvm.x86.sse2.pmulhu.w
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
return _mm_maskz_mulhi_epu16(__U, __A, __B);
}
__m256i test_mm256_mask_mulhi_epu16(__m256i __W, __mmask16 __U, __m256i __A, __m256i __B) {
// CHECK-LABEL: @test_mm256_mask_mulhi_epu16
- // CHECK: @llvm.x86.avx512.mask.pmulhu.w.256
+ // CHECK: @llvm.x86.avx2.pmulhu.w
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
return _mm256_mask_mulhi_epu16(__W, __U, __A, __B);
}
__m256i test_mm256_maskz_mulhi_epu16(__mmask16 __U, __m256i __A, __m256i __B) {
// CHECK-LABEL: @test_mm256_maskz_mulhi_epu16
- // CHECK: @llvm.x86.avx512.mask.pmulhu.w.256
+ // CHECK: @llvm.x86.avx2.pmulhu.w
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
return _mm256_maskz_mulhi_epu16(__U, __A, __B);
}
__m128i test_mm_mask_mulhi_epi16(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
// CHECK-LABEL: @test_mm_mask_mulhi_epi16
- // CHECK: @llvm.x86.avx512.mask.pmulh.w.128
+ // CHECK: @llvm.x86.sse2.pmulh.w
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
return _mm_mask_mulhi_epi16(__W, __U, __A, __B);
}
__m128i test_mm_maskz_mulhi_epi16(__mmask8 __U, __m128i __A, __m128i __B) {
// CHECK-LABEL: @test_mm_maskz_mulhi_epi16
- // CHECK: @llvm.x86.avx512.mask.pmulh.w.128
+ // CHECK: @llvm.x86.sse2.pmulh.w
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
return _mm_maskz_mulhi_epi16(__U, __A, __B);
}
__m256i test_mm256_mask_mulhi_epi16(__m256i __W, __mmask16 __U, __m256i __A, __m256i __B) {
// CHECK-LABEL: @test_mm256_mask_mulhi_epi16
- // CHECK: @llvm.x86.avx512.mask.pmulh.w.256
+ // CHECK: @llvm.x86.avx2.pmulh.w
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
return _mm256_mask_mulhi_epi16(__W, __U, __A, __B);
}
__m256i test_mm256_maskz_mulhi_epi16(__mmask16 __U, __m256i __A, __m256i __B) {
// CHECK-LABEL: @test_mm256_maskz_mulhi_epi16
- // CHECK: @llvm.x86.avx512.mask.pmulh.w.256
+ // CHECK: @llvm.x86.avx2.pmulh.w
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
return _mm256_maskz_mulhi_epi16(__U, __A, __B);
}
More information about the cfe-commits
mailing list