93 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X863; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X644 5define <8 x i32> @select00(i32 %a, <8 x i32> %b) nounwind {6; X86-LABEL: select00:7; X86: # %bb.0:8; X86-NEXT: cmpl $255, {{[0-9]+}}(%esp)9; X86-NEXT: vxorps %xmm1, %xmm1, %xmm110; X86-NEXT: je .LBB0_211; X86-NEXT: # %bb.1:12; X86-NEXT: vmovaps %ymm0, %ymm113; X86-NEXT: .LBB0_2:14; X86-NEXT: vxorps %ymm1, %ymm0, %ymm015; X86-NEXT: retl16;17; X64-LABEL: select00:18; X64: # %bb.0:19; X64-NEXT: cmpl $255, %edi20; X64-NEXT: vxorps %xmm1, %xmm1, %xmm121; X64-NEXT: je .LBB0_222; X64-NEXT: # %bb.1:23; X64-NEXT: vmovaps %ymm0, %ymm124; X64-NEXT: .LBB0_2:25; X64-NEXT: vxorps %ymm1, %ymm0, %ymm026; X64-NEXT: retq27 %cmpres = icmp eq i32 %a, 25528 %selres = select i1 %cmpres, <8 x i32> zeroinitializer, <8 x i32> %b29 %res = xor <8 x i32> %b, %selres30 ret <8 x i32> %res31}32 33define <4 x i64> @select01(i32 %a, <4 x i64> %b) nounwind {34; X86-LABEL: select01:35; X86: # %bb.0:36; X86-NEXT: cmpl $255, {{[0-9]+}}(%esp)37; X86-NEXT: vxorps %xmm1, %xmm1, %xmm138; X86-NEXT: je .LBB1_239; X86-NEXT: # %bb.1:40; X86-NEXT: vmovaps %ymm0, %ymm141; X86-NEXT: .LBB1_2:42; X86-NEXT: vxorps %ymm1, %ymm0, %ymm043; X86-NEXT: retl44;45; X64-LABEL: select01:46; X64: # %bb.0:47; X64-NEXT: cmpl $255, %edi48; X64-NEXT: vxorps %xmm1, %xmm1, %xmm149; X64-NEXT: je .LBB1_250; X64-NEXT: # %bb.1:51; X64-NEXT: vmovaps %ymm0, %ymm152; X64-NEXT: .LBB1_2:53; X64-NEXT: vxorps %ymm1, %ymm0, %ymm054; X64-NEXT: retq55 %cmpres = icmp eq i32 %a, 25556 %selres = select i1 %cmpres, <4 x i64> zeroinitializer, <4 x i64> %b57 %res = xor <4 x i64> %b, %selres58 ret <4 x i64> %res59}60 61; If a X86ISD::BLENDV node appears before legalization, constant fold using (mask < 0) instead of like a vselect (mask != 0).62define void @fold_blendv_mask(<4 x i32> %a0) {63; X86-LABEL: fold_blendv_mask:64; X86: # %bb.0: # %entry65; X86-NEXT: vmovaps {{.*#+}} ymm0 = [4294942349,7802,29242,15858,29361,4294951202,4294964216,4294941010]66; X86-NEXT: vmovaps %ymm0, (%eax)67; X86-NEXT: vzeroupper68; X86-NEXT: retl69;70; X64-LABEL: fold_blendv_mask:71; X64: # %bb.0: # %entry72; X64-NEXT: vmovaps {{.*#+}} ymm0 = [4294942349,7802,29242,15858,29361,4294951202,4294964216,4294941010]73; X64-NEXT: vmovaps %ymm0, (%rax)74; X64-NEXT: vzeroupper75; X64-NEXT: retq76entry:77 br label %head78 79head:80 %v0 = insertelement <4 x i32> %a0, i32 44158, i64 081 %v1 = insertelement <4 x i32> %v0, i32 54560, i64 182 %v2 = insertelement <4 x i32> %v1, i32 45291, i64 283 %v3 = insertelement <4 x i32> %v2, i32 18686, i64 384 %isneg = icmp slt <4 x i32> %v3, zeroinitializer85 %or0 = select <4 x i1> %isneg, <4 x i32> <i32 26146, i32 -1257, i32 -2, i32 -3052>, <4 x i32> <i32 -24947, i32 7802, i32 29242, i32 15858>86 %or1 = shufflevector <4 x i32> %or0, <4 x i32> <i32 29361, i32 -16094, i32 -3080, i32 -26286>, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>87 br i1 poison, label %exit, label %head88 89exit:90 store <8 x i32> %or1, ptr addrspace(1) undef, align 3291 ret void92}93