From aecc6a20776b59fb79777898b078cc8165be4fa2 Mon Sep 17 00:00:00 2001 From: Igor Breger Date: Mon, 14 Mar 2016 10:26:39 +0000 Subject: [PATCH] AVX512: icmp operation should be always lowered to CMPM (AVX-512) instruction on SKX. implemented by delena Differential Revision: http://reviews.llvm.org/D18054 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@263417 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/X86/X86ISelLowering.cpp | 43 +-- test/CodeGen/X86/avx512-ext.ll | 88 ++++- test/CodeGen/X86/avx512-vec-cmp.ll | 553 +++++++++++++++++++++++++---- 3 files changed, 590 insertions(+), 94 deletions(-) diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index 43784a1c454..ba443dcbea9 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -1445,6 +1445,8 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, setOperationAction(ISD::UINT_TO_FP, MVT::v4i32, Legal); setOperationAction(ISD::FP_TO_SINT, MVT::v4i32, Legal); setOperationAction(ISD::FP_TO_UINT, MVT::v4i32, Legal); + setOperationAction(ISD::ZERO_EXTEND, MVT::v4i32, Custom); + setOperationAction(ISD::ZERO_EXTEND, MVT::v2i64, Custom); } setOperationAction(ISD::TRUNCATE, MVT::v8i1, Custom); setOperationAction(ISD::TRUNCATE, MVT::v16i1, Custom); @@ -1884,7 +1886,8 @@ X86TargetLowering::getPreferredVectorAction(EVT VT) const { return TargetLoweringBase::getPreferredVectorAction(VT); } -EVT X86TargetLowering::getSetCCResultType(const DataLayout &DL, LLVMContext &, +EVT X86TargetLowering::getSetCCResultType(const DataLayout &DL, + LLVMContext& Context, EVT VT) const { if (!VT.isVector()) return Subtarget.hasAVX512() ? MVT::i1: MVT::i8; @@ -1892,7 +1895,7 @@ EVT X86TargetLowering::getSetCCResultType(const DataLayout &DL, LLVMContext &, if (VT.isSimple()) { MVT VVT = VT.getSimpleVT(); const unsigned NumElts = VVT.getVectorNumElements(); - const MVT EltVT = VVT.getVectorElementType(); + MVT EltVT = VVT.getVectorElementType(); if (VVT.is512BitVector()) { if (Subtarget.hasAVX512()) if (EltVT == MVT::i32 || EltVT == MVT::i64 || @@ -1909,23 +1912,20 @@ EVT X86TargetLowering::getSetCCResultType(const DataLayout &DL, LLVMContext &, } } - if (VVT.is256BitVector() || VVT.is128BitVector()) { - if (Subtarget.hasVLX()) - if (EltVT == MVT::i32 || EltVT == MVT::i64 || - EltVT == MVT::f32 || EltVT == MVT::f64) - switch(NumElts) { - case 2: return MVT::v2i1; - case 4: return MVT::v4i1; - case 8: return MVT::v8i1; - } - if (Subtarget.hasBWI() && Subtarget.hasVLX()) - if (EltVT == MVT::i8 || EltVT == MVT::i16) - switch(NumElts) { - case 8: return MVT::v8i1; - case 16: return MVT::v16i1; - case 32: return MVT::v32i1; - } + if (Subtarget.hasBWI() && Subtarget.hasVLX()) + return MVT::getVectorVT(MVT::i1, NumElts); + + if (!isTypeLegal(VT) && getTypeAction(Context, VT) == TypePromoteInteger) { + EVT LegalVT = getTypeToTransformTo(Context, VT); + EltVT = LegalVT.getVectorElementType().getSimpleVT(); } + + if (Subtarget.hasVLX() && EltVT.getSizeInBits() >= 32) + switch(NumElts) { + case 2: return MVT::v2i1; + case 4: return MVT::v4i1; + case 8: return MVT::v8i1; + } } return VT.changeVectorElementTypeToInteger(); @@ -6750,15 +6750,16 @@ X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const { // it to i32 first. if (ExtVT == MVT::i16 || ExtVT == MVT::i8) { Item = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Item); - if (VT.is256BitVector()) { + if (VT.getSizeInBits() >= 256) { + MVT ShufVT = MVT::getVectorVT(MVT::i32, VT.getSizeInBits()/32); if (Subtarget.hasAVX()) { - Item = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v8i32, Item); + Item = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, ShufVT, Item); Item = getShuffleVectorZeroOrUndef(Item, 0, true, Subtarget, DAG); } else { // Without AVX, we need to extend to a 128-bit vector and then // insert into the 256-bit vector. Item = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v4i32, Item); - SDValue ZeroVec = getZeroVector(MVT::v8i32, Subtarget, DAG, dl); + SDValue ZeroVec = getZeroVector(ShufVT, Subtarget, DAG, dl); Item = Insert128BitVector(ZeroVec, Item, 0, DAG, dl); } } else { diff --git a/test/CodeGen/X86/avx512-ext.ll b/test/CodeGen/X86/avx512-ext.ll index 24397111c83..fc4fcf92524 100644 --- a/test/CodeGen/X86/avx512-ext.ll +++ b/test/CodeGen/X86/avx512-ext.ll @@ -1900,6 +1900,67 @@ define <64 x i8> @zext_64xi1_to_64xi8(<64 x i8> %x, <64 x i8> %y) #0 { ret <64 x i8> %1 } +define <32 x i16> @zext_32xi1_to_32xi16(<32 x i16> %x, <32 x i16> %y) #0 { +; KNL-LABEL: zext_32xi1_to_32xi16: +; KNL: ## BB#0: +; KNL-NEXT: vpcmpeqw %ymm2, %ymm0, %ymm0 +; KNL-NEXT: vmovdqa {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] +; KNL-NEXT: vpand %ymm2, %ymm0, %ymm0 +; KNL-NEXT: vpcmpeqw %ymm3, %ymm1, %ymm1 +; KNL-NEXT: vpand %ymm2, %ymm1, %ymm1 +; KNL-NEXT: retq +; +; SKX-LABEL: zext_32xi1_to_32xi16: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqw %zmm1, %zmm0, %k1 +; SKX-NEXT: vmovdqu16 {{.