1; RUN: llc < %s -march=x86-64 -mattr=+mmx,+sse2 | FileCheck %s 2 3define i32 @test0(<1 x i64>* %v4) { 4; CHECK-LABEL: test0: 5; CHECK: # BB#0:{{.*}} %entry 6; CHECK: pshufw $238, (%[[REG:[a-z]+]]), %mm0 7; CHECK-NEXT: movd %mm0, %eax 8; CHECK-NEXT: addl $32, %eax 9; CHECK-NEXT: retq 10entry: 11 %v5 = load <1 x i64>, <1 x i64>* %v4, align 8 12 %v12 = bitcast <1 x i64> %v5 to <4 x i16> 13 %v13 = bitcast <4 x i16> %v12 to x86_mmx 14 %v14 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %v13, i8 -18) 15 %v15 = bitcast x86_mmx %v14 to <4 x i16> 16 %v16 = bitcast <4 x i16> %v15 to <1 x i64> 17 %v17 = extractelement <1 x i64> %v16, i32 0 18 %v18 = bitcast i64 %v17 to <2 x i32> 19 %v19 = extractelement <2 x i32> %v18, i32 0 20 %v20 = add i32 %v19, 32 21 ret i32 %v20 22} 23 24define i32 @test1(i32* nocapture readonly %ptr) { 25; CHECK-LABEL: test1: 26; CHECK: # BB#0:{{.*}} %entry 27; CHECK: movd (%[[REG]]), %mm0 28; CHECK-NEXT: pshufw $232, %mm0, %mm0 29; CHECK-NEXT: movd %mm0, %eax 30; CHECK-NEXT: emms 31; CHECK-NEXT: retq 32entry: 33 %0 = load i32, i32* %ptr, align 4 34 %1 = insertelement <2 x i32> undef, i32 %0, i32 0 35 %2 = insertelement <2 x i32> %1, i32 0, i32 1 36 %3 = bitcast <2 x i32> %2 to x86_mmx 37 %4 = bitcast x86_mmx %3 to i64 38 %5 = bitcast i64 %4 to <4 x i16> 39 %6 = bitcast <4 x i16> %5 to x86_mmx 40 %7 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %6, i8 -24) 41 %8 = bitcast x86_mmx %7 to <4 x i16> 42 %9 = bitcast <4 x i16> %8 to <1 x i64> 43 %10 = extractelement <1 x i64> %9, i32 0 44 %11 = bitcast i64 %10 to <2 x i32> 45 %12 = extractelement <2 x i32> %11, i32 0 46 tail call void @llvm.x86.mmx.emms() 47 ret i32 %12 48} 49 50define i32 @test2(i32* nocapture readonly %ptr) { 51; CHECK-LABEL: test2: 52; CHECK: # BB#0:{{.*}} %entry 53; CHECK: pshufw $232, (%[[REG]]), %mm0 54; CHECK-NEXT: movd %mm0, %eax 55; CHECK-NEXT: emms 56; CHECK-NEXT: retq 57entry: 58 %0 = bitcast i32* %ptr to x86_mmx* 59 %1 = load x86_mmx, x86_mmx* %0, align 8 60 %2 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %1, i8 -24) 61 %3 = bitcast x86_mmx %2 to <4 x i16> 62 %4 = bitcast <4 x i16> %3 to <1 x i64> 63 %5 = extractelement <1 x i64> %4, i32 0 64 %6 = bitcast i64 %5 to <2 x i32> 65 %7 = extractelement <2 x i32> %6, i32 0 66 tail call void @llvm.x86.mmx.emms() 67 ret i32 %7 68} 69 70declare x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx, i8) 71declare void @llvm.x86.mmx.emms() 72