1*67e74705SXin Li // RUN: %clang_cc1 %s -triple=x86_64-apple-darwin -target-feature +sse4a -emit-llvm -o - -Werror | FileCheck %s
2*67e74705SXin Li
3*67e74705SXin Li // Don't include mm_malloc.h, it's system specific.
4*67e74705SXin Li #define __MM_MALLOC_H
5*67e74705SXin Li
6*67e74705SXin Li #include <x86intrin.h>
7*67e74705SXin Li
8*67e74705SXin Li // NOTE: This should match the tests in llvm/test/CodeGen/X86/sse4a-intrinsics-fast-isel.ll
9*67e74705SXin Li
test_mm_extracti_si64(__m128i x)10*67e74705SXin Li __m128i test_mm_extracti_si64(__m128i x) {
11*67e74705SXin Li // CHECK-LABEL: test_mm_extracti_si64
12*67e74705SXin Li // CHECK: call <2 x i64> @llvm.x86.sse4a.extrqi(<2 x i64> %{{[^,]+}}, i8 3, i8 2)
13*67e74705SXin Li return _mm_extracti_si64(x, 3, 2);
14*67e74705SXin Li }
15*67e74705SXin Li
test_mm_extract_si64(__m128i x,__m128i y)16*67e74705SXin Li __m128i test_mm_extract_si64(__m128i x, __m128i y) {
17*67e74705SXin Li // CHECK-LABEL: test_mm_extract_si64
18*67e74705SXin Li // CHECK: call <2 x i64> @llvm.x86.sse4a.extrq(<2 x i64> %{{[^,]+}}, <16 x i8> %{{[^,]+}})
19*67e74705SXin Li return _mm_extract_si64(x, y);
20*67e74705SXin Li }
21*67e74705SXin Li
test_mm_inserti_si64(__m128i x,__m128i y)22*67e74705SXin Li __m128i test_mm_inserti_si64(__m128i x, __m128i y) {
23*67e74705SXin Li // CHECK-LABEL: test_mm_inserti_si64
24*67e74705SXin Li // CHECK: call <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64> %{{[^,]+}}, <2 x i64> %{{[^,]+}}, i8 5, i8 6)
25*67e74705SXin Li return _mm_inserti_si64(x, y, 5, 6);
26*67e74705SXin Li }
27*67e74705SXin Li
test_mm_insert_si64(__m128i x,__m128i y)28*67e74705SXin Li __m128i test_mm_insert_si64(__m128i x, __m128i y) {
29*67e74705SXin Li // CHECK-LABEL: test_mm_insert_si64
30*67e74705SXin Li // CHECK: call <2 x i64> @llvm.x86.sse4a.insertq(<2 x i64> %{{[^,]+}}, <2 x i64> %{{[^,]+}})
31*67e74705SXin Li return _mm_insert_si64(x, y);
32*67e74705SXin Li }
33*67e74705SXin Li
test_mm_stream_sd(double * p,__m128d a)34*67e74705SXin Li void test_mm_stream_sd(double *p, __m128d a) {
35*67e74705SXin Li // CHECK-LABEL: test_mm_stream_sd
36*67e74705SXin Li // CHECK: extractelement <2 x double> %{{.*}}, i64 0
37*67e74705SXin Li // CHECK: store double %{{.*}}, double* %{{.*}}, align 1, !nontemporal
38*67e74705SXin Li _mm_stream_sd(p, a);
39*67e74705SXin Li }
40*67e74705SXin Li
test_mm_stream_ss(float * p,__m128 a)41*67e74705SXin Li void test_mm_stream_ss(float *p, __m128 a) {
42*67e74705SXin Li // CHECK-LABEL: test_mm_stream_ss
43*67e74705SXin Li // CHECK: extractelement <4 x float> %{{.*}}, i64 0
44*67e74705SXin Li // CHECK: store float %{{.*}}, float* %{{.*}}, align 1, !nontemporal
45*67e74705SXin Li _mm_stream_ss(p, a);
46*67e74705SXin Li }
47