1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-unknown -mattr=+mmx,+sse2 | FileCheck %s --check-prefix=X32 3; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+mmx,+sse2 | FileCheck %s --check-prefix=X64 4 5define i32 @test0(<1 x i64>* %v4) nounwind { 6; X32-LABEL: test0: 7; X32: # %bb.0: # %entry 8; X32-NEXT: pushl %ebp 9; X32-NEXT: movl %esp, %ebp 10; X32-NEXT: andl $-8, %esp 11; X32-NEXT: subl $8, %esp 12; X32-NEXT: movl 8(%ebp), %eax 13; X32-NEXT: movl (%eax), %ecx 14; X32-NEXT: movl 4(%eax), %eax 15; X32-NEXT: movl %eax, {{[0-9]+}}(%esp) 16; X32-NEXT: movl %ecx, (%esp) 17; X32-NEXT: pshufw $238, (%esp), %mm0 # mm0 = mem[2,3,2,3] 18; X32-NEXT: movd %mm0, %eax 19; X32-NEXT: addl $32, %eax 20; X32-NEXT: movl %ebp, %esp 21; X32-NEXT: popl %ebp 22; X32-NEXT: retl 23; 24; X64-LABEL: test0: 25; X64: # %bb.0: # %entry 26; X64-NEXT: pshufw $238, (%rdi), %mm0 # mm0 = mem[2,3,2,3] 27; X64-NEXT: movd %mm0, %eax 28; X64-NEXT: addl $32, %eax 29; X64-NEXT: retq 30entry: 31 %v5 = load <1 x i64>, <1 x i64>* %v4, align 8 32 %v12 = bitcast <1 x i64> %v5 to <4 x i16> 33 %v13 = bitcast <4 x i16> %v12 to x86_mmx 34 %v14 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %v13, i8 -18) 35 %v15 = bitcast x86_mmx %v14 to <4 x i16> 36 %v16 = bitcast <4 x i16> %v15 to <1 x i64> 37 %v17 = extractelement <1 x i64> %v16, i32 0 38 %v18 = bitcast i64 %v17 to <2 x i32> 39 %v19 = extractelement <2 x i32> %v18, i32 0 40 %v20 = add i32 %v19, 32 41 ret i32 %v20 42} 43 44define i32 @test1(i32* nocapture readonly %ptr) nounwind { 45; X32-LABEL: test1: 46; X32: # %bb.0: # %entry 47; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 48; X32-NEXT: movd (%eax), %mm0 49; X32-NEXT: pshufw $232, %mm0, %mm0 # mm0 = mm0[0,2,2,3] 50; X32-NEXT: movd %mm0, %eax 51; X32-NEXT: emms 52; X32-NEXT: retl 53; 54; X64-LABEL: test1: 55; X64: # %bb.0: # %entry 56; X64-NEXT: movd (%rdi), %mm0 57; X64-NEXT: pshufw $232, %mm0, %mm0 # mm0 = mm0[0,2,2,3] 58; X64-NEXT: movd %mm0, %eax 59; X64-NEXT: emms 60; X64-NEXT: retq 61entry: 62 %0 = load i32, i32* %ptr, align 4 63 %1 = insertelement <2 x i32> undef, i32 %0, i32 0 64 %2 = insertelement <2 x i32> %1, i32 0, i32 1 65 %3 = bitcast <2 x i32> %2 to x86_mmx 66 %4 = bitcast x86_mmx %3 to i64 67 %5 = bitcast i64 %4 to <4 x i16> 68 %6 = bitcast <4 x i16> %5 to x86_mmx 69 %7 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %6, i8 -24) 70 %8 = bitcast x86_mmx %7 to <4 x i16> 71 %9 = bitcast <4 x i16> %8 to <1 x i64> 72 %10 = extractelement <1 x i64> %9, i32 0 73 %11 = bitcast i64 %10 to <2 x i32> 74 %12 = extractelement <2 x i32> %11, i32 0 75 tail call void @llvm.x86.mmx.emms() 76 ret i32 %12 77} 78 79define i32 @test2(i32* nocapture readonly %ptr) nounwind { 80; X32-LABEL: test2: 81; X32: # %bb.0: # %entry 82; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 83; X32-NEXT: pshufw $232, (%eax), %mm0 # mm0 = mem[0,2,2,3] 84; X32-NEXT: movd %mm0, %eax 85; X32-NEXT: emms 86; X32-NEXT: retl 87; 88; X64-LABEL: test2: 89; X64: # %bb.0: # %entry 90; X64-NEXT: pshufw $232, (%rdi), %mm0 # mm0 = mem[0,2,2,3] 91; X64-NEXT: movd %mm0, %eax 92; X64-NEXT: emms 93; X64-NEXT: retq 94entry: 95 %0 = bitcast i32* %ptr to x86_mmx* 96 %1 = load x86_mmx, x86_mmx* %0, align 8 97 %2 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %1, i8 -24) 98 %3 = bitcast x86_mmx %2 to <4 x i16> 99 %4 = bitcast <4 x i16> %3 to <1 x i64> 100 %5 = extractelement <1 x i64> %4, i32 0 101 %6 = bitcast i64 %5 to <2 x i32> 102 %7 = extractelement <2 x i32> %6, i32 0 103 tail call void @llvm.x86.mmx.emms() 104 ret i32 %7 105} 106 107define i32 @test3(x86_mmx %a) nounwind { 108; X32-LABEL: test3: 109; X32: # %bb.0: 110; X32-NEXT: movd %mm0, %eax 111; X32-NEXT: retl 112; 113; X64-LABEL: test3: 114; X64: # %bb.0: 115; X64-NEXT: movd %mm0, %eax 116; X64-NEXT: retq 117 %tmp0 = bitcast x86_mmx %a to <2 x i32> 118 %tmp1 = extractelement <2 x i32> %tmp0, i32 0 119 ret i32 %tmp1 120} 121 122; Verify we don't muck with extractelts from the upper lane. 123define i32 @test4(x86_mmx %a) nounwind { 124; X32-LABEL: test4: 125; X32: # %bb.0: 126; X32-NEXT: pushl %ebp 127; X32-NEXT: movl %esp, %ebp 128; X32-NEXT: andl $-8, %esp 129; X32-NEXT: subl $8, %esp 130; X32-NEXT: movq %mm0, (%esp) 131; X32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 132; X32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,0,1] 133; X32-NEXT: movd %xmm0, %eax 134; X32-NEXT: movl %ebp, %esp 135; X32-NEXT: popl %ebp 136; X32-NEXT: retl 137; 138; X64-LABEL: test4: 139; X64: # %bb.0: 140; X64-NEXT: movq %mm0, -{{[0-9]+}}(%rsp) 141; X64-NEXT: movq {{.*#+}} xmm0 = mem[0],zero 142; X64-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,3,0,1] 143; X64-NEXT: movd %xmm0, %eax 144; X64-NEXT: retq 145 %tmp0 = bitcast x86_mmx %a to <2 x i32> 146 %tmp1 = extractelement <2 x i32> %tmp0, i32 1 147 ret i32 %tmp1 148} 149 150declare x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx, i8) 151declare void @llvm.x86.mmx.emms() 152