1f4a2713aSLionel Sambuc; RUN: llc < %s -march=x86-64 -mcpu=corei7 -mattr=+sse4.1 | FileCheck %s 2f4a2713aSLionel Sambuc 3f4a2713aSLionel Sambuc; rdar://12721174 4f4a2713aSLionel Sambuc; We should not fold movss into pshufd since pshufd expects m128 while movss 5f4a2713aSLionel Sambuc; loads from m32. 6f4a2713aSLionel Sambucdefine void @sample_test(<4 x float>* %source, <2 x float>* %dest) nounwind { 7f4a2713aSLionel Sambuc; CHECK: sample_test 8*0a6a1f1dSLionel Sambuc; CHECK-NOT: movaps 9f4a2713aSLionel Sambuc; CHECK: insertps 10f4a2713aSLionel Sambucentry: 11f4a2713aSLionel Sambuc %source.addr = alloca <4 x float>*, align 8 12f4a2713aSLionel Sambuc %dest.addr = alloca <2 x float>*, align 8 13f4a2713aSLionel Sambuc %tmp = alloca <2 x float>, align 8 14f4a2713aSLionel Sambuc store <4 x float>* %source, <4 x float>** %source.addr, align 8 15f4a2713aSLionel Sambuc store <2 x float>* %dest, <2 x float>** %dest.addr, align 8 16f4a2713aSLionel Sambuc store <2 x float> zeroinitializer, <2 x float>* %tmp, align 8 17f4a2713aSLionel Sambuc %0 = load <4 x float>** %source.addr, align 8 18f4a2713aSLionel Sambuc %arrayidx = getelementptr inbounds <4 x float>* %0, i64 0 19f4a2713aSLionel Sambuc %1 = load <4 x float>* %arrayidx, align 16 20f4a2713aSLionel Sambuc %2 = extractelement <4 x float> %1, i32 0 21f4a2713aSLionel Sambuc %3 = load <2 x float>* %tmp, align 8 22f4a2713aSLionel Sambuc %4 = insertelement <2 x float> %3, float %2, i32 1 23f4a2713aSLionel Sambuc store <2 x float> %4, <2 x float>* %tmp, align 8 24f4a2713aSLionel Sambuc %5 = load <2 x float>* %tmp, align 8 25f4a2713aSLionel Sambuc %6 = load <2 x float>** %dest.addr, align 8 26f4a2713aSLionel Sambuc %arrayidx1 = getelementptr inbounds <2 x float>* %6, i64 0 27f4a2713aSLionel Sambuc store <2 x float> %5, <2 x float>* %arrayidx1, align 8 28f4a2713aSLionel Sambuc %7 = load <2 x float>** %dest.addr, align 8 29f4a2713aSLionel Sambuc %arrayidx2 = getelementptr inbounds <2 x float>* %7, i64 0 30f4a2713aSLionel Sambuc %8 = load <2 x float>* %arrayidx2, align 8 31f4a2713aSLionel Sambuc %vecext = extractelement <2 x float> %8, i32 0 32f4a2713aSLionel Sambuc %9 = load <2 x float>** %dest.addr, align 8 33f4a2713aSLionel Sambuc %arrayidx3 = getelementptr inbounds <2 x float>* %9, i64 0 34f4a2713aSLionel Sambuc %10 = load <2 x float>* %arrayidx3, align 8 35f4a2713aSLionel Sambuc %vecext4 = extractelement <2 x float> %10, i32 1 36f4a2713aSLionel Sambuc call void @ext(float %vecext, float %vecext4) 37f4a2713aSLionel Sambuc ret void 38f4a2713aSLionel Sambuc} 39f4a2713aSLionel Sambucdeclare void @ext(float, float) 40