*}}(%rip), %zmm0 {%k1} {z} +; SKX-NEXT: retq + %mask = icmp eq <32 x i16> %x, %y + %1 = zext <32 x i1> %mask to <32 x i16> + ret <32 x i16> %1 +} + +define <16 x i16> @zext_16xi1_to_16xi16(<16 x i16> %x, <16 x i16> %y) #0 { +; KNL-LABEL: zext_16xi1_to_16xi16: +; KNL: ## BB#0: +; KNL-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0 +; KNL-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0 +; KNL-NEXT: retq +; +; SKX-LABEL: zext_16xi1_to_16xi16: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqw %ymm1, %ymm0, %k1 +; SKX-NEXT: vmovdqu16 {{.*}}(%rip), %ymm0 {%k1} {z} +; SKX-NEXT: retq + %mask = icmp eq <16 x i16> %x, %y + %1 = zext <16 x i1> %mask to <16 x i16> + ret <16 x i16> %1 +} + + +define <32 x i8> @zext_32xi1_to_32xi8(<32 x i16> %x, <32 x i16> %y) #0 { +; KNL-LABEL: zext_32xi1_to_32xi8: +; KNL: ## BB#0: +; KNL-NEXT: vpcmpeqw %ymm2, %ymm0, %ymm0 +; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 +; KNL-NEXT: vpmovdb %zmm0, %xmm0 +; KNL-NEXT: vpcmpeqw %ymm3, %ymm1, %ymm1 +; KNL-NEXT: vpmovsxwd %ymm1, %zmm1 +; KNL-NEXT: vpmovdb %zmm1, %xmm1 +; KNL-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; KNL-NEXT: vandps {{.*}}(%rip), %ymm0, %ymm0 +; KNL-NEXT: retq +; +; SKX-LABEL: zext_32xi1_to_32xi8: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqw %zmm1, %zmm0, %k1 +; SKX-NEXT: vmovdqu8 {{.*}}(%rip), %ymm0 {%k1} {z} +; SKX-NEXT: retq + %mask = icmp eq <32 x i16> %x, %y + %1 = zext <32 x i1> %mask to <32 x i8> + ret <32 x i8> %1 +} + define <4 x i32> @zext_4xi1_to_4x32(<4 x i8> %x, <4 x i8> %y) #0 { ; KNL-LABEL: zext_4xi1_to_4x32: ; KNL: ## BB#0: @@ -1916,10 +1977,33 @@ define <4 x i32> @zext_4xi1_to_4x32(<4 x i8> %x, <4 x i8> %y) #0 { ; SKX-NEXT: vmovdqa64 {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0] ; SKX-NEXT: vpandq %xmm2, %xmm1, %xmm1 ; SKX-NEXT: vpandq %xmm2, %xmm0, %xmm0 -; SKX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 -; SKX-NEXT: vpandd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; SKX-NEXT: vpcmpeqd %xmm1, %xmm0, %k1 +; SKX-NEXT: vpbroadcastd {{.*}}(%rip), %xmm0 {%k1} {z} ; SKX-NEXT: retq %mask = icmp eq <4 x i8> %x, %y %1 = zext <4 x i1> %mask to <4 x i32> ret <4 x i32> %1 } + +define <2 x i64> @zext_2xi1_to_2xi64(<2 x i8> %x, <2 x i8> %y) #0 { +; KNL-LABEL: zext_2xi1_to_2xi64: +; KNL: ## BB#0: +; KNL-NEXT: vmovdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] +; KNL-NEXT: vpand %xmm2, %xmm1, %xmm1 +; KNL-NEXT: vpand %xmm2, %xmm0, %xmm0 +; KNL-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 +; KNL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: zext_2xi1_to_2xi64: +; SKX: ## BB#0: +; SKX-NEXT: vmovdqa64 {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] +; SKX-NEXT: vpandq %xmm2, %xmm1, %xmm1 +; SKX-NEXT: vpandq %xmm2, %xmm0, %xmm0 +; SKX-NEXT: vpcmpeqq %xmm1, %xmm0, %k1 +; SKX-NEXT: vmovdqa64 {{.*}}(%rip), %xmm0 {%k1} {z} +; SKX-NEXT: retq + %mask = icmp eq <2 x i8> %x, %y + %1 = zext <2 x i1> %mask to <2 x i64> + ret <2 x i64> %1 +} diff --git a/test/CodeGen/X86/avx512-vec-cmp.ll b/test/CodeGen/X86/avx512-vec-cmp.ll index 08d1e24ed82..0727a0956a6 100644 --- a/test/CodeGen/X86/avx512-vec-cmp.ll +++ b/test/CodeGen/X86/avx512-vec-cmp.ll @@ -1,3 +1,4 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s --check-prefix=KNL ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck %s --check-prefix=SKX @@ -8,6 +9,13 @@ define <16 x float> @test1(<16 x float> %x, <16 x float> %y) nounwind { ; KNL-NEXT: vmovaps %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test1: +; SKX: ## BB#0: +; SKX-NEXT: vcmpleps %zmm1, %zmm0, %k1 +; SKX-NEXT: vmovaps %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = fcmp ole <16 x float> %x, %y %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %y ret <16 x float> %max @@ -20,6 +28,13 @@ define <8 x double> @test2(<8 x double> %x, <8 x double> %y) nounwind { ; KNL-NEXT: vmovapd %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test2: +; SKX: ## BB#0: +; SKX-NEXT: vcmplepd %zmm1, %zmm0, %k1 +; SKX-NEXT: vmovapd %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = fcmp ole <8 x double> %x, %y %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %y ret <8 x double> %max @@ -32,6 +47,13 @@ define <16 x i32> @test3(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %yp) nounwin ; KNL-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test3: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqd (%rdi), %zmm0, %k1 +; SKX-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <16 x i32>, <16 x i32>* %yp, align 4 %mask = icmp eq <16 x i32> %x, %y %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 @@ -45,6 +67,13 @@ define <16 x i32> @test4_unsigned(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) ; KNL-NEXT: vmovdqa32 %zmm2, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test4_unsigned: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpnltud %zmm1, %zmm0, %k1 +; SKX-NEXT: vmovdqa32 %zmm2, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = icmp uge <16 x i32> %x, %y %max = select <16 x i1> %mask, <16 x i32> %x1, <16 x i32> %y ret <16 x i32> %max @@ -57,6 +86,13 @@ define <8 x i64> @test5(<8 x i64> %x, <8 x i64> %y) nounwind { ; KNL-NEXT: vmovdqa64 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test5: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqq %zmm1, %zmm0, %k1 +; SKX-NEXT: vmovdqa64 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = icmp eq <8 x i64> %x, %y %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %y ret <8 x i64> %max @@ -69,6 +105,13 @@ define <8 x i64> @test6_unsigned(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1) noun ; KNL-NEXT: vmovdqa64 %zmm2, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test6_unsigned: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpnleuq %zmm1, %zmm0, %k1 +; SKX-NEXT: vmovdqa64 %zmm2, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = icmp ugt <8 x i64> %x, %y %max = select <8 x i1> %mask, <8 x i64> %x1, <8 x i64> %y ret <8 x i64> %max @@ -81,13 +124,14 @@ define <4 x float> @test7(<4 x float> %a, <4 x float> %b) { ; KNL-NEXT: vcmpltps %xmm2, %xmm0, %xmm2 ; KNL-NEXT: vblendvps %xmm2, %xmm0, %xmm1, %xmm0 ; KNL-NEXT: retq +; ; SKX-LABEL: test7: ; SKX: ## BB#0: -; SKX: vxorps %xmm2, %xmm2, %xmm2 -; SKX: vcmpltps %xmm2, %xmm0, %k1 -; SKX: vmovaps %xmm0, %xmm1 {%k1} -; SKX: vmovaps %zmm1, %zmm0 -; SKX: retq +; SKX-NEXT: vxorps %xmm2, %xmm2, %xmm2 +; SKX-NEXT: vcmpltps %xmm2, %xmm0, %k1 +; SKX-NEXT: vmovaps %xmm0, %xmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = fcmp olt <4 x float> %a, zeroinitializer %c = select <4 x i1>%mask, <4 x float>%a, <4 x float>%b @@ -101,13 +145,14 @@ define <2 x double> @test8(<2 x double> %a, <2 x double> %b) { ; KNL-NEXT: vcmpltpd %xmm2, %xmm0, %xmm2 ; KNL-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0 ; KNL-NEXT: retq +; ; SKX-LABEL: test8: ; SKX: ## BB#0: -; SKX: vxorpd %xmm2, %xmm2, %xmm2 -; SKX: vcmpltpd %xmm2, %xmm0, %k1 -; SKX: vmovapd %xmm0, %xmm1 {%k1} -; SKX: vmovaps %zmm1, %zmm0 -; SKX: retq +; SKX-NEXT: vxorpd %xmm2, %xmm2, %xmm2 +; SKX-NEXT: vcmpltpd %xmm2, %xmm0, %k1 +; SKX-NEXT: vmovapd %xmm0, %xmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = fcmp olt <2 x double> %a, zeroinitializer %c = select <2 x i1>%mask, <2 x double>%a, <2 x double>%b ret <2 x double>%c @@ -119,6 +164,13 @@ define <8 x i32> @test9(<8 x i32> %x, <8 x i32> %y) nounwind { ; KNL-NEXT: vpcmpeqd %zmm1, %zmm0, %k1 ; KNL-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ; KNL-NEXT: retq +; +; SKX-LABEL: test9: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqd %ymm1, %ymm0, %k1 +; SKX-NEXT: vmovdqa32 %ymm0, %ymm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = icmp eq <8 x i32> %x, %y %max = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y ret <8 x i32> %max @@ -130,12 +182,13 @@ define <8 x float> @test10(<8 x float> %x, <8 x float> %y) nounwind { ; KNL-NEXT: vcmpeqps %zmm1, %zmm0, %k1 ; KNL-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ; KNL-NEXT: retq +; ; SKX-LABEL: test10: ; SKX: ## BB#0: -; SKX: vcmpeqps %ymm1, %ymm0, %k1 -; SKX: vmovaps %ymm0, %ymm1 {%k1} -; SKX: vmovaps %zmm1, %zmm0 -; SKX: retq +; SKX-NEXT: vcmpeqps %ymm1, %ymm0, %k1 +; SKX-NEXT: vmovaps %ymm0, %ymm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = fcmp oeq <8 x float> %x, %y %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %y @@ -147,6 +200,11 @@ define <8 x i32> @test11_unsigned(<8 x i32> %x, <8 x i32> %y) nounwind { ; KNL: ## BB#0: ; KNL-NEXT: vpmaxud %ymm1, %ymm0, %ymm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test11_unsigned: +; SKX: ## BB#0: +; SKX-NEXT: vpmaxud %ymm1, %ymm0, %ymm0 +; SKX-NEXT: retq %mask = icmp ugt <8 x i32> %x, %y %max = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y ret <8 x i32> %max @@ -160,6 +218,14 @@ define i16 @test12(<16 x i64> %a, <16 x i64> %b) nounwind { ; KNL-NEXT: kunpckbw %k0, %k1, %k0 ; KNL-NEXT: kmovw %k0, %eax ; KNL-NEXT: retq +; +; SKX-LABEL: test12: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqq %zmm2, %zmm0, %k0 +; SKX-NEXT: vpcmpeqq %zmm3, %zmm1, %k1 +; SKX-NEXT: kunpckbw %k0, %k1, %k0 +; SKX-NEXT: kmovw %k0, %eax +; SKX-NEXT: retq %res = icmp eq <16 x i64> %a, %b %res1 = bitcast <16 x i1> %res to i16 ret i16 %res1 @@ -197,6 +263,12 @@ define <16 x i32> @test13(<16 x float>%a, <16 x float>%b) ; KNL-NEXT: vcmpeqps %zmm1, %zmm0, %k1 ; KNL-NEXT: vpbroadcastd {{.