diff options
author | Hans Wennborg <hans@hanshq.net> | 2017-07-31 17:29:29 +0000 |
---|---|---|
committer | Hans Wennborg <hans@hanshq.net> | 2017-07-31 17:29:29 +0000 |
commit | 888cdb53e0f71201556f7d19d59c6bc3a582baf2 (patch) | |
tree | 7543977a30ddbc8c3b538074bc457a99a1cb4177 /test/CodeGen | |
parent | 6f5e1cc9f728bac74f436943ed9dab704c426ef5 (diff) |
Merging r309382:
------------------------------------------------------------------------
r309382 | rksimon | 2017-07-28 06:47:02 -0700 (Fri, 28 Jul 2017) | 3 lines
[X86] Add tests showing inability of vector non-temporal load/store intrinsic to force pointer alignment (PR33830)
Clang specifies a max type alignment of 16 bytes on darwin targets, meaning that the builtin nontemporal stores don't correctly align the loads/stores to 32 or 64 bytes when required, resulting in lowering to temporal unaligned loads/stores.
------------------------------------------------------------------------
Merging r309383:
------------------------------------------------------------------------
r309383 | rksimon | 2017-07-28 07:01:51 -0700 (Fri, 28 Jul 2017) | 1 line
Strip trailing whitespace. NFCI.
------------------------------------------------------------------------
Merging r309488:
------------------------------------------------------------------------
r309488 | rksimon | 2017-07-29 08:33:34 -0700 (Sat, 29 Jul 2017) | 7 lines
[X86][AVX] Ensure vector non-temporal load/store intrinsics force pointer alignment (PR33830)
Clang specifies a max type alignment of 16 bytes on darwin targets (annoyingly in the driver not via cc1), meaning that the builtin nontemporal stores don't correctly align the loads/stores to 32 or 64 bytes when required, resulting in lowering to temporal unaligned loads/stores.
This patch casts the vectors to explicitly aligned types prior to the load/store to ensure that the require alignment is respected.
Differential Revision: https://reviews.llvm.org/D35996
------------------------------------------------------------------------
git-svn-id: https://llvm.org/svn/llvm-project/cfe/branches/release_50@309588 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen')
-rw-r--r-- | test/CodeGen/x86-nontemporal.c | 93 |
1 files changed, 93 insertions, 0 deletions
diff --git a/test/CodeGen/x86-nontemporal.c b/test/CodeGen/x86-nontemporal.c new file mode 100644 index 0000000000..5e9e42c9f2 --- /dev/null +++ b/test/CodeGen/x86-nontemporal.c @@ -0,0 +1,93 @@ +// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +sse4.1 -target-feature +avx -target-feature +avx2 -target-feature +avx512f -emit-llvm -o - -Wall -Werror | FileCheck %s --check-prefix=CHECK +// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +sse4.1 -target-feature +avx -target-feature +avx2 -target-feature +avx512f -fno-signed-char -emit-llvm -o - -Wall -Werror | FileCheck %s --check-prefix=CHECK + +// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +sse4.1 -target-feature +avx -target-feature +avx2 -target-feature +avx512f -emit-llvm -o - -Wall -Werror -fmax-type-align=16 | FileCheck %s --check-prefix=CHECK +// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +sse4.1 -target-feature +avx -target-feature +avx2 -target-feature +avx512f -fno-signed-char -emit-llvm -o - -Wall -Werror -fmax-type-align=16 | FileCheck %s --check-prefix=CHECK + +#include <x86intrin.h> + +// (PR33830) Tests ensure the correct alignment of non-temporal load/stores on darwin targets where fmax-type-align is set to 16. + +// +// 128-bit vectors +// + +void test_mm_stream_pd(double* A, __m128d B) { + // CHECK-LABEL: test_mm_stream_pd + // CHECK: store <2 x double> %{{.*}}, <2 x double>* %{{.*}}, align 16, !nontemporal + _mm_stream_pd(A, B); +} + +void test_mm_stream_ps(float* A, __m128 B) { + // CHECK16-LABEL: test_mm_stream_ps + // CHECK16: store <4 x float> %{{.*}}, <4 x float>* %{{.*}}, align 16, !nontemporal + _mm_stream_ps(A, B); +} + +void test_mm_stream_si128(__m128i* A, __m128i B) { + // CHECK-LABEL: test_mm_stream_si128 + // CHECK: store <2 x i64> %{{.*}}, <2 x i64>* %{{.*}}, align 16, !nontemporal + _mm_stream_si128(A, B); +} + +__m128i test_mm_stream_load_si128(__m128i const *A) { + // CHECK-LABEL: test_mm_stream_load_si128 + // CHECK: load <2 x i64>, <2 x i64>* %{{.*}}, align 16, !nontemporal + return _mm_stream_load_si128(A); +} + +// +// 256-bit vectors +// + +void test_mm256_stream_pd(double* A, __m256d B) { + // CHECK-LABEL: test_mm256_stream_pd + // CHECK: store <4 x double> %{{.*}}, <4 x double>* %{{.*}}, align 32, !nontemporal + _mm256_stream_pd(A, B); +} + +void test_mm256_stream_ps(float* A, __m256 B) { + // CHECK-LABEL: test_mm256_stream_ps + // CHECK: store <8 x float> %{{.*}}, <8 x float>* %{{.*}}, align 32, !nontemporal + _mm256_stream_ps(A, B); +} + +void test_mm256_stream_si256(__m256i* A, __m256i B) { + // CHECK-LABEL: test_mm256_stream_si256 + // CHECK: store <4 x i64> %{{.*}}, <4 x i64>* %{{.*}}, align 32, !nontemporal + _mm256_stream_si256(A, B); +} + +__m256i test_mm256_stream_load_si256(__m256i const *A) { + // CHECK-LABEL: test_mm256_stream_load_si256 + // CHECK: load <4 x i64>, <4 x i64>* %{{.*}}, align 32, !nontemporal + return _mm256_stream_load_si256(A); +} + +// +// 512-bit vectors +// + +void test_mm512_stream_pd(double* A, __m512d B) { + // CHECK-LABEL: test_mm512_stream_pd + // CHECK: store <8 x double> %{{.*}}, <8 x double>* %{{.*}}, align 64, !nontemporal + _mm512_stream_pd(A, B); +} + +void test_mm512_stream_ps(float* A, __m512 B) { + // CHECK-LABEL: test_mm512_stream_ps + // CHECK: store <16 x float> %{{.*}}, <16 x float>* %{{.*}}, align 64, !nontemporal + _mm512_stream_ps(A, B); +} + +void test_mm512_stream_si512(__m512i* A, __m512i B) { + // CHECK-LABEL: test_mm512_stream_si512 + // CHECK: store <8 x i64> %{{.*}}, <8 x i64>* %{{.*}}, align 64, !nontemporal + _mm512_stream_si512(A, B); +} + +__m512i test_mm512_stream_load_si512(void *A) { + // CHECK-LABEL: test_mm512_stream_load_si512 + // CHECK: load <8 x i64>, <8 x i64>* %{{.*}}, align 64, !nontemporal + return _mm512_stream_load_si512(A); +} |