1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -fast-isel -mtriple=i386-unknown-unknown -mattr=+sse3 | FileCheck %s --check-prefix=ALL --check-prefix=X32 3; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+sse3 | FileCheck %s --check-prefix=ALL --check-prefix=X64 4 5; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/sse3-builtins.c 6 7define <2 x double> @test_mm_addsub_pd(<2 x double> %a0, <2 x double> %a1) { 8; X32-LABEL: test_mm_addsub_pd: 9; X32: # BB#0: 10; X32-NEXT: addsubpd %xmm1, %xmm0 11; X32-NEXT: retl 12; 13; X64-LABEL: test_mm_addsub_pd: 14; X64: # BB#0: 15; X64-NEXT: addsubpd %xmm1, %xmm0 16; X64-NEXT: retq 17 %res = call <2 x double> @llvm.x86.sse3.addsub.pd(<2 x double> %a0, <2 x double> %a1) 18 ret <2 x double> %res 19} 20declare <2 x double> @llvm.x86.sse3.addsub.pd(<2 x double>, <2 x double>) nounwind readnone 21 22define <4 x float> @test_mm_addsub_ps(<4 x float> %a0, <4 x float> %a1) { 23; X32-LABEL: test_mm_addsub_ps: 24; X32: # BB#0: 25; X32-NEXT: addsubps %xmm1, %xmm0 26; X32-NEXT: retl 27; 28; X64-LABEL: test_mm_addsub_ps: 29; X64: # BB#0: 30; X64-NEXT: addsubps %xmm1, %xmm0 31; X64-NEXT: retq 32 %res = call <4 x float> @llvm.x86.sse3.addsub.ps(<4 x float> %a0, <4 x float> %a1) 33 ret <4 x float> %res 34} 35declare <4 x float> @llvm.x86.sse3.addsub.ps(<4 x float>, <4 x float>) nounwind readnone 36 37define <2 x double> @test_mm_hadd_pd(<2 x double> %a0, <2 x double> %a1) { 38; X32-LABEL: test_mm_hadd_pd: 39; X32: # BB#0: 40; X32-NEXT: haddpd %xmm1, %xmm0 41; X32-NEXT: retl 42; 43; X64-LABEL: test_mm_hadd_pd: 44; X64: # BB#0: 45; X64-NEXT: haddpd %xmm1, %xmm0 46; X64-NEXT: retq 47 %res = call <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double> %a0, <2 x double> %a1) 48 ret <2 x double> %res 49} 50declare <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double>, <2 x double>) nounwind readnone 51 52define <4 x float> @test_mm_hadd_ps(<4 x float> %a0, <4 x float> %a1) { 53; X32-LABEL: test_mm_hadd_ps: 54; X32: # BB#0: 55; X32-NEXT: haddps %xmm1, %xmm0 56; X32-NEXT: retl 57; 58; X64-LABEL: test_mm_hadd_ps: 59; X64: # BB#0: 60; X64-NEXT: haddps %xmm1, %xmm0 61; X64-NEXT: retq 62 %res = call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %a0, <4 x float> %a1) 63 ret <4 x float> %res 64} 65declare <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float>, <4 x float>) nounwind readnone 66 67define <2 x double> @test_mm_hsub_pd(<2 x double> %a0, <2 x double> %a1) { 68; X32-LABEL: test_mm_hsub_pd: 69; X32: # BB#0: 70; X32-NEXT: hsubpd %xmm1, %xmm0 71; X32-NEXT: retl 72; 73; X64-LABEL: test_mm_hsub_pd: 74; X64: # BB#0: 75; X64-NEXT: hsubpd %xmm1, %xmm0 76; X64-NEXT: retq 77 %res = call <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double> %a0, <2 x double> %a1) 78 ret <2 x double> %res 79} 80declare <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double>, <2 x double>) nounwind readnone 81 82define <4 x float> @test_mm_hsub_ps(<4 x float> %a0, <4 x float> %a1) { 83; X32-LABEL: test_mm_hsub_ps: 84; X32: # BB#0: 