1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-linux -mcpu=penryn -o - | FileCheck %s --check-prefix=X64 3; RUN: llc < %s -mcpu=yonah -march=x86 -mtriple=i386-linux-gnu -o - | FileCheck %s --check-prefix=X32 4 5; PR7518 6define void @test1(<2 x float> %Q, float *%P2) nounwind { 7; X64-LABEL: test1: 8; X64: # BB#0: 9; X64-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] 10; X64-NEXT: addss %xmm0, %xmm1 11; X64-NEXT: movss %xmm1, (%rdi) 12; X64-NEXT: retq 13; 14; X32-LABEL: test1: 15; X32: # BB#0: 16; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 17; X32-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] 18; X32-NEXT: addss %xmm0, %xmm1 19; X32-NEXT: movss %xmm1, (%eax) 20; X32-NEXT: retl 21 %a = extractelement <2 x float> %Q, i32 0 22 %b = extractelement <2 x float> %Q, i32 1 23 %c = fadd float %a, %b 24 store float %c, float* %P2 25 ret void 26} 27 28define <2 x float> @test2(<2 x float> %Q, <2 x float> %R, <2 x float> *%P) nounwind { 29; X64-LABEL: test2: 30; X64: # BB#0: 31; X64-NEXT: addps %xmm1, %xmm0 32; X64-NEXT: retq 33; 34; X32-LABEL: test2: 35; X32: # BB#0: 36; X32-NEXT: addps %xmm1, %xmm0 37; X32-NEXT: retl 38 %Z = fadd <2 x float> %Q, %R 39 ret <2 x float> %Z 40} 41 42define <2 x float> @test3(<4 x float> %A) nounwind { 43; X64-LABEL: test3: 44; X64: # BB#0: 45; X64-NEXT: addps %xmm0, %xmm0 46; X64-NEXT: retq 47; 48; X32-LABEL: test3: 49; X32: # BB#0: 50; X32-NEXT: addps %xmm0, %xmm0 51; X32-NEXT: retl 52 %B = shufflevector <4 x float> %A, <4 x float> undef, <2 x i32> <i32 0, i32 1> 53 %C = fadd <2 x float> %B, %B 54 ret <2 x float> %C 55} 56 57define <2 x float> @test4(<2 x float> %A) nounwind { 58; X64-LABEL: test4: 59; X64: # BB#0: 60; X64-NEXT: addps %xmm0, %xmm0 61; X64-NEXT: retq 62; 63; X32-LABEL: test4: 64; X32: # BB#0: 65; X32-NEXT: addps %xmm0, %xmm0 66; X32-NEXT: retl 67 %C = fadd <2 x float> %A, %A 68 ret <2 x float> %C 69} 70 71define <4 x float> @test5(<4 x float> %A) nounwind { 72; X64-LABEL: test5: 73; X64: # BB#0: 74; X64-NEXT: addps %xmm0, %xmm0 75; X64-NEXT: addps %xmm0, %xmm0 76; X64-NEXT: retq 77; 78; X32-LABEL: test5: 79; X32: # BB#0: 80; X32-NEXT: addps %xmm0, %xmm0 81; X32-NEXT: addps %xmm0, %xmm0 82; X32-NEXT: retl 83 %B = shufflevector <4 x float> %A, <4 x float> undef, <2 x i32> <i32 0, i32 1> 84 %C = fadd <2 x float> %B, %B 85 br label %BB 86 87BB: 88 %D = fadd <2 x float> %C, %C 89 %E = shufflevector <2 x float> %D, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 90 ret <4 x float> %E 91} 92 93 94