*}}(%rip), %zmm0 {%k1} {z} ; KNL-NEXT: retq +; +; SKX-LABEL: test13: +; SKX: ## BB#0: +; SKX-NEXT: vcmpeqps %zmm1, %zmm0, %k1 +; SKX-NEXT: vpbroadcastd {{.*}}(%rip), %zmm0 {%k1} {z} +; SKX-NEXT: retq { %cmpvector_i = fcmp oeq <16 x float> %a, %b %conv = zext <16 x i1> %cmpvector_i to <16 x i32> @@ -212,6 +284,15 @@ define <16 x i32> @test14(<16 x i32>%a, <16 x i32>%b) { ; KNL-NEXT: knotw %k0, %k1 ; KNL-NEXT: vmovdqa32 %zmm1, %zmm0 {%k1} {z} ; KNL-NEXT: retq +; +; SKX-LABEL: test14: +; SKX: ## BB#0: +; SKX-NEXT: vpsubd %zmm1, %zmm0, %zmm1 +; SKX-NEXT: vpcmpgtd %zmm0, %zmm1, %k0 +; SKX-NEXT: knotw %k0, %k0 +; SKX-NEXT: knotw %k0, %k1 +; SKX-NEXT: vmovdqa32 %zmm1, %zmm0 {%k1} {z} +; SKX-NEXT: retq %sub_r = sub <16 x i32> %a, %b %cmp.i2.i = icmp sgt <16 x i32> %sub_r, %a %sext.i3.i = sext <16 x i1> %cmp.i2.i to <16 x i32> @@ -229,6 +310,15 @@ define <8 x i64> @test15(<8 x i64>%a, <8 x i64>%b) { ; KNL-NEXT: knotw %k0, %k1 ; KNL-NEXT: vmovdqa64 %zmm1, %zmm0 {%k1} {z} ; KNL-NEXT: retq +; +; SKX-LABEL: test15: +; SKX: ## BB#0: +; SKX-NEXT: vpsubq %zmm1, %zmm0, %zmm1 +; SKX-NEXT: vpcmpgtq %zmm0, %zmm1, %k0 +; SKX-NEXT: knotb %k0, %k0 +; SKX-NEXT: knotw %k0, %k1 +; SKX-NEXT: vmovdqa64 %zmm1, %zmm0 {%k1} {z} +; SKX-NEXT: retq %sub_r = sub <8 x i64> %a, %b %cmp.i2.i = icmp sgt <8 x i64> %sub_r, %a %sext.i3.i = sext <8 x i1> %cmp.i2.i to <8 x i64> @@ -244,6 +334,13 @@ define <16 x i32> @test16(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) nounwind ; KNL-NEXT: vmovdqa32 %zmm2, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test16: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpled %zmm0, %zmm1, %k1 +; SKX-NEXT: vmovdqa32 %zmm2, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = icmp sge <16 x i32> %x, %y %max = select <16 x i1> %mask, <16 x i32> %x1, <16 x i32> %y ret <16 x i32> %max @@ -256,6 +353,13 @@ define <16 x i32> @test17(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nou ; KNL-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test17: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpgtd (%rdi), %zmm0, %k1 +; SKX-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 %mask = icmp sgt <16 x i32> %x, %y %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 @@ -269,6 +373,13 @@ define <16 x i32> @test18(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nou ; KNL-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test18: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpled (%rdi), %zmm0, %k1 +; SKX-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 %mask = icmp sle <16 x i32> %x, %y %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 @@ -282,6 +393,13 @@ define <16 x i32> @test19(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nou ; KNL-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test19: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpleud (%rdi), %zmm0, %k1 +; SKX-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 %mask = icmp ule <16 x i32> %x, %y %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 @@ -296,6 +414,14 @@ define <16 x i32> @test20(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i3 ; KNL-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test20: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqd %zmm1, %zmm0, %k1 +; SKX-NEXT: vpcmpeqd %zmm3, %zmm2, %k1 {%k1} +; SKX-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask1 = icmp eq <16 x i32> %x1, %y1 %mask0 = icmp eq <16 x i32> %x, %y %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer @@ -311,6 +437,14 @@ define <8 x i64> @test21(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y ; KNL-NEXT: vmovdqa64 %zmm0, %zmm2 {%k1} ; KNL-NEXT: vmovaps %zmm2, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test21: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpleq %zmm1, %zmm0, %k1 +; SKX-NEXT: vpcmpleq %zmm2, %zmm3, %k1 {%k1} +; SKX-NEXT: vmovdqa64 %zmm0, %zmm2 {%k1} +; SKX-NEXT: vmovaps %zmm2, %zmm0 +; SKX-NEXT: retq %mask1 = icmp sge <8 x i64> %x1, %y1 %mask0 = icmp sle <8 x i64> %x, %y %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer @@ -326,6 +460,14 @@ define <8 x i64> @test22(<8 x i64> %x, <8 x i64>* %y.ptr, <8 x i64> %x1, <8 x i6 ; KNL-NEXT: vmovdqa64 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test22: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpgtq %zmm2, %zmm1, %k1 +; SKX-NEXT: