From 7b49e8ac359bc35f95af548fbed4b7afd625caab Mon Sep 17 00:00:00 2001 From: Craig Topper Date: Wed, 16 Oct 2019 02:50:04 +0000 Subject: [PATCH] [LegalizeTypes] Don't call PromoteTargetBoolean from SplitVecOp_VSETCC. PromoteTargetBoolean calls getSetccResultType to get the return type. But we were passing it the setcc result type rather than the setcc input type. This causes an issue on X86 with avx512vl where the setcc result type for vXf16 vectors is vXi16 while the result type for vXi16 vectors is vXi1. There's really no guarantee that getSetccResultType is the type we need here. So now we just grab the extend type from getExtendForContent and extend to the original result VT of the node we're splitting. llvm-svn: 374970 --- .../CodeGen/SelectionDAG/LegalizeVectorTypes.cpp | 6 +- llvm/test/CodeGen/X86/avx512-vec-cmp.ll | 107 +++++++++++++++++++++ 2 files changed, 112 insertions(+), 1 deletion(-) diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp index 2407127f..2f8fa97 100644 --- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp +++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp @@ -2639,7 +2639,11 @@ SDValue DAGTypeLegalizer::SplitVecOp_VSETCC(SDNode *N) { LoRes = DAG.getNode(ISD::SETCC, DL, PartResVT, Lo0, Lo1, N->getOperand(2)); HiRes = DAG.getNode(ISD::SETCC, DL, PartResVT, Hi0, Hi1, N->getOperand(2)); SDValue Con = DAG.getNode(ISD::CONCAT_VECTORS, DL, WideResVT, LoRes, HiRes); - return PromoteTargetBoolean(Con, N->getValueType(0)); + + EVT OpVT = N->getOperand(0).getValueType(); + ISD::NodeType ExtendCode = + TargetLowering::getExtendForContent(TLI.getBooleanContents(OpVT)); + return DAG.getNode(ExtendCode, DL, N->getValueType(0), Con); } diff --git a/llvm/test/CodeGen/X86/avx512-vec-cmp.ll b/llvm/test/CodeGen/X86/avx512-vec-cmp.ll index 0b518ea..b60db5d 100644 --- a/llvm/test/CodeGen/X86/avx512-vec-cmp.ll +++ b/llvm/test/CodeGen/X86/avx512-vec-cmp.ll @@ -1424,3 +1424,110 @@ define <4 x i32> @zext_bool_logic(<4 x i64> %cond1, <4 x i64> %cond2, <4 x i32> %e = add <4 x i32> %d, %x ret <4 x i32> %e } + +; This used to crash in WidenVecRes_SETCC due to generating the wrong +; result type. +define void @half_vec_compare(<2 x half>* %x, <2 x i8>* %y) { +; KNL-LABEL: half_vec_compare: +; KNL: ## %bb.0: ## %entry +; KNL-NEXT: movzwl (%rdi), %eax ## encoding: [0x0f,0xb7,0x07] +; KNL-NEXT: movzwl 2(%rdi), %ecx ## encoding: [0x0f,0xb7,0x4f,0x02] +; KNL-NEXT: cwtl ## encoding: [0x98] +; KNL-NEXT: vmovd %eax, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc0] +; KNL-NEXT: vcvtph2ps %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x13,0xc0] +; KNL-NEXT: movswl %cx, %eax ## encoding: [0x0f,0xbf,0xc1] +; KNL-NEXT: vmovd %eax, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc8] +; KNL-NEXT: vcvtph2ps %xmm1, %xmm1 ## encoding: [0xc4,0xe2,0x79,0x13,0xc9] +; KNL-NEXT: vxorps %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe8,0x57,0xd2] +; KNL-NEXT: