From 0a70c3c7af8c579f1b487928b6e5b7089d539171 Mon Sep 17 00:00:00 2001 From: Craig Topper Date: Tue, 20 Feb 2018 07:28:18 +0000 Subject: [PATCH] [X86] Remove mask from 512 bit pmulhrsw/pmulhw/pmulhuw builtins. We now use a vselect node in IR around an unmasked builtin. This makes it consistent with the 128 and 256 bit versions. llvm-svn: 325560 --- clang/include/clang/Basic/BuiltinsX86.def | 6 +- clang/lib/Headers/avx512bwintrin.h | 75 +++++++++-------------- clang/test/CodeGen/avx512bw-builtins.c | 24 +++++--- 3 files changed, 47 insertions(+), 58 deletions(-) diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def index b7a058182c30..74c0c039c9d4 100644 --- a/clang/include/clang/Basic/BuiltinsX86.def +++ b/clang/include/clang/Basic/BuiltinsX86.def @@ -1138,9 +1138,9 @@ TARGET_BUILTIN(__builtin_ia32_vpermt2varhi128_maskz, "V8sV8sV8sV8sUc", "", "avx5 TARGET_BUILTIN(__builtin_ia32_vpermt2varhi256_mask, "V16sV16sV16sV16sUs", "", "avx512vl,avx512bw") TARGET_BUILTIN(__builtin_ia32_vpermt2varhi256_maskz, "V16sV16sV16sV16sUs", "", "avx512vl,avx512bw") -TARGET_BUILTIN(__builtin_ia32_pmulhrsw512_mask, "V32sV32sV32sV32sUi", "", "avx512bw") -TARGET_BUILTIN(__builtin_ia32_pmulhuw512_mask, "V32sV32sV32sV32sUi", "", "avx512bw") -TARGET_BUILTIN(__builtin_ia32_pmulhw512_mask, "V32sV32sV32sV32sUi", "", "avx512bw") +TARGET_BUILTIN(__builtin_ia32_pmulhrsw512, "V32sV32sV32s", "", "avx512bw") +TARGET_BUILTIN(__builtin_ia32_pmulhuw512, "V32sV32sV32s", "", "avx512bw") +TARGET_BUILTIN(__builtin_ia32_pmulhw512, "V32sV32sV32s", "", "avx512bw") TARGET_BUILTIN(__builtin_ia32_addpd512_mask, "V8dV8dV8dV8dUcIi", "", "avx512f") TARGET_BUILTIN(__builtin_ia32_addps512_mask, "V16fV16fV16fV16fUsIi", "", "avx512f") diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h index 064300a48798..2aa72144037c 100644 --- a/clang/lib/Headers/avx512bwintrin.h +++ b/clang/lib/Headers/avx512bwintrin.h @@ -1008,87 +1008,70 @@ _mm512_maskz_permutex2var_epi16 (__mmask32 __U, __m512i __A, } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mulhrs_epi16 (__m512i __A, __m512i __B) +_mm512_mulhrs_epi16(__m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_pmulhrsw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) _mm512_setzero_hi(), - (__mmask32) -1); + return (__m512i)__builtin_ia32_pmulhrsw512((__v32hi)__A, (__v32hi)__B); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mask_mulhrs_epi16 (__m512i __W, __mmask32 __U, __m512i __A, - __m512i __B) +_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); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_mulhrs_epi16(__A, __B), + (__v32hi)__W); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_maskz_mulhrs_epi16 (__mmask32 __U, __m512i __A, __m512i __B) +_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); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_mulhrs_epi16(__A, __B), + (__v32hi)_mm512_setzero_hi()); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mulhi_epi16 (__m512i __A, __m512i __B) +_mm512_mulhi_epi16(__m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_pmulhw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) _mm512_setzero_hi(), - (__mmask32) -1); + return (__m512i)__builtin_ia32_pmulhw512((__v32hi) __A, (__v32hi) __B); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mask_mulhi_epi16 (__m512i __W, __mmask32 __U, __m512i __A, +_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); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_mulhi_epi16(__A, __B), + (__v32hi)__W); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_maskz_mulhi_epi16 (__mmask32 __U, __m512i __A, __m512i __B) +_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); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_mulhi_epi16(__A, __B), + (__v32hi)_mm512_setzero_hi()); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mulhi_epu16 (__m512i __A, __m512i __B) +_mm512_mulhi_epu16(__m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_pmulhuw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) _mm512_setzero_hi(), - (__mmask32) -1); + return (__m512i)__builtin_ia32_pmulhuw512((__v32hi) __A, (__v32hi) __B); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mask_mulhi_epu16 (__m512i __W, __mmask32 __U, __m512i __A, - __m512i __B) +_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); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_mulhi_epu16(__A, __B), + (__v32hi)__W); } 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); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_mulhi_epu16(__A, __B), + (__v32hi)_mm512_setzero_hi()); } static __inline__ __m512i __DEFAULT_FN_ATTRS diff --git a/clang/test/CodeGen/avx512bw-builtins.c b/clang/test/CodeGen/avx512bw-builtins.c index 4288c086924b..3b601f479135 100644 --- a/clang/test/CodeGen/avx512bw-builtins.c +++ b/clang/test/CodeGen/avx512bw-builtins.c @@ -984,47 +984,53 @@ __m512i test_mm512_maskz_permutex2var_epi16(__mmask32 __U, __m512i __A, __m512i __m512i test_mm512_mulhrs_epi16(__m512i __A, __m512i __B) { // CHECK-LABEL: @test_mm512_mulhrs_epi16 - // CHECK: @llvm.x86.avx512.mask.pmul.hr.sw.512 + // CHECK: @llvm.x86.avx512.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 + // CHECK: @llvm.x86.avx512.pmul.hr.sw.512 + // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}} 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 + // CHECK: @llvm.x86.avx512.pmul.hr.sw.512 + // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}} 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 + // CHECK: @llvm.x86.avx512.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 + // CHECK: @llvm.x86.avx512.pmulh.w.512 + // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}} 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 + // CHECK: @llvm.x86.avx512.pmulh.w.512 + // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}} 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 + // CHECK: @llvm.x86.avx512.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 + // CHECK: @llvm.x86.avx512.pmulhu.w.512 + // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}} 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 + // CHECK: @llvm.x86.avx512.pmulhu.w.512 + // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}} return _mm512_maskz_mulhi_epu16(__U,__A,__B); } -- 2.34.1