vpcmpgtq (%rdi), %zmm0, %k1 {%k1} +; SKX-NEXT: vmovdqa64 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask1 = icmp sgt <8 x i64> %x1, %y1 %y = load <8 x i64>, <8 x i64>* %y.ptr, align 4 %mask0 = icmp sgt <8 x i64> %x, %y @@ -342,6 +484,14 @@ define <16 x i32> @test23(<16 x i32> %x, <16 x i32>* %y.ptr, <16 x i32> %x1, <16 ; KNL-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test23: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpled %zmm1, %zmm2, %k1 +; SKX-NEXT: vpcmpleud (%rdi), %zmm0, %k1 {%k1} +; SKX-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask1 = icmp sge <16 x i32> %x1, %y1 %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 %mask0 = icmp ule <16 x i32> %x, %y @@ -357,6 +507,13 @@ define <8 x i64> @test24(<8 x i64> %x, <8 x i64> %x1, i64* %yb.ptr) nounwind { ; KNL-NEXT: vmovdqa64 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test24: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpeqq (%rdi){1to8}, %zmm0, %k1 +; SKX-NEXT: vmovdqa64 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %yb = load i64, i64* %yb.ptr, align 4 %y.0 = insertelement <8 x i64> undef, i64 %yb, i32 0 %y = shufflevector <8 x i64> %y.0, <8 x i64> undef, <8 x i32> zeroinitializer @@ -372,6 +529,13 @@ define <16 x i32> @test25(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1) nounwind ; KNL-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test25: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpled (%rdi){1to16}, %zmm0, %k1 +; SKX-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %yb = load i32, i32* %yb.ptr, align 4 %y.0 = insertelement <16 x i32> undef, i32 %yb, i32 0 %y = shufflevector <16 x i32> %y.0, <16 x i32> undef, <16 x i32> zeroinitializer @@ -388,6 +552,14 @@ define <16 x i32> @test26(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1, <16 x i32 ; KNL-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test26: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpled %zmm1, %zmm2, %k1 +; SKX-NEXT: vpcmpgtd (%rdi){1to16}, %zmm0, %k1 {%k1} +; SKX-NEXT: vmovdqa32 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask1 = icmp sge <16 x i32> %x1, %y1 %yb = load i32, i32* %yb.ptr, align 4 %y.0 = insertelement <16 x i32> undef, i32 %yb, i32 0 @@ -401,11 +573,19 @@ define <16 x i32> @test26(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1, <16 x i32 define <8 x i64> @test27(<8 x i64> %x, i64* %yb.ptr, <8 x i64> %x1, <8 x i64> %y1) nounwind { ; KNL-LABEL: test27: ; KNL: ## BB#0: -; KNL-NEXT: vpcmpleq %zmm1, %zmm2, %k1 -; KNL-NEXT: vpcmpleq (%rdi){1to8}, %zmm0, %k1 {%k1} +; KNL-NEXT: vpcmpleq %zmm1, %zmm2, %k1 +; KNL-NEXT: vpcmpleq (%rdi){1to8}, %zmm0, %k1 {%k1} ; KNL-NEXT: vmovdqa64 %zmm0, %zmm1 {%k1} ; KNL-NEXT: vmovaps %zmm1, %zmm0 ; KNL-NEXT: retq +; +; SKX-LABEL: test27: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpleq %zmm1, %zmm2, %k1 +; SKX-NEXT: vpcmpleq (%rdi){1to8}, %zmm0, %k1 {%k1} +; SKX-NEXT: vmovdqa64 %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask1 = icmp sge <8 x i64> %x1, %y1 %yb = load i64, i64* %yb.ptr, align 4 %y.0 = insertelement <8 x i64> undef, i64 %yb, i32 0 @@ -416,11 +596,23 @@ define <8 x i64> @test27(<8 x i64> %x, i64* %yb.ptr, <8 x i64> %x1, <8 x i64> %y ret <8 x i64> %max } -; KNL-LABEL: test28 -; KNL: vpcmpgtq -; KNL: vpcmpgtq -; KNL: kxnorw define <8 x i32>@test28(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1) { +; KNL-LABEL: test28: +; KNL: ## BB#0: +; KNL-NEXT: vpcmpgtq %zmm1, %zmm0, %k0 +; KNL-NEXT: vpcmpgtq %zmm3, %zmm2, %k1 +; KNL-NEXT: kxnorw %k1, %k0, %k1 +; KNL-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0 {%k1} {z} +; KNL-NEXT: vpmovqd %zmm0, %ymm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test28: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpgtq %zmm1, %zmm0, %k0 +; SKX-NEXT: vpcmpgtq %zmm3, %zmm2, %k1 +; SKX-NEXT: kxnorb %k1, %k0, %k0 +; SKX-NEXT: vpmovm2d %k0, %ymm0 +; SKX-NEXT: retq %x_gt_y = icmp sgt <8 x i64> %x, %y %x1_gt_y1 = icmp sgt <8 x i64> %x1, %y1 %res = icmp eq <8 x i1>%x_gt_y, %x1_gt_y1 @@ -428,11 +620,23 @@ define <8 x i32>@test28(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1 ret <8 x i32> %resse } -; KNL-LABEL: test29 -; KNL: vpcmpgtd -; KNL: vpcmpgtd -; KNL: kxorw define <16 x i8>@test29(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i32> %y1) { +; KNL-LABEL: test29: +; KNL: ## BB#0: +; KNL-NEXT: vpcmpgtd %zmm1, %zmm0, %k0 +; KNL-NEXT: vpcmpgtd %zmm3, %zmm2, %k1 +; KNL-NEXT: kxorw %k1, %k0, %k1 +; KNL-NEXT: vpbroadcastd {{.