85; X32-NEXT: hsubps %xmm1, %xmm0 86; X32-NEXT: retl 87; 88; X64-LABEL: test_mm_hsub_ps: 89; X64: # BB#0: 90; X64-NEXT: hsubps %xmm1, %xmm0 91; X64-NEXT: retq 92 %res = call <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float> %a0, <4 x float> %a1) 93 ret <4 x float> %res 94} 95declare <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float>, <4 x float>) nounwind readnone 96 97define <2 x i64> @test_mm_lddqu_si128(<2 x i64>* %a0) { 98; X32-LABEL: test_mm_lddqu_si128: 99; X32: # BB#0: 100; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 101; X32-NEXT: lddqu (%eax), %xmm0 102; X32-NEXT: retl 103; 104; X64-LABEL: test_mm_lddqu_si128: 105; X64: # BB#0: 106; X64-NEXT: lddqu (%rdi), %xmm0 107; X64-NEXT: retq 108 %bc = bitcast <2 x i64>* %a0 to i8* 109 %call = call <16 x i8> @llvm.x86.sse3.ldu.dq(i8* %bc) 110 %res = bitcast <16 x i8> %call to <2 x i64> 111 ret <2 x i64> %res 112} 113declare <16 x i8> @llvm.x86.sse3.ldu.dq(i8*) nounwind readonly 114 115define <2 x double> @test_mm_loaddup_pd(double* %a0) { 116; X32-LABEL: test_mm_loaddup_pd: 117; X32: # BB#0: 118; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 119; X32-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] 120; X32-NEXT: retl 121; 122; X64-LABEL: test_mm_loaddup_pd: 123; X64: # BB#0: 124; X64-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] 125; X64-NEXT: retq 126 %ld = load double, double* %a0 127 %res0 = insertelement <2 x double> undef, double %ld, i32 0 128 %res1 = insertelement <2 x double> %res0, double %ld, i32 1 129 ret <2 x double> %res1 130} 131 132define <2 x double> @test_mm_movedup_pd(<2 x double> %a0) { 133; X32-LABEL: test_mm_movedup_pd: 134; X32: # BB#0: 135; X32-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0] 136; X32-NEXT: retl 137; 138; X64-LABEL: test_mm_movedup_pd: 139; X64: # BB#0: 140; X64-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0] 141; X64-NEXT: retq 142 %res = shufflevector <2 x double> %a0, <2 x double> %a0, <2 x i32> zeroinitializer 143 ret <2 x double> %res 144} 145 146define <4 x float> @test_mm_movehdup_ps(<4 x float> %a0) { 147; X32-LABEL: test_mm_movehdup_ps: 148; X32: # BB#0: 149; X32-NEXT: movshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] 150; X32-NEXT: retl 151; 152; X64-LABEL: test_mm_movehdup_ps: 153; X64: # BB#0: 154; X64-NEXT: movshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] 155; X64-NEXT: retq 156 %res = shufflevector <4 x float> %a0, <4 x float> %a0, <4 x i32> <i32 1, i32 1, i32 3, i32 3> 157 ret <4 x float> %res 158} 159 160define <4 x float> @test_mm_moveldup_ps(<4 x float> %a0) { 161; X32-LABEL: test_mm_moveldup_ps: 162; X32: # BB#0: 163; X32-NEXT: movsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] 164; X32-NEXT: retl 165; 166; X64-LABEL: test_mm_moveldup_ps: 167; X64: # BB#0: 168; X64-NEXT: movsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] 169; X64-NEXT: retq 170 %res = shufflevector <4 x float> %a0, <4 x float> %a0, <4 x i32> <i32 0, i32 0, i32 2, i32 2> 171 ret <4 x float> %res 172} 173