416 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -disable-peephole -mtriple=i686-apple-darwin9 -mattr=+sse,+sse2,+sse4.1 | FileCheck %s --check-prefixes=X863; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin9 -mattr=+sse,+sse2,+sse4.1 | FileCheck %s --check-prefixes=X644; RUN: llc < %s -disable-peephole -mtriple=i686-apple-darwin9 -mattr=+avx | FileCheck %s --check-prefixes=X86_AVX,X86_AVX15; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin9 -mattr=+avx | FileCheck %s --check-prefixes=X64_AVX,X64_AVX16; RUN: llc < %s -disable-peephole -mtriple=i686-apple-darwin9 -mattr=+avx512f | FileCheck %s --check-prefixes=X86_AVX,X86_AVX5127; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin9 -mattr=+avx512f | FileCheck %s --check-prefixes=X64_AVX,X64_AVX5128 9define i16 @test1(float %f) nounwind {10; X86-LABEL: test1:11; X86: ## %bb.0:12; X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero13; X86-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm014; X86-NEXT: mulss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm015; X86-NEXT: xorps %xmm1, %xmm116; X86-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]17; X86-NEXT: minss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm018; X86-NEXT: maxss %xmm1, %xmm019; X86-NEXT: cvttss2si %xmm0, %eax20; X86-NEXT: ## kill: def $ax killed $ax killed $eax21; X86-NEXT: retl22;23; X64-LABEL: test1:24; X64: ## %bb.0:25; X64-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm026; X64-NEXT: mulss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm027; X64-NEXT: xorps %xmm1, %xmm128; X64-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]29; X64-NEXT: minss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm030; X64-NEXT: maxss %xmm1, %xmm031; X64-NEXT: cvttss2si %xmm0, %eax32; X64-NEXT: ## kill: def $ax killed $ax killed $eax33; X64-NEXT: retq34;35; X86_AVX1-LABEL: test1:36; X86_AVX1: ## %bb.0:37; X86_AVX1-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero38; X86_AVX1-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm039; X86_AVX1-NEXT: vmulss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm040; X86_AVX1-NEXT: vxorps %xmm1, %xmm1, %xmm141; X86_AVX1-NEXT: vmovss {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]42; X86_AVX1-NEXT: vminss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm043; X86_AVX1-NEXT: vmaxss %xmm1, %xmm0, %xmm044; X86_AVX1-NEXT: vcvttss2si %xmm0, %eax45; X86_AVX1-NEXT: ## kill: def $ax killed $ax killed $eax46; X86_AVX1-NEXT: retl47;48; X64_AVX1-LABEL: test1:49; X64_AVX1: ## %bb.0:50; X64_AVX1-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm051; X64_AVX1-NEXT: vmulss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm052; X64_AVX1-NEXT: vxorps %xmm1, %xmm1, %xmm153; X64_AVX1-NEXT: vmovss {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]54; X64_AVX1-NEXT: vminss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm055; X64_AVX1-NEXT: vmaxss %xmm1, %xmm0, %xmm056; X64_AVX1-NEXT: vcvttss2si %xmm0, %eax57; X64_AVX1-NEXT: ## kill: def $ax killed $ax killed $eax58; X64_AVX1-NEXT: retq59;60; X86_AVX512-LABEL: test1:61; X86_AVX512: ## %bb.0:62; X86_AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero63; X86_AVX512-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm064; X86_AVX512-NEXT: vmulss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm065; X86_AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm166; X86_AVX512-NEXT: vmovss {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]67; X86_AVX512-NEXT: vminss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm068; X86_AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm169; X86_AVX512-NEXT: vmaxss %xmm1, %xmm0, %xmm070; X86_AVX512-NEXT: vcvttss2si %xmm0, %eax71; X86_AVX512-NEXT: ## kill: def $ax killed $ax killed $eax72; X86_AVX512-NEXT: retl73;74; X64_AVX512-LABEL: test1:75; X64_AVX512: ## %bb.0:76; X64_AVX512-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm077; X64_AVX512-NEXT: vmulss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm078; X64_AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm179; X64_AVX512-NEXT: vmovss {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]80; X64_AVX512-NEXT: vminss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm081; X64_AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm182; X64_AVX512-NEXT: vmaxss %xmm1, %xmm0, %xmm083; X64_AVX512-NEXT: vcvttss2si %xmm0, %eax84; X64_AVX512-NEXT: ## kill: def $ax killed $ax killed $eax85; X64_AVX512-NEXT: retq86 %tmp = insertelement <4 x float> undef, float %f, i32 0 ; <<4 x float>> [#uses=1]87 %tmp10 = insertelement <4 x float> %tmp, float 0.000000e+00, i32 1 ; <<4 x float>> [#uses=1]88 %tmp11 = insertelement <4 x float> %tmp10, float 0.000000e+00, i32 2 ; <<4 x float>> [#uses=1]89 %tmp12 = insertelement <4 x float> %tmp11, float 0.000000e+00, i32 3 ; <<4 x float>> [#uses=1]90 %tmp28 = tail call <4 x float> @llvm.x86.sse.sub.ss( <4 x float> %tmp12, <4 x float> < float 1.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00 > ) ; <<4 x float>> [#uses=1]91 %tmp37 = tail call <4 x float> @llvm.x86.sse.mul.ss( <4 x float> %tmp28, <4 x float> < float 5.000000e-01, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00 > ) ; <<4 x float>> [#uses=1]92 %tmp48 = tail call <4 x float> @llvm.x86.sse.min.ss( <4 x float> %tmp37, <4 x float> < float 6.553500e+04, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00 > ) ; <<4 x float>> [#uses=1]93 %tmp59 = tail call <4 x float> @llvm.x86.sse.max.ss( <4 x float> %tmp48, <4 x float> zeroinitializer ) ; <<4 x float>> [#uses=1]94 %tmp.upgrd.1 = tail call i32 @llvm.x86.sse.cvttss2si( <4 x float> %tmp59 ) ; <i32> [#uses=1]95 %tmp69 = trunc i32 %tmp.upgrd.1 to i16 ; <i16> [#uses=1]96 ret i16 %tmp6997}98 99define i16 @test2(float %f) nounwind {100; X86-LABEL: test2:101; X86: ## %bb.0:102; X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero103; X86-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0104; X86-NEXT: mulss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0105; X86-NEXT: minss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0106; X86-NEXT: xorps %xmm1, %xmm1107; X86-NEXT: maxss %xmm1, %xmm0108; X86-NEXT: cvttss2si %xmm0, %eax109; X86-NEXT: ## kill: def $ax killed $ax killed $eax110; X86-NEXT: retl111;112; X64-LABEL: test2:113; X64: ## %bb.0:114; X64-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0115; X64-NEXT: mulss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0116; X64-NEXT: minss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0117; X64-NEXT: xorps %xmm1, %xmm1118; X64-NEXT: maxss %xmm1, %xmm0119; X64-NEXT: cvttss2si %xmm0, %eax120; X64-NEXT: ## kill: def $ax killed $ax killed $eax121; X64-NEXT: retq122;123; X86_AVX-LABEL: test2:124; X86_AVX: ## %bb.0:125; X86_AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero126; X86_AVX-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0127; X86_AVX-NEXT: vmulss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0128; X86_AVX-NEXT: vminss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0129; X86_AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1130; X86_AVX-NEXT: vmaxss %xmm1, %xmm0, %xmm0131; X86_AVX-NEXT: vcvttss2si %xmm0, %eax132; X86_AVX-NEXT: ## kill: def $ax killed $ax killed $eax133; X86_AVX-NEXT: retl134;135; X64_AVX-LABEL: test2:136; X64_AVX: ## %bb.0:137; X64_AVX-NEXT: vaddss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0138; X64_AVX-NEXT: vmulss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0139; X64_AVX-NEXT: vminss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0140; X64_AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1141; X64_AVX-NEXT: vmaxss %xmm1, %xmm0, %xmm0142; X64_AVX-NEXT: vcvttss2si %xmm0, %eax143; X64_AVX-NEXT: ## kill: def $ax killed $ax killed $eax144; X64_AVX-NEXT: retq145 %tmp28 = fsub float %f, 1.000000e+00 ; <float> [#uses=1]146 %tmp37 = fmul float %tmp28, 5.000000e-01 ; <float> [#uses=1]147 %tmp375 = insertelement <4 x float> undef, float %tmp37, i32 0 ; <<4 x float>> [#uses=1]148 %tmp48 = tail call <4 x float> @llvm.x86.sse.min.ss( <4 x float> %tmp375, <4 x float> < float 6.553500e+04, float undef, float undef, float undef > ) ; <<4 x float>> [#uses=1]149 %tmp59 = tail call <4 x float> @llvm.x86.sse.max.ss( <4 x float> %tmp48, <4 x float> < float 0.000000e+00, float undef, float undef, float undef > ) ; <<4 x float>> [#uses=1]150 %tmp = tail call i32 @llvm.x86.sse.cvttss2si( <4 x float> %tmp59 ) ; <i32> [#uses=1]151 %tmp69 = trunc i32 %tmp to i16 ; <i16> [#uses=1]152 ret i16 %tmp69153}154 155declare <4 x float> @llvm.x86.sse.sub.ss(<4 x float>, <4 x float>)156 157declare <4 x float> @llvm.x86.sse.mul.ss(<4 x float>, <4 x float>)158 159declare <4 x float> @llvm.x86.sse.min.ss(<4 x float>, <4 x float>)160 161declare <4 x float> @llvm.x86.sse.max.ss(<4 x float>, <4 x float>)162 163declare i32 @llvm.x86.sse.cvttss2si(<4 x float>)164 165declare <4 x float> @llvm.x86.sse41.round.ss(<4 x float>, <4 x float>, i32)166 167declare <4 x float> @f()168 169define <4 x float> @test3(<4 x float> %A, ptr%b, i32 %C) nounwind {170; X86-LABEL: test3:171; X86: ## %bb.0:172; X86-NEXT: movl {{[0-9]+}}(%esp), %eax173; X86-NEXT: roundss $4, (%eax), %xmm0174; X86-NEXT: retl175;176; X64-LABEL: test3:177; X64: ## %bb.0:178; X64-NEXT: roundss $4, (%rdi), %xmm0179; X64-NEXT: retq180;181; X86_AVX-LABEL: test3:182; X86_AVX: ## %bb.0:183; X86_AVX-NEXT: movl {{[0-9]+}}(%esp), %eax184; X86_AVX-NEXT: vroundss $4, (%eax), %xmm0, %xmm0185; X86_AVX-NEXT: retl186;187; X64_AVX-LABEL: test3:188; X64_AVX: ## %bb.0:189; X64_AVX-NEXT: vroundss $4, (%rdi), %xmm0, %xmm0190; X64_AVX-NEXT: retq191 %a = load float , ptr%b192 %B = insertelement <4 x float> undef, float %a, i32 0193 %X = call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> %A, <4 x float> %B, i32 4)194 ret <4 x float> %X195}196 197define <4 x float> @test4(<4 x float> %A, ptr%b, i32 %C) nounwind {198; X86-LABEL: test4:199; X86: ## %bb.0:200; X86-NEXT: subl $28, %esp201; X86-NEXT: movl {{[0-9]+}}(%esp), %eax202; X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero203; X86-NEXT: movaps %xmm0, (%esp) ## 16-byte Spill204; X86-NEXT: calll _f205; X86-NEXT: roundss $4, (%esp), %xmm0 ## 16-byte Folded Reload206; X86-NEXT: addl $28, %esp207; X86-NEXT: retl208;209; X64-LABEL: test4:210; X64: ## %bb.0:211; X64-NEXT: subq $24, %rsp212; X64-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero213; X64-NEXT: movaps %xmm0, (%rsp) ## 16-byte Spill214; X64-NEXT: callq _f215; X64-NEXT: roundss $4, (%rsp), %xmm0 ## 16-byte Folded Reload216; X64-NEXT: addq $24, %rsp217; X64-NEXT: retq218;219; X86_AVX-LABEL: test4:220; X86_AVX: ## %bb.0:221; X86_AVX-NEXT: subl $28, %esp222; X86_AVX-NEXT: movl {{[0-9]+}}(%esp), %eax223; X86_AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero224; X86_AVX-NEXT: vmovaps %xmm0, (%esp) ## 16-byte Spill225; X86_AVX-NEXT: calll _f226; X86_AVX-NEXT: vroundss $4, (%esp), %xmm0, %xmm0 ## 16-byte Folded Reload227; X86_AVX-NEXT: addl $28, %esp228; X86_AVX-NEXT: retl229;230; X64_AVX-LABEL: test4:231; X64_AVX: ## %bb.0:232; X64_AVX-NEXT: subq $24, %rsp233; X64_AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero234; X64_AVX-NEXT: vmovaps %xmm0, (%rsp) ## 16-byte Spill235; X64_AVX-NEXT: callq _f236; X64_AVX-NEXT: vroundss $4, (%rsp), %xmm0, %xmm0 ## 16-byte Folded Reload237; X64_AVX-NEXT: addq $24, %rsp238; X64_AVX-NEXT: retq239 %a = load float , ptr%b240 %B = insertelement <4 x float> undef, float %a, i32 0241 %q = call <4 x float> @f()242 %X = call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> %q, <4 x float> %B, i32 4)243 ret <4 x float> %X244}245 246; PR13576247define <2 x double> @test5() nounwind uwtable readnone noinline {248; X86-LABEL: test5:249; X86: ## %bb.0: ## %entry250; X86-NEXT: movaps {{.*#+}} xmm0 = [1.28E+2,1.23321E+2]251; X86-NEXT: retl252;253; X64-LABEL: test5:254; X64: ## %bb.0: ## %entry255; X64-NEXT: movaps {{.*#+}} xmm0 = [1.28E+2,1.23321E+2]256; X64-NEXT: retq257;258; X86_AVX-LABEL: test5:259; X86_AVX: ## %bb.0: ## %entry260; X86_AVX-NEXT: vmovaps {{.*#+}} xmm0 = [1.28E+2,1.23321E+2]261; X86_AVX-NEXT: retl262;263; X64_AVX-LABEL: test5:264; X64_AVX: ## %bb.0: ## %entry265; X64_AVX-NEXT: vmovaps {{.*#+}} xmm0 = [1.28E+2,1.23321E+2]266; X64_AVX-NEXT: retq267entry:268 %0 = tail call <2 x double> @llvm.x86.sse2.cvtsi2sd(<2 x double> <double 4.569870e+02, double 1.233210e+02>, i32 128) nounwind readnone269 ret <2 x double> %0270}271 272declare <2 x double> @llvm.x86.sse2.cvtsi2sd(<2 x double>, i32) nounwind readnone273 274define <4 x float> @minss_fold(ptr %x, <4 x float> %y) {275; X86-LABEL: minss_fold:276; X86: ## %bb.0: ## %entry277; X86-NEXT: movl {{[0-9]+}}(%esp), %eax278; X86-NEXT: minss (%eax), %xmm0279; X86-NEXT: retl280;281; X64-LABEL: minss_fold:282; X64: ## %bb.0: ## %entry283; X64-NEXT: minss (%rdi), %xmm0284; X64-NEXT: retq285;286; X86_AVX-LABEL: minss_fold:287; X86_AVX: ## %bb.0: ## %entry288; X86_AVX-NEXT: movl {{[0-9]+}}(%esp), %eax289; X86_AVX-NEXT: vminss (%eax), %xmm0, %xmm0290; X86_AVX-NEXT: retl291;292; X64_AVX-LABEL: minss_fold:293; X64_AVX: ## %bb.0: ## %entry294; X64_AVX-NEXT: vminss (%rdi), %xmm0, %xmm0295; X64_AVX-NEXT: retq296entry:297 %0 = load float, ptr %x, align 1298 %vecinit.i = insertelement <4 x float> undef, float %0, i32 0299 %vecinit2.i = insertelement <4 x float> %vecinit.i, float 0.000000e+00, i32 1300 %vecinit3.i = insertelement <4 x float> %vecinit2.i, float 0.000000e+00, i32 2301 %vecinit4.i = insertelement <4 x float> %vecinit3.i, float 0.000000e+00, i32 3302 %1 = tail call <4 x float> @llvm.x86.sse.min.ss(<4 x float> %y, <4 x float> %vecinit4.i)303 ret <4 x float> %1304}305 306define <4 x float> @maxss_fold(ptr %x, <4 x float> %y) {307; X86-LABEL: maxss_fold:308; X86: ## %bb.0: ## %entry309; X86-NEXT: movl {{[0-9]+}}(%esp), %eax310; X86-NEXT: maxss (%eax), %xmm0311; X86-NEXT: retl312;313; X64-LABEL: maxss_fold:314; X64: ## %bb.0: ## %entry315; X64-NEXT: maxss (%rdi), %xmm0316; X64-NEXT: retq317;318; X86_AVX-LABEL: maxss_fold:319; X86_AVX: ## %bb.0: ## %entry320; X86_AVX-NEXT: movl {{[0-9]+}}(%esp), %eax321; X86_AVX-NEXT: vmaxss (%eax), %xmm0, %xmm0322; X86_AVX-NEXT: retl323;324; X64_AVX-LABEL: maxss_fold:325; X64_AVX: ## %bb.0: ## %entry326; X64_AVX-NEXT: vmaxss (%rdi), %xmm0, %xmm0327; X64_AVX-NEXT: retq328entry:329 %0 = load float, ptr %x, align 1330 %vecinit.i = insertelement <4 x float> undef, float %0, i32 0331 %vecinit2.i = insertelement <4 x float> %vecinit.i, float 0.000000e+00, i32 1332 %vecinit3.i = insertelement <4 x float> %vecinit2.i, float 0.000000e+00, i32 2333 %vecinit4.i = insertelement <4 x float> %vecinit3.i, float 0.000000e+00, i32 3334 %1 = tail call <4 x float> @llvm.x86.sse.max.ss(<4 x float> %y, <4 x float> %vecinit4.i)335 ret <4 x float> %1336}337 338define <4 x float> @cmpss_fold(ptr %x, <4 x float> %y) {339; X86-LABEL: cmpss_fold:340; X86: ## %bb.0: ## %entry341; X86-NEXT: movl {{[0-9]+}}(%esp), %eax342; X86-NEXT: cmpeqss (%eax), %xmm0343; X86-NEXT: retl344;345; X64-LABEL: cmpss_fold:346; X64: ## %bb.0: ## %entry347; X64-NEXT: cmpeqss (%rdi), %xmm0348; X64-NEXT: retq349;350; X86_AVX-LABEL: cmpss_fold:351; X86_AVX: ## %bb.0: ## %entry352; X86_AVX-NEXT: movl {{[0-9]+}}(%esp), %eax353; X86_AVX-NEXT: vcmpeqss (%eax), %xmm0, %xmm0354; X86_AVX-NEXT: retl355;356; X64_AVX-LABEL: cmpss_fold:357; X64_AVX: ## %bb.0: ## %entry358; X64_AVX-NEXT: vcmpeqss (%rdi), %xmm0, %xmm0359; X64_AVX-NEXT: retq360entry:361 %0 = load float, ptr %x, align 1362 %vecinit.i = insertelement <4 x float> undef, float %0, i32 0363 %vecinit2.i = insertelement <4 x float> %vecinit.i, float 0.000000e+00, i32 1364 %vecinit3.i = insertelement <4 x float> %vecinit2.i, float 0.000000e+00, i32 2365 %vecinit4.i = insertelement <4 x float> %vecinit3.i, float 0.000000e+00, i32 3366 %1 = tail call <4 x float> @llvm.x86.sse.cmp.ss(<4 x float> %y, <4 x float> %vecinit4.i, i8 0)367 ret <4 x float> %1368}369declare <4 x float> @llvm.x86.sse.cmp.ss(<4 x float>, <4 x float>, i8) nounwind readnone370 371 372define <4 x float> @double_fold(ptr %x, <4 x float> %y) {373; X86-LABEL: double_fold:374; X86: ## %bb.0: ## %entry375; X86-NEXT: movl {{[0-9]+}}(%esp), %eax376; X86-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero377; X86-NEXT: movaps %xmm0, %xmm2378; X86-NEXT: minss %xmm1, %xmm2379; X86-NEXT: maxss %xmm1, %xmm0380; X86-NEXT: addps %xmm2, %xmm0381; X86-NEXT: retl382;383; X64-LABEL: double_fold:384; X64: ## %bb.0: ## %entry385; X64-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero386; X64-NEXT: movaps %xmm0, %xmm2387; X64-NEXT: minss %xmm1, %xmm2388; X64-NEXT: maxss %xmm1, %xmm0389; X64-NEXT: addps %xmm2, %xmm0390; X64-NEXT: retq391;392; X86_AVX-LABEL: double_fold:393; X86_AVX: ## %bb.0: ## %entry394; X86_AVX-NEXT: movl {{[0-9]+}}(%esp), %eax395; X86_AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero396; X86_AVX-NEXT: vminss %xmm1, %xmm0, %xmm2397; X86_AVX-NEXT: vmaxss %xmm1, %xmm0, %xmm0398; X86_AVX-NEXT: vaddps %xmm0, %xmm2, %xmm0399; X86_AVX-NEXT: retl400;401; X64_AVX-LABEL: double_fold:402; X64_AVX: ## %bb.0: ## %entry403; X64_AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero404; X64_AVX-NEXT: vminss %xmm1, %xmm0, %xmm2405; X64_AVX-NEXT: vmaxss %xmm1, %xmm0, %xmm0406; X64_AVX-NEXT: vaddps %xmm0, %xmm2, %xmm0407; X64_AVX-NEXT: retq408entry:409 %0 = load float, ptr %x, align 1410 %vecinit.i = insertelement <4 x float> undef, float %0, i32 0411 %1 = tail call <4 x float> @llvm.x86.sse.min.ss(<4 x float> %y, <4 x float> %vecinit.i)412 %2 = tail call <4 x float> @llvm.x86.sse.max.ss(<4 x float> %y, <4 x float> %vecinit.i)413 %3 = fadd <4 x float> %1, %2414 ret <4 x float> %3415}416