From 0064d9a014c14decfacf4cba6086efd0ff77e14c Mon Sep 17 00:00:00 2001 From: Craig Topper Date: Wed, 30 Aug 2017 16:15:12 +0000 Subject: [PATCH] [X86] Implement broadcastf32x2 and broadcasti32x2 intrinsics using __builtin_shufflevector instead builtins This patch implements the broadcastf32x2/broadcasti32x2 intrinsics using __builtin_shufflevector. Differential Revision: https://reviews.llvm.org/D37287 git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@312135 91177308-0d34-0410-b5e6-96231b3b80d8 --- include/clang/Basic/BuiltinsX86.def | 5 --- lib/Headers/avx512dqintrin.h | 38 ++++++++++---------- lib/Headers/avx512vldqintrin.h | 54 ++++++++++++++--------------- test/CodeGen/avx512dq-builtins.c | 16 +++++---- test/CodeGen/avx512vldq-builtins.c | 24 ++++++++----- 5 files changed, 72 insertions(+), 65 deletions(-) diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def index f39048c1a5..46d71c3ecb 100644 --- a/include/clang/Basic/BuiltinsX86.def +++ b/include/clang/Basic/BuiltinsX86.def @@ -1596,11 +1596,6 @@ TARGET_BUILTIN(__builtin_ia32_broadcastmb128, "V2LLiUc","","avx512cd,avx512vl") TARGET_BUILTIN(__builtin_ia32_broadcastmb256, "V4LLiUc","","avx512cd,avx512vl") TARGET_BUILTIN(__builtin_ia32_broadcastmw128, "V4iUs","","avx512cd,avx512vl") TARGET_BUILTIN(__builtin_ia32_broadcastmw256, "V8iUs","","avx512cd,avx512vl") -TARGET_BUILTIN(__builtin_ia32_broadcastf32x2_512_mask, "V16fV4fV16fUs","","avx512dq") -TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_512_mask, "V16iV4iV16iUs","","avx512dq") -TARGET_BUILTIN(__builtin_ia32_broadcastf32x2_256_mask, "V8fV4fV8fUc","","avx512dq,avx512vl") -TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_128_mask, "V4iV4iV4iUc","","avx512dq,avx512vl") -TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_256_mask, "V8iV4iV8iUc","","avx512dq,avx512vl") TARGET_BUILTIN(__builtin_ia32_pbroadcastw512_gpr_mask, "V32shV32sUi","","avx512bw") TARGET_BUILTIN(__builtin_ia32_pbroadcastw256_gpr_mask, "V16shV16sUs","","avx512bw,avx512vl") TARGET_BUILTIN(__builtin_ia32_pbroadcastw128_gpr_mask, "V8ssV8sUc","","avx512bw,avx512vl") diff --git a/lib/Headers/avx512dqintrin.h b/lib/Headers/avx512dqintrin.h index 4fd1add773..2c431d9740 100644 --- a/lib/Headers/avx512dqintrin.h +++ b/lib/Headers/avx512dqintrin.h @@ -973,25 +973,26 @@ _mm512_movepi64_mask (__m512i __A) static __inline__ __m512 __DEFAULT_FN_ATTRS _mm512_broadcast_f32x2 (__m128 __A) { - return (__m512) __builtin_ia32_broadcastf32x2_512_mask ((__v4sf) __A, - (__v16sf)_mm512_undefined_ps(), - (__mmask16) -1); + return (__m512)__builtin_shufflevector((__v4sf)__A, + (__v4sf)_mm_undefined_ps(), + 0, 1, 0, 1, 0, 1, 0, 1, + 0, 1, 0, 1, 0, 1, 0, 1); } static __inline__ __m512 __DEFAULT_FN_ATTRS _mm512_mask_broadcast_f32x2 (__m512 __O, __mmask16 __M, __m128 __A) { - return (__m512) __builtin_ia32_broadcastf32x2_512_mask ((__v4sf) __A, - (__v16sf) - __O, __M); + return (__m512)__builtin_ia32_selectps_512((__mmask16)__M, + (__v16sf)_mm512_broadcast_f32x2(__A), + (__v16sf)__O); } static __inline__ __m512 __DEFAULT_FN_ATTRS _mm512_maskz_broadcast_f32x2 (__mmask16 __M, __m128 __A) { - return (__m512) __builtin_ia32_broadcastf32x2_512_mask ((__v4sf) __A, - (__v16sf)_mm512_setzero_ps (), - __M); + return (__m512)__builtin_ia32_selectps_512((__mmask16)__M, + (__v16sf)_mm512_broadcast_f32x2(__A), + (__v16sf)_mm512_setzero_ps()); } static __inline__ __m512 __DEFAULT_FN_ATTRS @@ -1044,25 +1045,26 @@ _mm512_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A) static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_broadcast_i32x2 (__m128i __A) { - return (__m512i) __builtin_ia32_broadcasti32x2_512_mask ((__v4si) __A, - (__v16si)_mm512_setzero_si512(), - (__mmask16) -1); + return (__m512i)__builtin_shufflevector((__v4si)__A, + (__v4si)_mm_undefined_si128(), + 0, 1, 0, 1, 0, 1, 0, 1, + 0, 1, 0, 1, 0, 1, 0, 1); } static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask_broadcast_i32x2 (__m512i __O, __mmask16 __M, __m128i __A) { - return (__m512i) __builtin_ia32_broadcasti32x2_512_mask ((__v4si) __A, - (__v16si) - __O, __M); + return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M, + (__v16si)_mm512_broadcast_i32x2(__A), + (__v16si)__O); } static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_broadcast_i32x2 (__mmask16 __M, __m128i __A) { - return (__m512i) __builtin_ia32_broadcasti32x2_512_mask ((__v4si) __A, - (__v16si)_mm512_setzero_si512 (), - __M); + return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M, + (__v16si)_mm512_broadcast_i32x2(__A), + (__v16si)_mm512_setzero_si512()); } static __inline__ __m512i __DEFAULT_FN_ATTRS diff --git a/lib/Headers/avx512vldqintrin.h b/lib/Headers/avx512vldqintrin.h index aecd7df34d..d80df9eaff 100644 --- a/lib/Headers/avx512vldqintrin.h +++ b/lib/Headers/avx512vldqintrin.h @@ -978,25 +978,25 @@ _mm256_movepi64_mask (__m256i __A) static __inline__ __m256 __DEFAULT_FN_ATTRS _mm256_broadcast_f32x2 (__m128 __A) { - return (__m256) __builtin_ia32_broadcastf32x2_256_mask ((__v4sf) __A, - (__v8sf)_mm256_undefined_ps(), - (__mmask8) -1); + return (__m256)__builtin_shufflevector((__v4sf)__A, + (__v4sf)_mm_undefined_ps(), + 0, 1, 0, 1, 0, 1, 0, 1); } static __inline__ __m256 __DEFAULT_FN_ATTRS _mm256_mask_broadcast_f32x2 (__m256 __O, __mmask8 __M, __m128 __A) { - return (__m256) __builtin_ia32_broadcastf32x2_256_mask ((__v4sf) __A, - (__v8sf) __O, - __M); + return (__m256)__builtin_ia32_selectps_256((__mmask8)__M, + (__v8sf)_mm256_broadcast_f32x2(__A), + (__v8sf)__O); } static __inline__ __m256 __DEFAULT_FN_ATTRS _mm256_maskz_broadcast_f32x2 (__mmask8 __M, __m128 __A) { - return (__m256) __builtin_ia32_broadcastf32x2_256_mask ((__v4sf) __A, - (__v8sf) _mm256_setzero_ps (), - __M); + return (__m256)__builtin_ia32_selectps_256((__mmask8)__M, + (__v8sf)_mm256_broadcast_f32x2(__A), + (__v8sf)_mm256_setzero_ps()); } static __inline__ __m256d __DEFAULT_FN_ATTRS @@ -1025,49 +1025,49 @@ _mm256_maskz_broadcast_f64x2 (__mmask8 __M, __m128d __A) static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_broadcast_i32x2 (__m128i __A) { - return (__m128i) __builtin_ia32_broadcasti32x2_128_mask ((__v4si) __A, - (__v4si)_mm_undefined_si128(), - (__mmask8) -1); + return (__m128i)__builtin_shufflevector((__v4si)__A, + (__v4si)_mm_undefined_si128(), + 0, 1, 0, 1); } static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_mask_broadcast_i32x2 (__m128i __O, __mmask8 __M, __m128i __A) { - return (__m128i) __builtin_ia32_broadcasti32x2_128_mask ((__v4si) __A, - (__v4si) __O, - __M); + return (__m128i)__builtin_ia32_selectd_128((__mmask8)__M, + (__v4si)_mm_broadcast_i32x2(__A), + (__v4si)__O); } static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_maskz_broadcast_i32x2 (__mmask8 __M, __m128i __A) { - return (__m128i) __builtin_ia32_broadcasti32x2_128_mask ((__v4si) __A, - (__v4si) _mm_setzero_si128 (), - __M); + return (__m128i)__builtin_ia32_selectd_128((__mmask8)__M, + (__v4si)_mm_broadcast_i32x2(__A), + (__v4si)_mm_setzero_si128()); } static __inline__ __m256i __DEFAULT_FN_ATTRS _mm256_broadcast_i32x2 (__m128i __A) { - return (__m256i) __builtin_ia32_broadcasti32x2_256_mask ((__v4si) __A, - (__v8si)_mm256_undefined_si256(), - (__mmask8) -1); + return (__m256i)__builtin_shufflevector((__v4si)__A, + (__v4si)_mm_undefined_si128(), + 0, 1, 0, 1, 0, 1, 0, 1); } static __inline__ __m256i __DEFAULT_FN_ATTRS _mm256_mask_broadcast_i32x2 (__m256i __O, __mmask8 __M, __m128i __A) { - return (__m256i) __builtin_ia32_broadcasti32x2_256_mask ((__v4si) __A, - (__v8si) __O, - __M); + return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M, + (__v8si)_mm256_broadcast_i32x2(__A), + (__v8si)__O); } static __inline__ __m256i __DEFAULT_FN_ATTRS _mm256_maskz_broadcast_i32x2 (__mmask8 __M, __m128i __A) { - return (__m256i) __builtin_ia32_broadcasti32x2_256_mask ((__v4si) __A, - (__v8si) _mm256_setzero_si256 (), - __M); + return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M, + (__v8si)_mm256_broadcast_i32x2(__A), + (__v8si)_mm256_setzero_si256()); } static __inline__ __m256i __DEFAULT_FN_ATTRS diff --git a/test/CodeGen/avx512dq-builtins.c b/test/CodeGen/avx512dq-builtins.c index ca8566c597..1b21ca3c43 100644 --- a/test/CodeGen/avx512dq-builtins.c +++ b/test/CodeGen/avx512dq-builtins.c @@ -949,19 +949,21 @@ __mmask8 test_mm512_movepi64_mask(__m512i __A) { __m512 test_mm512_broadcast_f32x2(__m128 __A) { // CHECK-LABEL: @test_mm512_broadcast_f32x2 - // CHECK: @llvm.x86.avx512.mask.broadcastf32x2 + // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> zeroinitializer, <16 x i32> return _mm512_broadcast_f32x2(__A); } __m512 test_mm512_mask_broadcast_f32x2(__m512 __O, __mmask16 __M, __m128 __A) { // CHECK-LABEL: @test_mm512_mask_broadcast_f32x2 - // CHECK: @llvm.x86.avx512.mask.broadcastf32x2 + // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> zeroinitializer, <16 x i32> + // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}} return _mm512_mask_broadcast_f32x2(__O, __M, __A); } __m512 test_mm512_maskz_broadcast_f32x2(__mmask16 __M, __m128 __A) { // CHECK-LABEL: @test_mm512_maskz_broadcast_f32x2 - // CHECK: @llvm.x86.avx512.mask.broadcastf32x2 + // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> zeroinitializer, <16 x i32> + // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}} return _mm512_maskz_broadcast_f32x2(__M, __A); } @@ -1007,19 +1009,21 @@ __m512d test_mm512_maskz_broadcast_f64x2(__mmask8 __M, double const* __A) { __m512i test_mm512_broadcast_i32x2(__m128i __A) { // CHECK-LABEL: @test_mm512_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <16 x i32> return _mm512_broadcast_i32x2(__A); } __m512i test_mm512_mask_broadcast_i32x2(__m512i __O, __mmask16 __M, __m128i __A) { // CHECK-LABEL: @test_mm512_mask_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <16 x i32> + // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}} return _mm512_mask_broadcast_i32x2(__O, __M, __A); } __m512i test_mm512_maskz_broadcast_i32x2(__mmask16 __M, __m128i __A) { // CHECK-LABEL: @test_mm512_maskz_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <16 x i32> + // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}} return _mm512_maskz_broadcast_i32x2(__M, __A); } diff --git a/test/CodeGen/avx512vldq-builtins.c b/test/CodeGen/avx512vldq-builtins.c index b18c811f84..3ca4b2135e 100644 --- a/test/CodeGen/avx512vldq-builtins.c +++ b/test/CodeGen/avx512vldq-builtins.c @@ -909,19 +909,21 @@ __mmask8 test_mm256_movepi64_mask(__m256i __A) { __m256 test_mm256_broadcast_f32x2(__m128 __A) { // CHECK-LABEL: @test_mm256_broadcast_f32x2 - // CHECK: @llvm.x86.avx512.mask.broadcastf32x2 + // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> zeroinitializer, <8 x i32> return _mm256_broadcast_f32x2(__A); } __m256 test_mm256_mask_broadcast_f32x2(__m256 __O, __mmask8 __M, __m128 __A) { // CHECK-LABEL: @test_mm256_mask_broadcast_f32x2 - // CHECK: @llvm.x86.avx512.mask.broadcastf32x2 + // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> zeroinitializer, <8 x i32> + // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}} return _mm256_mask_broadcast_f32x2(__O, __M, __A); } __m256 test_mm256_maskz_broadcast_f32x2(__mmask8 __M, __m128 __A) { // CHECK-LABEL: @test_mm256_maskz_broadcast_f32x2 - // CHECK: @llvm.x86.avx512.mask.broadcastf32x2 + // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> zeroinitializer, <8 x i32> + // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}} return _mm256_maskz_broadcast_f32x2(__M, __A); } @@ -947,37 +949,41 @@ __m256d test_mm256_maskz_broadcast_f64x2(__mmask8 __M, double const* __A) { __m128i test_mm_broadcast_i32x2(__m128i __A) { // CHECK-LABEL: @test_mm_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <4 x i32> return _mm_broadcast_i32x2(__A); } __m128i test_mm_mask_broadcast_i32x2(__m128i __O, __mmask8 __M, __m128i __A) { // CHECK-LABEL: @test_mm_mask_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <4 x i32> + // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}} return _mm_mask_broadcast_i32x2(__O, __M, __A); } __m128i test_mm_maskz_broadcast_i32x2(__mmask8 __M, __m128i __A) { // CHECK-LABEL: @test_mm_maskz_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <4 x i32> + // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}} return _mm_maskz_broadcast_i32x2(__M, __A); } __m256i test_mm256_broadcast_i32x2(__m128i __A) { // CHECK-LABEL: @test_mm256_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <8 x i32> return _mm256_broadcast_i32x2(__A); } __m256i test_mm256_mask_broadcast_i32x2(__m256i __O, __mmask8 __M, __m128i __A) { // CHECK-LABEL: @test_mm256_mask_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <8 x i32> + // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}} return _mm256_mask_broadcast_i32x2(__O, __M, __A); } __m256i test_mm256_maskz_broadcast_i32x2(__mmask8 __M, __m128i __A) { // CHECK-LABEL: @test_mm256_maskz_broadcast_i32x2 - // CHECK: @llvm.x86.avx512.mask.broadcasti32x2 + // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> zeroinitializer, <8 x i32> + // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}} return _mm256_maskz_broadcast_i32x2(__M, __A); } -- 2.50.1