*}}(%rip), %zmm0 {%k1} {z} +; KNL-NEXT: vpmovdb %zmm0, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test29: +; SKX: ## BB#0: +; SKX-NEXT: vpcmpgtd %zmm1, %zmm0, %k0 +; SKX-NEXT: vpcmpgtd %zmm3, %zmm2, %k1 +; SKX-NEXT: kxorw %k1, %k0, %k0 +; SKX-NEXT: vpmovm2b %k0, %xmm0 +; SKX-NEXT: retq %x_gt_y = icmp sgt <16 x i32> %x, %y %x1_gt_y1 = icmp sgt <16 x i32> %x1, %y1 %res = icmp ne <16 x i1>%x_gt_y, %x1_gt_y1 @@ -441,9 +645,18 @@ define <16 x i8>@test29(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i32> } define <4 x double> @test30(<4 x double> %x, <4 x double> %y) nounwind { +; KNL-LABEL: test30: +; KNL: ## BB#0: +; KNL-NEXT: vcmpeqpd %ymm1, %ymm0, %ymm2 +; KNL-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0 +; KNL-NEXT: retq +; ; SKX-LABEL: test30: -; SKX: vcmpeqpd %ymm1, %ymm0, %k1 -; SKX: vmovapd %ymm0, %ymm1 {%k1} +; SKX: ## BB#0: +; SKX-NEXT: vcmpeqpd %ymm1, %ymm0, %k1 +; SKX-NEXT: vmovapd %ymm0, %ymm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %mask = fcmp oeq <4 x double> %x, %y %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %y @@ -451,9 +664,18 @@ define <4 x double> @test30(<4 x double> %x, <4 x double> %y) nounwind { } define <2 x double> @test31(<2 x double> %x, <2 x double> %x1, <2 x double>* %yp) nounwind { -; SKX-LABEL: test31: -; SKX: vcmpltpd (%rdi), %xmm0, %k1 -; SKX: vmovapd %xmm0, %xmm1 {%k1} +; KNL-LABEL: test31: +; KNL: ## BB#0: +; KNL-NEXT: vcmpltpd (%rdi), %xmm0, %xmm2 +; KNL-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test31: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltpd (%rdi), %xmm0, %k1 +; SKX-NEXT: vmovapd %xmm0, %xmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <2 x double>, <2 x double>* %yp, align 4 %mask = fcmp olt <2 x double> %x, %y @@ -462,9 +684,18 @@ define <2 x double> @test31(<2 x double> %x, <2 x double> %x1, <2 x double>* %yp } define <4 x double> @test32(<4 x double> %x, <4 x double> %x1, <4 x double>* %yp) nounwind { +; KNL-LABEL: test32: +; KNL: ## BB#0: +; KNL-NEXT: vcmpltpd (%rdi), %ymm0, %ymm2 +; KNL-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0 +; KNL-NEXT: retq +; ; SKX-LABEL: test32: -; SKX: vcmpltpd (%rdi), %ymm0, %k1 -; SKX: vmovapd %ymm0, %ymm1 {%k1} +; SKX: ## BB#0: +; SKX-NEXT: vcmpltpd (%rdi), %ymm0, %k1 +; SKX-NEXT: vmovapd %ymm0, %ymm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <4 x double>, <4 x double>* %yp, align 4 %mask = fcmp ogt <4 x double> %y, %x @@ -473,9 +704,19 @@ define <4 x double> @test32(<4 x double> %x, <4 x double> %x1, <4 x double>* %yp } define <8 x double> @test33(<8 x double> %x, <8 x double> %x1, <8 x double>* %yp) nounwind { -; SKX-LABEL: test33: -; SKX: vcmpltpd (%rdi), %zmm0, %k1 -; SKX: vmovapd %zmm0, %zmm1 {%k1} +; KNL-LABEL: test33: +; KNL: ## BB#0: +; KNL-NEXT: vcmpltpd (%rdi), %zmm0, %k1 +; KNL-NEXT: vmovapd %zmm0, %zmm1 {%k1} +; KNL-NEXT: vmovaps %zmm1, %zmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test33: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltpd (%rdi), %zmm0, %k1 +; SKX-NEXT: vmovapd %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <8 x double>, <8 x double>* %yp, align 4 %mask = fcmp olt <8 x double> %x, %y %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1 @@ -483,9 +724,18 @@ define <8 x double> @test33(<8 x double> %x, <8 x double> %x1, <8 x double>* %yp } define <4 x float> @test34(<4 x float> %x, <4 x float> %x1, <4 x float>* %yp) nounwind { -; SKX-LABEL: test34: -; SKX: vcmpltps (%rdi), %xmm0, %k1 -; SKX: vmovaps %xmm0, %xmm1 {%k1} +; KNL-LABEL: test34: +; KNL: ## BB#0: +; KNL-NEXT: vcmpltps (%rdi), %xmm0, %xmm2 +; KNL-NEXT: vblendvps %xmm2, %xmm0, %xmm1, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test34: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltps (%rdi), %xmm0, %k1 +; SKX-NEXT: vmovaps %xmm0, %xmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <4 x float>, <4 x float>* %yp, align 4 %mask = fcmp olt <4 x float> %x, %y %max = select <4 x i1> %mask, <4 x float> %x, <4 x float> %x1 @@ -493,9 +743,19 @@ define <4 x float> @test34(<4 x float> %x, <4 x float> %x1, <4 x float>* %yp) no } define <8 x float> @test35(<8 x float> %x, <8 x float> %x1, <8 x float>* %yp) nounwind { +; KNL-LABEL: test35: +; KNL: ## BB#0: +; KNL-NEXT: vmovups (%rdi), %ymm2 +; KNL-NEXT: vcmpltps %zmm2, %zmm0, %k1 +; KNL-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} +; KNL-NEXT: retq +; ; SKX-LABEL: test35: -; SKX: vcmpltps (%rdi), %ymm0, %k1 -; SKX: vmovaps %ymm0, %ymm1 {%k1} +; SKX: ## BB#0: +; SKX-NEXT: vcmpltps (%rdi), %ymm0, %k1 +; SKX-NEXT: vmovaps %ymm0, %ymm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <8 x float>, <8 x float>* %yp, align 4 %mask = fcmp ogt <8 x float> %y, %x @@ -504,9 +764,19 @@ define <8 x float> @test35(<8 x float> %x, <8 x float> %x1, <8 x float>* %yp) no } define <16 x float> @test36(<16 