198 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --no_x86_scrub_mem_shuffle2; RUN: llc < %s -mtriple=i686-apple-darwin -mcpu=corei7-avx | FileCheck %s --check-prefixes=CHECK,X863; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx | FileCheck %s --check-prefixes=CHECK,X644 5define <4 x i32> @blendvb_fallback_v4i32(<4 x i1> %mask, <4 x i32> %x, <4 x i32> %y) {6; CHECK-LABEL: blendvb_fallback_v4i32:7; CHECK: ## %bb.0:8; CHECK-NEXT: vpslld $31, %xmm0, %xmm09; CHECK-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm010; CHECK-NEXT: ret{{[l|q]}}11 %ret = select <4 x i1> %mask, <4 x i32> %x, <4 x i32> %y12 ret <4 x i32> %ret13}14 15define <8 x i32> @blendvb_fallback_v8i32(<8 x i1> %mask, <8 x i32> %x, <8 x i32> %y) {16; CHECK-LABEL: blendvb_fallback_v8i32:17; CHECK: ## %bb.0:18; CHECK-NEXT: vpmovzxwd {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero19; CHECK-NEXT: vpslld $31, %xmm3, %xmm320; CHECK-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]21; CHECK-NEXT: vpslld $31, %xmm0, %xmm022; CHECK-NEXT: vinsertf128 $1, %xmm0, %ymm3, %ymm023; CHECK-NEXT: vblendvps %ymm0, %ymm1, %ymm2, %ymm024; CHECK-NEXT: ret{{[l|q]}}25 %ret = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y26 ret <8 x i32> %ret27}28 29define <8 x float> @blendvb_fallback_v8f32(<8 x i1> %mask, <8 x float> %x, <8 x float> %y) {30; CHECK-LABEL: blendvb_fallback_v8f32:31; CHECK: ## %bb.0:32; CHECK-NEXT: vpmovzxwd {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero33; CHECK-NEXT: vpslld $31, %xmm3, %xmm334; CHECK-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]35; CHECK-NEXT: vpslld $31, %xmm0, %xmm036; CHECK-NEXT: vinsertf128 $1, %xmm0, %ymm3, %ymm037; CHECK-NEXT: vblendvps %ymm0, %ymm1, %ymm2, %ymm038; CHECK-NEXT: ret{{[l|q]}}39 %ret = select <8 x i1> %mask, <8 x float> %x, <8 x float> %y40 ret <8 x float> %ret41}42 43declare <4 x float> @llvm.x86.sse41.insertps(<4 x float>, <4 x float>, i32) nounwind readnone44 45define <4 x float> @insertps_from_vector_load(<4 x float> %a, ptr nocapture readonly %pb) {46; X86-LABEL: insertps_from_vector_load:47; X86: ## %bb.0:48; X86-NEXT: movl {{[0-9]+}}(%esp), %eax49; X86-NEXT: vinsertps $48, (%eax), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]50; X86-NEXT: retl51;52; X64-LABEL: insertps_from_vector_load:53; X64: ## %bb.0:54; X64-NEXT: vinsertps $48, (%rdi), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]55; X64-NEXT: retq56 %1 = load <4 x float>, ptr %pb, align 1657 %2 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %1, i32 48)58 ret <4 x float> %259}60 61;; Use a non-zero CountS for insertps62define <4 x float> @insertps_from_vector_load_offset(<4 x float> %a, ptr nocapture readonly %pb) {63; X86-LABEL: insertps_from_vector_load_offset:64; X86: ## %bb.0:65; X86-NEXT: movl {{[0-9]+}}(%esp), %eax66; X86-NEXT: vinsertps $32, 4(%eax), %xmm0, %xmm0 ## xmm0 = xmm0[0,1],mem[0],xmm0[3]67; X86-NEXT: retl68;69; X64-LABEL: insertps_from_vector_load_offset:70; X64: ## %bb.0:71; X64-NEXT: vinsertps $32, 4(%rdi), %xmm0, %xmm0 ## xmm0 = xmm0[0,1],mem[0],xmm0[3]72; X64-NEXT: retq73 %1 = load <4 x float>, ptr %pb, align 1674 %2 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %1, i32 96)75 ret <4 x float> %276}77 78define <4 x float> @insertps_from_vector_load_offset_2(<4 x float> %a, ptr nocapture readonly %pb, i64 %index) {79; X86-LABEL: insertps_from_vector_load_offset_2:80; X86: ## %bb.0:81; X86-NEXT: movl {{[0-9]+}}(%esp), %eax82; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx83; X86-NEXT: shll $4, %ecx84; X86-NEXT: vinsertps $0, 12(%eax,%ecx), %xmm0, %xmm0 ## xmm0 = mem[0],xmm0[1,2,3]85; X86-NEXT: retl86;87; X64-LABEL: insertps_from_vector_load_offset_2:88; X64: ## %bb.0:89; X64-NEXT: shlq $4, %rsi90; X64-NEXT: vinsertps $0, 12(%rdi,%rsi), %xmm0, %xmm0 ## xmm0 = mem[0],xmm0[1,2,3]91; X64-NEXT: retq92 %1 = getelementptr inbounds <4 x float>, ptr %pb, i64 %index93 %2 = load <4 x float>, ptr %1, align 1694 %3 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %2, i32 192)95 ret <4 x float> %396}97 98define <4 x float> @insertps_from_broadcast_loadf32(<4 x float> %a, ptr nocapture readonly %fb, i64 %index) {99; X86-LABEL: insertps_from_broadcast_loadf32:100; X86: ## %bb.0:101; X86-NEXT: movl {{[0-9]+}}(%esp), %eax102; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx103; X86-NEXT: vinsertps $48, (%ecx,%eax,4), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]104; X86-NEXT: retl105;106; X64-LABEL: insertps_from_broadcast_loadf32:107; X64: ## %bb.0:108; X64-NEXT: vinsertps $48, (%rdi,%rsi,4), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]109; X64-NEXT: retq110 %1 = getelementptr inbounds float, ptr %fb, i64 %index111 %2 = load float, ptr %1, align 4112 %3 = insertelement <4 x float> undef, float %2, i32 0113 %4 = insertelement <4 x float> %3, float %2, i32 1114 %5 = insertelement <4 x float> %4, float %2, i32 2115 %6 = insertelement <4 x float> %5, float %2, i32 3116 %7 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %6, i32 48)117 ret <4 x float> %7118}119 120define <4 x float> @insertps_from_broadcast_loadv4f32(<4 x float> %a, ptr nocapture readonly %b) {121; X86-LABEL: insertps_from_broadcast_loadv4f32:122; X86: ## %bb.0:123; X86-NEXT: movl {{[0-9]+}}(%esp), %eax124; X86-NEXT: vinsertps $48, (%eax), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]125; X86-NEXT: retl126;127; X64-LABEL: insertps_from_broadcast_loadv4f32:128; X64: ## %bb.0:129; X64-NEXT: vinsertps $48, (%rdi), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]130; X64-NEXT: retq131 %1 = load <4 x float>, ptr %b, align 4132 %2 = extractelement <4 x float> %1, i32 0133 %3 = insertelement <4 x float> undef, float %2, i32 0134 %4 = insertelement <4 x float> %3, float %2, i32 1135 %5 = insertelement <4 x float> %4, float %2, i32 2136 %6 = insertelement <4 x float> %5, float %2, i32 3137 %7 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %6, i32 48)138 ret <4 x float> %7139}140 141define <4 x float> @insertps_from_broadcast_multiple_use(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x float> %d, ptr nocapture readonly %fb, i64 %index) {142; X86-LABEL: insertps_from_broadcast_multiple_use:143; X86: ## %bb.0:144; X86-NEXT: movl {{[0-9]+}}(%esp), %eax145; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx146; X86-NEXT: vbroadcastss (%ecx,%eax,4), %xmm4147; X86-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0,1,2],xmm4[3]148; X86-NEXT: vblendps {{.*#+}} xmm1 = xmm1[0,1,2],xmm4[3]149; X86-NEXT: vaddps %xmm1, %xmm0, %xmm0150; X86-NEXT: vblendps {{.*#+}} xmm1 = xmm2[0,1,2],xmm4[3]151; X86-NEXT: vblendps {{.*#+}} xmm2 = xmm3[0,1,2],xmm4[3]152; X86-NEXT: vaddps %xmm2, %xmm1, %xmm1153; X86-NEXT: vaddps %xmm1, %xmm0, %xmm0154; X86-NEXT: retl155;156; X64-LABEL: insertps_from_broadcast_multiple_use:157; X64: ## %bb.0:158; X64-NEXT: vbroadcastss (%rdi,%rsi,4), %xmm4159; X64-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0,1,2],xmm4[3]160; X64-NEXT: vblendps {{.*#+}} xmm1 = xmm1[0,1,2],xmm4[3]161; X64-NEXT: vaddps %xmm1, %xmm0, %xmm0162; X64-NEXT: vblendps {{.*#+}} xmm1 = xmm2[0,1,2],xmm4[3]163; X64-NEXT: vblendps {{.*#+}} xmm2 = xmm3[0,1,2],xmm4[3]164; X64-NEXT: vaddps %xmm2, %xmm1, %xmm1165; X64-NEXT: vaddps %xmm1, %xmm0, %xmm0166; X64-NEXT: retq167 %1 = getelementptr inbounds float, ptr %fb, i64 %index168 %2 = load float, ptr %1, align 4169 %3 = insertelement <4 x float> undef, float %2, i32 0170 %4 = insertelement <4 x float> %3, float %2, i32 1171 %5 = insertelement <4 x float> %4, float %2, i32 2172 %6 = insertelement <4 x float> %5, float %2, i32 3173 %7 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %6, i32 48)174 %8 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %b, <4 x float> %6, i32 48)175 %9 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %c, <4 x float> %6, i32 48)176 %10 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %d, <4 x float> %6, i32 48)177 %11 = fadd <4 x float> %7, %8178 %12 = fadd <4 x float> %9, %10179 %13 = fadd <4 x float> %11, %12180 ret <4 x float> %13181}182 183define <4 x float> @nofold_insertps(ptr %a, <4 x float> %b) {184; X86-LABEL: nofold_insertps:185; X86: ## %bb.0:186; X86-NEXT: movl {{[0-9]+}}(%esp), %eax187; X86-NEXT: vinsertps $48, 8(%eax), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]188; X86-NEXT: retl189;190; X64-LABEL: nofold_insertps:191; X64: ## %bb.0:192; X64-NEXT: vinsertps $48, 8(%rdi), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]193; X64-NEXT: retq194 %1 = load <4 x float>, ptr %a, align 1195 %2 = shufflevector <4 x float> %b, <4 x float> %1, <4 x i32> <i32 0, i32 1, i32 2, i32 6>196 ret <4 x float> %2197}198