1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s 3 4define void @foo(<3 x float> %in, <4 x i8>* nocapture %out) nounwind { 5; CHECK-LABEL: foo: 6; CHECK: # BB#0: 7; CHECK-NEXT: cvttps2dq %xmm0, %xmm0 8; CHECK-NEXT: movl $255, %eax 9; CHECK-NEXT: pinsrd $3, %eax, %xmm0 10; CHECK-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u] 11; CHECK-NEXT: movd %xmm0, (%rdi) 12; CHECK-NEXT: retq 13 %t0 = fptoui <3 x float> %in to <3 x i8> 14 %t1 = shufflevector <3 x i8> %t0, <3 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 undef> 15 %t2 = insertelement <4 x i8> %t1, i8 -1, i32 3 16 store <4 x i8> %t2, <4 x i8>* %out, align 4 17 ret void 18} 19 20; Verify that the DAGCombiner doesn't wrongly fold a build_vector into a 21; blend with a zero vector if the build_vector contains negative zero. 22; 23; TODO: the codegen for function 'test_negative_zero_1' is sub-optimal. 24; Ideally, we should generate a single shuffle blend operation. 25 26define <4 x float> @test_negative_zero_1(<4 x float> %A) { 27; CHECK-LABEL: test_negative_zero_1: 28; CHECK: # BB#0: # %entry 29; CHECK-NEXT: movapd %xmm0, %xmm1 30; CHECK-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1,0] 31; CHECK-NEXT: xorps %xmm2, %xmm2 32; CHECK-NEXT: blendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3] 33; CHECK-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero 34; CHECK-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] 35; CHECK-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0] 36; CHECK-NEXT: retq 37entry: 38 %0 = extractelement <4 x float> %A, i32 0 39 %1 = insertelement <4 x float> undef, float %0, i32 0 40 %2 = insertelement <4 x float> %1, float -0.0, i32 1 41 %3 = extractelement <4 x float> %A, i32 2 42 %4 = insertelement <4 x float> %2, float %3, i32 2 43 %5 = insertelement <4 x float> %4, float 0.0, i32 3 44 ret <4 x float> %5 45} 46 47define <2 x double> @test_negative_zero_2(<2 x double> %A) { 48; CHECK-LABEL: test_negative_zero_2: 49; CHECK: # BB#0: # %entry 50; CHECK-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] 51; CHECK-NEXT: retq 52entry: 53 %0 = extractelement <2 x double> %A, i32 0 54 %1 = insertelement <2 x double> undef, double %0, i32 0 55 %2 = insertelement <2 x double> %1, double -0.0, i32 1 56 ret <2 x double> %2 57} 58