34 lines · plain
1; RUN: llc -mtriple=i686-unknown-unknown -mattr=+sse4.1 < %s | FileCheck %s -check-prefix=X322; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 < %s | FileCheck %s -check-prefix=X643 4; Test for case where insertps was folding the load of the insertion element, but a later optimization5; was then manipulating the load.6 7define <4 x float> @insertps_unfold(ptr %v0, ptr %v1) {8; X32-LABEL: insertps_unfold:9; X32: # %bb.0:10; X32-NEXT: movl {{[0-9]+}}(%esp), %eax11; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx12; X32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero13; X32-NEXT: movaps (%eax), %xmm014; X32-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0]15; X32-NEXT: addps %xmm1, %xmm016; X32-NEXT: retl17;18; X64-LABEL: insertps_unfold:19; X64: # %bb.0:20; X64-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero21; X64-NEXT: movaps (%rdi), %xmm022; X64-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0]23; X64-NEXT: addps %xmm1, %xmm024; X64-NEXT: retq25 %a = getelementptr inbounds <4 x float>, ptr %v1, i64 0, i64 126 %b = load float, ptr %a, align 427 %c = insertelement <4 x float> undef, float %b, i32 028 %d = load <4 x float>, ptr %v1, align 1629 %e = load <4 x float>, ptr %v0, align 1630 %f = shufflevector <4 x float> %e, <4 x float> %d, <4 x i32> <i32 0, i32 1, i32 2, i32 5>31 %g = fadd <4 x float> %c, %f32 ret <4 x float> %g33}34