x float> %x, <16 x float> %x1, <16 x float>* %yp) nounwind { -; SKX-LABEL: test36: -; SKX: vcmpltps (%rdi), %zmm0, %k1 -; SKX: vmovaps %zmm0, %zmm1 {%k1} +; KNL-LABEL: test36: +; KNL: ## BB#0: +; KNL-NEXT: vcmpltps (%rdi), %zmm0, %k1 +; KNL-NEXT: vmovaps %zmm0, %zmm1 {%k1} +; KNL-NEXT: vmovaps %zmm1, %zmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test36: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltps (%rdi), %zmm0, %k1 +; SKX-NEXT: vmovaps %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %y = load <16 x float>, <16 x float>* %yp, align 4 %mask = fcmp olt <16 x float> %x, %y %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %x1 @@ -514,9 +784,19 @@ define <16 x float> @test36(<16 x float> %x, <16 x float> %x1, <16 x float>* %yp } define <8 x double> @test37(<8 x double> %x, <8 x double> %x1, double* %ptr) nounwind { -; SKX-LABEL: test37: -; SKX: vcmpltpd (%rdi){1to8}, %zmm0, %k1 -; SKX: vmovapd %zmm0, %zmm1 {%k1} +; KNL-LABEL: test37: +; KNL: ## BB#0: +; KNL-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 +; KNL-NEXT: vmovapd %zmm0, %zmm1 {%k1} +; KNL-NEXT: vmovaps %zmm1, %zmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test37: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 +; SKX-NEXT: vmovapd %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %a = load double, double* %ptr %v = insertelement <8 x double> undef, double %a, i32 0 @@ -528,28 +808,48 @@ define <8 x double> @test37(<8 x double> %x, <8 x double> %x1, double* %ptr) nou } define <4 x double> @test38(<4 x double> %x, <4 x double> %x1, double* %ptr) nounwind { -; SKX-LABEL: test38: -; SKX: vcmpltpd (%rdi){1to4}, %ymm0, %k1 -; SKX: vmovapd %ymm0, %ymm1 {%k1} +; KNL-LABEL: test38: +; KNL: ## BB#0: +; KNL-NEXT: vbroadcastsd (%rdi), %ymm2 +; KNL-NEXT: vcmpltpd %ymm2, %ymm0, %ymm2 +; KNL-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test38: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltpd (%rdi){1to4}, %ymm0, %k1 +; SKX-NEXT: vmovapd %ymm0, %ymm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %a = load double, double* %ptr %v = insertelement <4 x double> undef, double %a, i32 0 %shuffle = shufflevector <4 x double> %v, <4 x double> undef, <4 x i32> zeroinitializer - + %mask = fcmp ogt <4 x double> %shuffle, %x %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %x1 ret <4 x double> %max } define <2 x double> @test39(<2 x double> %x, <2 x double> %x1, double* %ptr) nounwind { -; SKX-LABEL: test39: -; SKX: vcmpltpd (%rdi){1to2}, %xmm0, %k1 -; SKX: vmovapd %xmm0, %xmm1 {%k1} +; KNL-LABEL: test39: +; KNL: ## BB#0: +; KNL-NEXT: vmovddup {{.*#+}} xmm2 = mem[0,0] +; KNL-NEXT: vcmpltpd %xmm2, %xmm0, %xmm2 +; KNL-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test39: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltpd (%rdi){1to2}, %xmm0, %k1 +; SKX-NEXT: vmovapd %xmm0, %xmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %a = load double, double* %ptr %v = insertelement <2 x double> undef, double %a, i32 0 %shuffle = shufflevector <2 x double> %v, <2 x double> undef, <2 x i32> - + %mask = fcmp ogt <2 x double> %shuffle, %x %max = select <2 x i1> %mask, <2 x double> %x, <2 x double> %x1 ret <2 x double> %max @@ -557,59 +857,170 @@ define <2 x double> @test39(<2 x double> %x, <2 x double> %x1, double* %ptr) nou define <16 x float> @test40(<16 x float> %x, <16 x float> %x1, float* %ptr) nounwind { -; SKX-LABEL: test40: -; SKX: vcmpltps (%rdi){1to16}, %zmm0, %k1 -; SKX: vmovaps %zmm0, %zmm1 {%k1} +; KNL-LABEL: test40: +; KNL: ## BB#0: +; KNL-NEXT: vcmpltps (%rdi){1to16}, %zmm0, %k1 +; KNL-NEXT: vmovaps %zmm0, %zmm1 {%k1} +; KNL-NEXT: vmovaps %zmm1, %zmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test40: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltps (%rdi){1to16}, %zmm0, %k1 +; SKX-NEXT: vmovaps %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %a = load float, float* %ptr %v = insertelement <16 x float> undef, float %a, i32 0 %shuffle = shufflevector <16 x float> %v, <16 x float> undef, <16 x i32> - + %mask = fcmp ogt <16 x float> %shuffle, %x %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %x1 ret <16 x float> %max } define <8 x float> @test41(<8 x float> %x, <8 x float> %x1, float* %ptr) nounwind { -; SKX-LABEL: test41: -; SKX: vcmpltps (%rdi){1to8}, %ymm0, %k1 -; SKX: vmovaps %ymm0, %ymm1 {%k1} +; KNL-LABEL: test41: +; KNL: ## BB#0: +; KNL-NEXT: vbroadcastss (%rdi), %ymm2 +; KNL-NEXT: vcmpltps %zmm2, %zmm0, %k1 +; KNL-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} +; KNL-NEXT: retq +; +; SKX-LABEL: test41: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltps (%rdi){1to8}, %ymm0, %k1 +; SKX-NEXT: vmovaps %ymm0, %ymm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %a = load float, float* %ptr %v = insertelement <8 x float> undef, float %a, i32 0 %shuffle = shufflevector <8 x float> %v, <8 x float> undef, <8 x i32> - + %mask = fcmp ogt <8 x float> %shuffle, %x %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %x1 ret <8 x float> %max } define <4 x float> @test42(<4 x float> %x, <4 x float> %x1, float* %ptr) nounwind { -; SKX-LABEL: test42: -; SKX: vcmpltps (%rdi){1to4}, %xmm0, %k1 -; SKX: vmovaps %xmm0, %xmm1 {%k1} - +; KNL-LABEL: test42: +; KNL: ## BB#0: +; KNL-NEXT: vbroadcastss (%rdi), %xmm2 +; KNL-NEXT: vcmpltps %xmm2, %xmm0, %xmm2 +; KNL-NEXT: vblendvps %xmm2, %xmm0, %xmm1, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test42: +; SKX: ## BB#0: +; SKX-NEXT: vcmpltps (%rdi){1to4}, %xmm0, %k1 +; SKX-NEXT: vmovaps %xmm0, %xmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq + %a = load float, float* %ptr %v = insertelement <4 x float> undef, float %a, i32 0 %shuffle = shufflevector <4 x float> %v, <4 x float> undef, <4 x i32> - + %mask = fcmp ogt <4 x float> %shuffle, %x %max = select <4 x i1> %mask, <4 x float> %x, <4 x float> %x1 ret <4 x float> %max } define <8 x double> @test43(<8 x double> %x, <8 x double> %x1, double* %ptr,<8 x i1> %mask_in) nounwind { -; SKX-LABEL: test43: -; SKX: vpmovw2m %xmm2, %k1 -; SKX: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} -; SKX: vmovapd %zmm0, %zmm1 {%k1} +; KNL-LABEL: test43: +; KNL: ## BB#0: +; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 +; KNL-NEXT: vpsllq $63, %zmm2, %zmm2 +; KNL-NEXT: vptestmq %zmm2, %zmm2, %k1 +; KNL-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} +; KNL-NEXT: vmovapd %zmm0, %zmm1 {%k1} +; KNL-NEXT: vmovaps %zmm1, %zmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test43: +; SKX: ## BB#0: +; SKX-NEXT: vpsllw $15, %xmm2, %xmm2 +; SKX-NEXT: vpmovw2m %xmm2, %k1 +; SKX-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} +; SKX-NEXT: vmovapd %zmm0, %zmm1 {%k1} +; SKX-NEXT: vmovaps %zmm1, %zmm0 +; SKX-NEXT: retq %a = load double, double* %ptr %v = insertelement <8 x double> undef, double %a, i32 0 %shuffle = shufflevector <8 x double> %v, <8 x double> undef, <8 x i32> zeroinitializer - + %mask_cmp = fcmp ogt <8 x double> %shuffle, %x %mask = and <8 x i1> %mask_cmp, %mask_in %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1 ret <8 x double> %max } + +define <4 x i32> @test44(<4 x i16> %x, <4 x i16> %y) #0 { +; KNL-LABEL: test44: +; KNL: ## BB#0: +; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; KNL-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; KNL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; KNL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test44: +; SKX: ## BB#0: +; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; SKX-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; SKX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; SKX-NEXT: vpcmpeqd %xmm1, %xmm0, %k0 +; SKX-NEXT: vpmovm2d %k0, %xmm0 +; SKX-NEXT: retq + %mask = icmp eq <4 x i16> %x, %y + %1 = sext <4 x i1> %mask to <4 x i32> + ret <4 x i32> %1 +} + +define <2 x i64> @test45(<2 x i16> %x, <2 x i16> %y) #0 { +; KNL-LABEL: test45: +; KNL: ## BB#0: +; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; KNL-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; KNL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; KNL-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 +; KNL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test45: +; SKX: ## BB#0: +; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; SKX-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; SKX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; SKX-NEXT: vpcmpeqq %xmm1, %xmm0, %k1 +; SKX-NEXT: vmovdqa64 {{.*}}(%rip), %xmm0 {%k1} {z} +; SKX-NEXT: retq + %mask = icmp eq <2 x i16> %x, %y + %1 = zext <2 x i1> %mask to <2 x i64> + ret <2 x i64> %1 +} + +define <2 x i64> @test46(<2 x float> %x, <2 x float> %y) #0 { +; KNL-LABEL: test46: +; KNL: ## BB#0: +; KNL-NEXT: vcmpeqps %xmm1, %xmm0, %xmm0 +; KNL-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero +; KNL-NEXT: vpsllq $32, %xmm0, %xmm0 +; KNL-NEXT: vpsrad $31, %xmm0, %xmm1 +; KNL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3] +; KNL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; KNL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; KNL-NEXT: retq +; +; SKX-LABEL: test46: +; SKX: ## BB#0: +; SKX-NEXT: vcmpeqps %xmm1, %xmm0, %k1 +; SKX-NEXT: vmovdqa64 {{.*}}(%rip), %xmm0 {%k1} {z} +; SKX-NEXT: retq + %mask = fcmp oeq <2 x float> %x, %y + %1 = zext <2 x i1> %mask to <2 x i64> + ret <2 x i64> %1 +}