vucomiss %xmm2, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x2e,0xca] +; KNL-NEXT: setp %al ## encoding: [0x0f,0x9a,0xc0] +; KNL-NEXT: setne %cl ## encoding: [0x0f,0x95,0xc1] +; KNL-NEXT: orb %al, %cl ## encoding: [0x08,0xc1] +; KNL-NEXT: kmovw %ecx, %k0 ## encoding: [0xc5,0xf8,0x92,0xc1] +; KNL-NEXT: kshiftlw $1, %k0, %k0 ## encoding: [0xc4,0xe3,0xf9,0x32,0xc0,0x01] +; KNL-NEXT: vucomiss %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x2e,0xc2] +; KNL-NEXT: setp %al ## encoding: [0x0f,0x9a,0xc0] +; KNL-NEXT: setne %cl ## encoding: [0x0f,0x95,0xc1] +; KNL-NEXT: orb %al, %cl ## encoding: [0x08,0xc1] +; KNL-NEXT: andl $1, %ecx ## encoding: [0x83,0xe1,0x01] +; KNL-NEXT: kmovw %ecx, %k1 ## encoding: [0xc5,0xf8,0x92,0xc9] +; KNL-NEXT: korw %k0, %k1, %k1 ## encoding: [0xc5,0xf4,0x45,0xc8] +; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] +; KNL-NEXT: vpmovdw %zmm0, %ymm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x33,0xc0] +; KNL-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x63,0xc0] +; KNL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xdb,0x05,A,A,A,A] +; KNL-NEXT: ## fixup A - offset: 4, value: LCPI68_0-4, kind: reloc_riprel_4byte +; KNL-NEXT: vpextrw $0, %xmm0, (%rsi) ## encoding: [0xc4,0xe3,0x79,0x15,0x06,0x00] +; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] +; KNL-NEXT: retq ## encoding: [0xc3] +; +; AVX512BW-LABEL: half_vec_compare: +; AVX512BW: ## %bb.0: ## %entry +; AVX512BW-NEXT: movzwl (%rdi), %eax ## encoding: [0x0f,0xb7,0x07] +; AVX512BW-NEXT: movzwl 2(%rdi), %ecx ## encoding: [0x0f,0xb7,0x4f,0x02] +; AVX512BW-NEXT: cwtl ## encoding: [0x98] +; AVX512BW-NEXT: vmovd %eax, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc0] +; AVX512BW-NEXT: vcvtph2ps %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x13,0xc0] +; AVX512BW-NEXT: movswl %cx, %eax ## encoding: [0x0f,0xbf,0xc1] +; AVX512BW-NEXT: vmovd %eax, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc8] +; AVX512BW-NEXT: vcvtph2ps %xmm1, %xmm1 ## encoding: [0xc4,0xe2,0x79,0x13,0xc9] +; AVX512BW-NEXT: vxorps %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe8,0x57,0xd2] +; AVX512BW-NEXT: vucomiss %xmm2, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x2e,0xca] +; AVX512BW-NEXT: setp %al ## encoding: [0x0f,0x9a,0xc0] +; AVX512BW-NEXT: setne %cl ## encoding: [0x0f,0x95,0xc1] +; AVX512BW-NEXT: orb %al, %cl ## encoding: [0x08,0xc1] +; AVX512BW-NEXT: kmovd %ecx, %k0 ## encoding: [0xc5,0xfb,0x92,0xc1] +; AVX512BW-NEXT: kshiftlw $1, %k0, %k0 ## encoding: [0xc4,0xe3,0xf9,0x32,0xc0,0x01] +; AVX512BW-NEXT: vucomiss %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x2e,0xc2] +; AVX512BW-NEXT: setp %al ## encoding: [0x0f,0x9a,0xc0] +; AVX512BW-NEXT: setne %cl ## encoding: [0x0f,0x95,0xc1] +; AVX512BW-NEXT: orb %al, %cl ## encoding: [0x08,0xc1] +; AVX512BW-NEXT: andl $1, %ecx ## encoding: [0x83,0xe1,0x01] +; AVX512BW-NEXT: kmovw %ecx, %k1 ## encoding: [0xc5,0xf8,0x92,0xc9] +; AVX512BW-NEXT: korw %k0, %k1, %k0 ## encoding: [0xc5,0xf4,0x45,0xc0] +; AVX512BW-NEXT: vpmovm2w %k0, %zmm0 ## encoding: [0x62,0xf2,0xfe,0x48,0x28,0xc0] +; AVX512BW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x63,0xc0] +; AVX512BW-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xdb,0x05,A,A,A,A] +; AVX512BW-NEXT: ## fixup A - offset: 4, value: LCPI68_0-4, kind: reloc_riprel_4byte +; AVX512BW-NEXT: vpextrw $0, %xmm0, (%rsi) ## EVEX TO VEX Compression encoding: [0xc4,0xe3,0x79,0x15,0x06,0x00] +; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] +; AVX512BW-NEXT: retq ## encoding: [0xc3] +; +; SKX-LABEL: half_vec_compare: +; SKX: ## %bb.0: ## %entry +; SKX-NEXT: movzwl (%rdi), %eax ## encoding: [0x0f,0xb7,0x07] +; SKX-NEXT: movzwl 2(%rdi), %ecx ## encoding: [0x0f,0xb7,0x4f,0x02] +; SKX-NEXT: cwtl ## encoding: [0x98] +; SKX-NEXT: vmovd %eax, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc0] +; SKX-NEXT: vcvtph2ps %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x13,0xc0] +; SKX-NEXT: movswl %cx, %eax ## encoding: [0x0f,0xbf,0xc1] +; SKX-NEXT: vmovd %eax, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc8] +; SKX-NEXT: vcvtph2ps %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x13,0xc9] +; SKX-NEXT: vxorps %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe8,0x57,0xd2] +; SKX-NEXT: vucomiss %xmm2, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x2e,0xca] +; SKX-NEXT: setp %al ## encoding: [0x0f,0x9a,0xc0] +; SKX-NEXT: setne %cl ## encoding: [0x0f,0x95,0xc1] +; SKX-NEXT: orb %al, %cl ## encoding: [0x08,0xc1] +; SKX-NEXT: kmovd %ecx, %k0 ## encoding: [0xc5,0xfb,0x92,0xc1] +; SKX-NEXT: kshiftlb $1, %k0, %k0 ## encoding: [0xc4,0xe3,0x79,0x32,0xc0,0x01] +; SKX-NEXT: vucomiss %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x2e,0xc2] +; SKX-NEXT: setp %al ## encoding: [0x0f,0x9a,0xc0] +; SKX-NEXT: setne %cl ## encoding: [0x0f,0x95,0xc1] +; SKX-NEXT: orb %al, %cl ## encoding: [0x08,0xc1] +; SKX-NEXT: kmovd %ecx, %k1 ## encoding: [0xc5,0xfb,0x92,0xc9] +; SKX-NEXT: kshiftlb $7, %k1, %k1 ## encoding: [0xc4,0xe3,0x79,0x32,0xc9,0x07] +; SKX-NEXT: kshiftrb $7, %k1, %k1 ## encoding: [0xc4,0xe3,0x79,0x30,0xc9,0x07] +; SKX-NEXT: korw %k0, %k1, %k0 ## encoding: [0xc5,0xf4,0x45,0xc0] +; SKX-NEXT: vpmovm2w %k0, %xmm0 ## encoding: [0x62,0xf2,0xfe,0x08,0x28,0xc0] +; SKX-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x63,0xc0] +; SKX-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xdb,0x05,A,A,A,A] +; SKX-NEXT: ## fixup A - offset: 4, value: LCPI68_0-4, kind: reloc_riprel_4byte +; SKX-NEXT: vpextrw $0, %xmm0, (%rsi) ## EVEX TO VEX Compression encoding: [0xc4,0xe3,0x79,0x15,0x06,0x00] +; SKX-NEXT: retq ## encoding: [0xc3] +entry: + %0 = load <2 x half>, <2 x half>* %x + %1 = fcmp une <2 x half> %0, zeroinitializer + %2 = zext <2 x i1> %1 to <2 x i8> + store <2 x i8> %2, <2 x i8>* %y + ret void +} -- 2.7.4