1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X86 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64 4 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, %xmm1 10; X86-NEXT: je .LBB0_2 11; X86-NEXT: # %bb.1: 12; X86-NEXT: vmovaps %ymm0, %ymm1 13; X86-NEXT: .LBB0_2: 14; X86-NEXT: vxorps %ymm1, %ymm0, %ymm0 15; X86-NEXT: retl 16; 17; X64-LABEL: select00: 18; X64: # %bb.0: 19; X64-NEXT: cmpl $255, %edi 20; X64-NEXT: vxorps %xmm1, %xmm1, %xmm1 21; X64-NEXT: je .LBB0_2 22; X64-NEXT: # %bb.1: 23; X64-NEXT: vmovaps %ymm0, %ymm1 24; X64-NEXT: .LBB0_2: 25; X64-NEXT: vxorps %ymm1, %ymm0, %ymm0 26; X64-NEXT: retq 27 %cmpres = icmp eq i32 %a, 255 28 %selres = select i1 %cmpres, <8 x i32> zeroinitializer, <8 x i32> %b 29 %res = xor <8 x i32> %b, %selres 30 ret <8 x i32> %res 31} 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, %xmm1 38; X86-NEXT: je .LBB1_2 39; X86-NEXT: # %bb.1: 40; X86-NEXT: vmovaps %ymm0, %ymm1 41; X86-NEXT: .LBB1_2: 42; X86-NEXT: vxorps %ymm1, %ymm0, %ymm0 43; X86-NEXT: retl 44; 45; X64-LABEL: select01: 46; X64: # %bb.0: 47; X64-NEXT: cmpl $255, %edi 48; X64-NEXT: vxorps %xmm1, %xmm1, %xmm1 49; X64-NEXT: je .LBB1_2 50; X64-NEXT: # %bb.1: 51; X64-NEXT: vmovaps %ymm0, %ymm1 52; X64-NEXT: .LBB1_2: 53; X64-NEXT: vxorps %ymm1, %ymm0, %ymm0 54; X64-NEXT: retq 55 %cmpres = icmp eq i32 %a, 255 56 %selres = select i1 %cmpres, <4 x i64> zeroinitializer, <4 x i64> %b 57 %res = xor <4 x i64> %b, %selres 58 ret <4 x i64> %res 59} 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: # %entry 65; X86-NEXT: vmovaps {{.*#+}} ymm0 = [4294942349,7802,29242,15858,29361,4294951202,4294964216,4294941010] 66; X86-NEXT: vmovaps %ymm0, (%eax) 67; X86-NEXT: vzeroupper 68; X86-NEXT: retl 69; 70; X64-LABEL: fold_blendv_mask: 71; X64: # %bb.0: # %entry 72; X64-NEXT: vmovaps {{.*#+}} ymm0 = [4294942349,7802,29242,15858,29361,4294951202,4294964216,4294941010] 73; X64-NEXT: vmovaps %ymm0, (%rax) 74; X64-NEXT: vzeroupper 75; X64-NEXT: retq 76entry: 77 br label %head 78 79head: 80 %v0 = insertelement <4 x i32> %a0, i32 44158, i64 0 81 %v1 = insertelement <4 x i32> %v0, i32 54560, i64 1 82 %v2 = insertelement <4 x i32> %v1, i32 45291, i64 2 83 %v3 = insertelement <4 x i32> %v2, i32 18686, i64 3 84 %isneg = icmp slt <4 x i32> %v3, zeroinitializer 85 %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 %head 88 89exit: 90 store <8 x i32> %or1, ptr addrspace(1) undef, align 32 91 ret void 92} 93