1*67e74705SXin Li // RUN: %clang_cc1 %s -triple=x86_64-apple-darwin -target-feature +sse3 -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/sse3-intrinsics-fast-isel.ll
9*67e74705SXin Li
test_mm_addsub_pd(__m128d A,__m128d B)10*67e74705SXin Li __m128d test_mm_addsub_pd(__m128d A, __m128d B) {
11*67e74705SXin Li // CHECK-LABEL: test_mm_addsub_pd
12*67e74705SXin Li // CHECK: call <2 x double> @llvm.x86.sse3.addsub.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}})
13*67e74705SXin Li return _mm_addsub_pd(A, B);
14*67e74705SXin Li }
15*67e74705SXin Li
test_mm_addsub_ps(__m128 A,__m128 B)16*67e74705SXin Li __m128 test_mm_addsub_ps(__m128 A, __m128 B) {
17*67e74705SXin Li // CHECK-LABEL: test_mm_addsub_ps
18*67e74705SXin Li // CHECK: call <4 x float> @llvm.x86.sse3.addsub.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}})
19*67e74705SXin Li return _mm_addsub_ps(A, B);
20*67e74705SXin Li }
21*67e74705SXin Li
test_mm_hadd_pd(__m128d A,__m128d B)22*67e74705SXin Li __m128d test_mm_hadd_pd(__m128d A, __m128d B) {
23*67e74705SXin Li // CHECK-LABEL: test_mm_hadd_pd
24*67e74705SXin Li // CHECK: call <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}})
25*67e74705SXin Li return _mm_hadd_pd(A, B);
26*67e74705SXin Li }
27*67e74705SXin Li
test_mm_hadd_ps(__m128 A,__m128 B)28*67e74705SXin Li __m128 test_mm_hadd_ps(__m128 A, __m128 B) {
29*67e74705SXin Li // CHECK-LABEL: test_mm_hadd_ps
30*67e74705SXin Li // CHECK: call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}})
31*67e74705SXin Li return _mm_hadd_ps(A, B);
32*67e74705SXin Li }
33*67e74705SXin Li
test_mm_hsub_pd(__m128d A,__m128d B)34*67e74705SXin Li __m128d test_mm_hsub_pd(__m128d A, __m128d B) {
35*67e74705SXin Li // CHECK-LABEL: test_mm_hsub_pd
36*67e74705SXin Li // CHECK: call <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}})
37*67e74705SXin Li return _mm_hsub_pd(A, B);
38*67e74705SXin Li }
39*67e74705SXin Li
test_mm_hsub_ps(__m128 A,__m128 B)40*67e74705SXin Li __m128 test_mm_hsub_ps(__m128 A, __m128 B) {
41*67e74705SXin Li // CHECK-LABEL: test_mm_hsub_ps
42*67e74705SXin Li // CHECK: call <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}})
43*67e74705SXin Li return _mm_hsub_ps(A, B);
44*67e74705SXin Li }
45*67e74705SXin Li
test_mm_lddqu_si128(__m128i const * P)46*67e74705SXin Li __m128i test_mm_lddqu_si128(__m128i const* P) {
47*67e74705SXin Li // CHECK-LABEL: test_mm_lddqu_si128
48*67e74705SXin Li // CHECK: call <16 x i8> @llvm.x86.sse3.ldu.dq(i8* %{{.*}})
49*67e74705SXin Li return _mm_lddqu_si128(P);
50*67e74705SXin Li }
51*67e74705SXin Li
test_mm_loaddup_pd(double const * P)52*67e74705SXin Li __m128d test_mm_loaddup_pd(double const* P) {
53*67e74705SXin Li // CHECK-LABEL: test_mm_loaddup_pd
54*67e74705SXin Li // CHECK: load double*
55*67e74705SXin Li // CHECK: insertelement <2 x double> undef, double %{{.*}}, i32 0
56*67e74705SXin Li // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 1
57*67e74705SXin Li return _mm_loaddup_pd(P);
58*67e74705SXin Li }
59*67e74705SXin Li
test_mm_movedup_pd(__m128d A)60*67e74705SXin Li __m128d test_mm_movedup_pd(__m128d A) {
61*67e74705SXin Li // CHECK-LABEL: test_mm_movedup_pd
62*67e74705SXin Li // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x i32> zeroinitializer
63*67e74705SXin Li return _mm_movedup_pd(A);
64*67e74705SXin Li }
65*67e74705SXin Li
test_mm_movehdup_ps(__m128 A)66*67e74705SXin Li __m128 test_mm_movehdup_ps(__m128 A) {
67*67e74705SXin Li // CHECK-LABEL: test_mm_movehdup_ps
68*67e74705SXin Li // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x i32> <i32 1, i32 1, i32 3, i32 3>
69*67e74705SXin Li return _mm_movehdup_ps(A);
70*67e74705SXin Li }
71*67e74705SXin Li
test_mm_moveldup_ps(__m128 A)72*67e74705SXin Li __m128 test_mm_moveldup_ps(__m128 A) {
73*67e74705SXin Li // CHECK-LABEL: test_mm_moveldup_ps
74*67e74705SXin Li // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
75*67e74705SXin Li return _mm_moveldup_ps(A);
76*67e74705SXin Li }
77