1 // 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
2 // 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
3
4 // 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
5 // 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
6
7 #include <immintrin.h>
8
9 // (PR33830) Tests ensure the correct alignment of non-temporal load/stores on darwin targets where fmax-type-align is set to 16.
10
11 //
12 // 128-bit vectors
13 //
14
test_mm_stream_pd(double * A,__m128d B)15 void test_mm_stream_pd(double* A, __m128d B) {
16 // CHECK-LABEL: test_mm_stream_pd
17 // CHECK: store <2 x double> %{{.*}}, <2 x double>* %{{.*}}, align 16, !nontemporal
18 _mm_stream_pd(A, B);
19 }
20
test_mm_stream_ps(float * A,__m128 B)21 void test_mm_stream_ps(float* A, __m128 B) {
22 // CHECK16-LABEL: test_mm_stream_ps
23 // CHECK16: store <4 x float> %{{.*}}, <4 x float>* %{{.*}}, align 16, !nontemporal
24 _mm_stream_ps(A, B);
25 }
26
test_mm_stream_si128(__m128i * A,__m128i B)27 void test_mm_stream_si128(__m128i* A, __m128i B) {
28 // CHECK-LABEL: test_mm_stream_si128
29 // CHECK: store <2 x i64> %{{.*}}, <2 x i64>* %{{.*}}, align 16, !nontemporal
30 _mm_stream_si128(A, B);
31 }
32
test_mm_stream_load_si128(__m128i const * A)33 __m128i test_mm_stream_load_si128(__m128i const *A) {
34 // CHECK-LABEL: test_mm_stream_load_si128
35 // CHECK: load <2 x i64>, <2 x i64>* %{{.*}}, align 16, !nontemporal
36 return _mm_stream_load_si128(A);
37 }
38
39 //
40 // 256-bit vectors
41 //
42
test_mm256_stream_pd(double * A,__m256d B)43 void test_mm256_stream_pd(double* A, __m256d B) {
44 // CHECK-LABEL: test_mm256_stream_pd
45 // CHECK: store <4 x double> %{{.*}}, <4 x double>* %{{.*}}, align 32, !nontemporal
46 _mm256_stream_pd(A, B);
47 }
48
test_mm256_stream_ps(float * A,__m256 B)49 void test_mm256_stream_ps(float* A, __m256 B) {
50 // CHECK-LABEL: test_mm256_stream_ps
51 // CHECK: store <8 x float> %{{.*}}, <8 x float>* %{{.*}}, align 32, !nontemporal
52 _mm256_stream_ps(A, B);
53 }
54
test_mm256_stream_si256(__m256i * A,__m256i B)55 void test_mm256_stream_si256(__m256i* A, __m256i B) {
56 // CHECK-LABEL: test_mm256_stream_si256
57 // CHECK: store <4 x i64> %{{.*}}, <4 x i64>* %{{.*}}, align 32, !nontemporal
58 _mm256_stream_si256(A, B);
59 }
60
test_mm256_stream_load_si256(__m256i const * A)61 __m256i test_mm256_stream_load_si256(__m256i const *A) {
62 // CHECK-LABEL: test_mm256_stream_load_si256
63 // CHECK: load <4 x i64>, <4 x i64>* %{{.*}}, align 32, !nontemporal
64 return _mm256_stream_load_si256(A);
65 }
66
67 //
68 // 512-bit vectors
69 //
70
test_mm512_stream_pd(double * A,__m512d B)71 void test_mm512_stream_pd(double* A, __m512d B) {
72 // CHECK-LABEL: test_mm512_stream_pd
73 // CHECK: store <8 x double> %{{.*}}, <8 x double>* %{{.*}}, align 64, !nontemporal
74 _mm512_stream_pd(A, B);
75 }
76
test_mm512_stream_ps(float * A,__m512 B)77 void test_mm512_stream_ps(float* A, __m512 B) {
78 // CHECK-LABEL: test_mm512_stream_ps
79 // CHECK: store <16 x float> %{{.*}}, <16 x float>* %{{.*}}, align 64, !nontemporal
80 _mm512_stream_ps(A, B);
81 }
82
test_mm512_stream_si512(__m512i * A,__m512i B)83 void test_mm512_stream_si512(__m512i* A, __m512i B) {
84 // CHECK-LABEL: test_mm512_stream_si512
85 // CHECK: store <8 x i64> %{{.*}}, <8 x i64>* %{{.*}}, align 64, !nontemporal
86 _mm512_stream_si512(A, B);
87 }
88
test_mm512_stream_load_si512(void * A)89 __m512i test_mm512_stream_load_si512(void *A) {
90 // CHECK-LABEL: test_mm512_stream_load_si512
91 // CHECK: load <8 x i64>, <8 x i64>* %{{.*}}, align 64, !nontemporal
92 return _mm512_stream_load_si512(A);
93 }
94