From 8da0402210232ba439493bf197865835fbf0600e Mon Sep 17 00:00:00 2001 From: Craig Topper Date: Thu, 18 Jul 2019 06:18:06 +0000 Subject: [PATCH] [X86] Disable combineConcatVectors for vXi1 vectors. I'm not convinced the code this calls is properly vetted for vXi1 vectors. Experimental vector widening legalization testing for D55251 is now hitting an assertion failure inside EltsFromConsecutiveLoads. This is occurring from a v2i1 load having a store size different than its VT size. Hopefully this commit will keep such issues from happening. llvm-svn: 366405 --- llvm/lib/Target/X86/X86ISelLowering.cpp | 4 + .../CodeGen/X86/avx512vl-intrinsics-upgrade.ll | 228 ++++++++++----------- 2 files changed, 116 insertions(+), 116 deletions(-) diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 15d4bde..0b4bf68 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -43730,6 +43730,10 @@ static SDValue combineConcatVectors(SDNode *N, SelectionDAG &DAG, EVT SrcVT = N->getOperand(0).getValueType(); const TargetLowering &TLI = DAG.getTargetLoweringInfo(); + // Don't do anything for i1 vectors. + if (VT.getVectorElementType() == MVT::i1) + return SDValue(); + if (Subtarget.hasAVX() && TLI.isTypeLegal(VT) && TLI.isTypeLegal(SrcVT)) { SmallVector Ops(N->op_begin(), N->op_end()); if (SDValue R = combineConcatVectorOps(SDLoc(N), VT.getSimpleVT(), Ops, DAG, diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll index fa82fe6..8d0aafc 100644 --- a/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll +++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll @@ -8356,19 +8356,19 @@ define <8 x i8> @test_mask_cmp_q_256(<4 x i64> %a0, <4 x i64> %a1, i8 %mask) { ; X86-NEXT: kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c] ; X86-NEXT: kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0] ; X86-NEXT: vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0] -; X86-NEXT: vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00] +; X86-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x00] ; X86-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] -; X86-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02] +; X86-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x01] ; X86-NEXT: kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3] -; X86-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04] +; X86-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x02] ; X86-NEXT: kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4] -; X86-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x08] +; X86-NEXT: vpinsrw $4, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x04] ; X86-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] -; X86-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0a] +; X86-NEXT: vpinsrw $5, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x05] ; X86-NEXT: kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6] -; X86-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0c] +; X86-NEXT: vpinsrw $6, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x06] ; X86-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] -; X86-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0e] +; X86-NEXT: vpinsrw $7, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x07] ; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77] ; X86-NEXT: retl # encoding: [0xc3] ; @@ -8385,19 +8385,19 @@ define <8 x i8> @test_mask_cmp_q_256(<4 x i64> %a0, <4 x i64> %a1, i8 %mask) { ; X64-NEXT: kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c] ; X64-NEXT: kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0] ; X64-NEXT: vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0] -; X64-NEXT: vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00] +; X64-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x00] ; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] -; X64-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02] +; X64-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x01] ; X64-NEXT: kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3] -; X64-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04] +; X64-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x02] ; X64-NEXT: kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4] -; X64-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x08] +; X64-NEXT: vpinsrw $4, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x04] ; X64-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] -; X64-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0a] +; X64-NEXT: vpinsrw $5, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x05] ; X64-NEXT: kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6] -; X64-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0c] +; X64-NEXT: vpinsrw $6, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x06] ; X64-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] -; X64-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0e] +; X64-NEXT: vpinsrw $7, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x07] ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77] ; X64-NEXT: retq # encoding: [0xc3] %res0 = call i8 @llvm.x86.avx512.mask.cmp.q.256(<4 x i64> %a0, <4 x i64> %a1, i32 0, i8 %mask) @@ -8470,61 +8470,59 @@ define <8 x i8> @test_mask_ucmp_q_256(<4 x i64> %a0, <4 x i64> %a1, i8 %mask) { ; X86-LABEL: test_mask_ucmp_q_256: ; X86: # %bb.0: ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04] -; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8] -; X86-NEXT: vpcmpeqq %ymm1, %ymm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x29,0xc1] -; X86-NEXT: vpcmpltuq %ymm1, %ymm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xe9,0x01] -; X86-NEXT: vpcmpleuq %ymm1, %ymm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xd9,0x02] -; X86-NEXT: vpcmpneqq %ymm1, %ymm0, %k2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xd1,0x04] -; X86-NEXT: vpcmpnltuq %ymm1, %ymm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xe1,0x05] -; X86-NEXT: vpcmpnleuq %ymm1, %ymm0, %k6 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xf1,0x06] -; X86-NEXT: kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c] -; X86-NEXT: kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c] -; X86-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] -; X86-NEXT: kmovw %k0, %ecx # encoding: [0xc5,0xf8,0x93,0xc8] -; X86-NEXT: vmovd %ecx, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc1] -; X86-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02] +; X86-NEXT: kmovw %eax, %k2 # encoding: [0xc5,0xf8,0x92,0xd0] +; X86-NEXT: vpcmpeqq %ymm1, %ymm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x2a,0x29,0xc1] +; X86-NEXT: vpcmpltuq %ymm1, %ymm0, %k1 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xc9,0x01] +; X86-NEXT: vpcmpleuq %ymm1, %ymm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xd9,0x02] +; X86-NEXT: vpcmpneqq %ymm1, %ymm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xe1,0x04] +; X86-NEXT: vpcmpnltuq %ymm1, %ymm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xe9,0x05] +; X86-NEXT: vpcmpnleuq %ymm1, %ymm0, %k6 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xf1,0x06] +; X86-NEXT: kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c] +; X86-NEXT: kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c] +; X86-NEXT: kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0] +; X86-NEXT: vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0] +; X86-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x00] +; X86-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] +; X86-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x01] ; X86-NEXT: kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3] -; X86-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04] -; X86-NEXT: xorl %eax, %eax # encoding: [0x31,0xc0] -; X86-NEXT: vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06] -; X86-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] -; X86-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x08] +; X86-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x02] ; X86-NEXT: kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4] -; X86-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0a] +; X86-NEXT: vpinsrw $4, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x04] +; X86-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] +; X86-NEXT: vpinsrw $5, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x05] ; X86-NEXT: kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6] -; X86-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0c] -; X86-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] -; X86-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0e] +; X86-NEXT: vpinsrw $6, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x06] +; X86-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] +; X86-NEXT: vpinsrw $7, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x07] ; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77] ; X86-NEXT: retl # encoding: [0xc3] ; ; X64-LABEL: test_mask_ucmp_q_256: ; X64: # %bb.0: -; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf] -; X64-NEXT: vpcmpeqq %ymm1, %ymm0, %k0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x29,0xc1] -; X64-NEXT: vpcmpltuq %ymm1, %ymm0, %k5 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xe9,0x01] -; X64-NEXT: vpcmpleuq %ymm1, %ymm0, %k3 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xd9,0x02] -; X64-NEXT: vpcmpneqq %ymm1, %ymm0, %k2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1f,0xd1,0x04] -; X64-NEXT: vpcmpnltuq %ymm1, %ymm0, %k4 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xe1,0x05] -; X64-NEXT: vpcmpnleuq %ymm1, %ymm0, %k6 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x1e,0xf1,0x06] -; X64-NEXT: kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c] -; X64-NEXT: kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c] -; X64-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] -; X64-NEXT: kmovw %k0, %ecx # encoding: [0xc5,0xf8,0x93,0xc8] -; X64-NEXT: vmovd %ecx, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc1] -; X64-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02] +; X64-NEXT: kmovw %edi, %k2 # encoding: [0xc5,0xf8,0x92,0xd7] +; X64-NEXT: vpcmpeqq %ymm1, %ymm0, %k0 {%k2} # encoding: [0x62,0xf2,0xfd,0x2a,0x29,0xc1] +; X64-NEXT: vpcmpltuq %ymm1, %ymm0, %k1 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xc9,0x01] +; X64-NEXT: vpcmpleuq %ymm1, %ymm0, %k3 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xd9,0x02] +; X64-NEXT: vpcmpneqq %ymm1, %ymm0, %k4 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1f,0xe1,0x04] +; X64-NEXT: vpcmpnltuq %ymm1, %ymm0, %k5 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xe9,0x05] +; X64-NEXT: vpcmpnleuq %ymm1, %ymm0, %k6 {%k2} # encoding: [0x62,0xf3,0xfd,0x2a,0x1e,0xf1,0x06] +; X64-NEXT: kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c] +; X64-NEXT: kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c] +; X64-NEXT: kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0] +; X64-NEXT: vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0] +; X64-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x00] +; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] +; X64-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x01] ; X64-NEXT: kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3] -; X64-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04] -; X64-NEXT: xorl %eax, %eax # encoding: [0x31,0xc0] -; X64-NEXT: vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06] -; X64-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] -; X64-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x08] +; X64-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x02] ; X64-NEXT: kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4] -; X64-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0a] +; X64-NEXT: vpinsrw $4, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x04] +; X64-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] +; X64-NEXT: vpinsrw $5, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x05] ; X64-NEXT: kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6] -; X64-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0c] -; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] -; X64-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0e] +; X64-NEXT: vpinsrw $6, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x06] +; X64-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] +; X64-NEXT: vpinsrw $7, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x07] ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77] ; X64-NEXT: retq # encoding: [0xc3] %res0 = call i8 @llvm.x86.avx512.mask.ucmp.q.256(<4 x i64> %a0, <4 x i64> %a1, i32 0, i8 %mask) @@ -8607,19 +8605,19 @@ define <8 x i8> @test_mask_cmp_d_128(<4 x i32> %a0, <4 x i32> %a1, i8 %mask) { ; X86-NEXT: kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c] ; X86-NEXT: kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0] ; X86-NEXT: vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0] -; X86-NEXT: vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00] +; X86-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x00] ; X86-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] -; X86-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02] +; X86-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x01] ; X86-NEXT: kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3] -; X86-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04] +; X86-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x02] ; X86-NEXT: kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4] -; X86-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x08] +; X86-NEXT: vpinsrw $4, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x04] ; X86-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] -; X86-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0a] +; X86-NEXT: vpinsrw $5, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x05] ; X86-NEXT: kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6] -; X86-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0c] +; X86-NEXT: vpinsrw $6, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x06] ; X86-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] -; X86-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0e] +; X86-NEXT: vpinsrw $7, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x07] ; X86-NEXT: retl # encoding: [0xc3] ; ; X64-LABEL: test_mask_cmp_d_128: @@ -8635,19 +8633,19 @@ define <8 x i8> @test_mask_cmp_d_128(<4 x i32> %a0, <4 x i32> %a1, i8 %mask) { ; X64-NEXT: kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c] ; X64-NEXT: kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0] ; X64-NEXT: vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0] -; X64-NEXT: vpinsrb $0, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00] +; X64-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x00] ; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] -; X64-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02] +; X64-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x01] ; X64-NEXT: kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3] -; X64-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04] +; X64-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x02] ; X64-NEXT: kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4] -; X64-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x08] +; X64-NEXT: vpinsrw $4, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x04] ; X64-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] -; X64-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0a] +; X64-NEXT: vpinsrw $5, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x05] ; X64-NEXT: kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6] -; X64-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0c] +; X64-NEXT: vpinsrw $6, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x06] ; X64-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] -; X64-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0e] +; X64-NEXT: vpinsrw $7, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x07] ; X64-NEXT: retq # encoding: [0xc3] %res0 = call i8 @llvm.x86.avx512.mask.cmp.d.128(<4 x i32> %a0, <4 x i32> %a1, i32 0, i8 %mask) %vec0 = insertelement <8 x i8> undef, i8 %res0, i32 0 @@ -8718,60 +8716,58 @@ define <8 x i8> @test_mask_ucmp_d_128(<4 x i32> %a0, <4 x i32> %a1, i8 %mask) { ; X86-LABEL: test_mask_ucmp_d_128: ; X86: # %bb.0: ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04] -; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8] -; X86-NEXT: vpcmpeqd %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x76,0xc1] -; X86-NEXT: vpcmpltud %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xe9,0x01] -; X86-NEXT: vpcmpleud %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xd9,0x02] -; X86-NEXT: vpcmpneqd %xmm1, %xmm0, %k2 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xd1,0x04] -; X86-NEXT: vpcmpnltud %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xe1,0x05] -; X86-NEXT: vpcmpnleud %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xf1,0x06] -; X86-NEXT: kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c] -; X86-NEXT: kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c] -; X86-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] -; X86-NEXT: kmovw %k0, %ecx # encoding: [0xc5,0xf8,0x93,0xc8] -; X86-NEXT: vmovd %ecx, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc1] -; X86-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02] +; X86-NEXT: kmovw %eax, %k2 # encoding: [0xc5,0xf8,0x92,0xd0] +; X86-NEXT: vpcmpeqd %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf1,0x7d,0x0a,0x76,0xc1] +; X86-NEXT: vpcmpltud %xmm1, %xmm0, %k1 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xc9,0x01] +; X86-NEXT: vpcmpleud %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xd9,0x02] +; X86-NEXT: vpcmpneqd %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xe1,0x04] +; X86-NEXT: vpcmpnltud %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xe9,0x05] +; X86-NEXT: vpcmpnleud %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xf1,0x06] +; X86-NEXT: kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c] +; X86-NEXT: kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c] +; X86-NEXT: kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0] +; X86-NEXT: vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0] +; X86-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x00] +; X86-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] +; X86-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x01] ; X86-NEXT: kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3] -; X86-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04] -; X86-NEXT: xorl %eax, %eax # encoding: [0x31,0xc0] -; X86-NEXT: vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06] -; X86-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] -; X86-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x08] +; X86-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x02] ; X86-NEXT: kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4] -; X86-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0a] +; X86-NEXT: vpinsrw $4, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x04] +; X86-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] +; X86-NEXT: vpinsrw $5, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x05] ; X86-NEXT: kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6] -; X86-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0c] -; X86-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] -; X86-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0e] +; X86-NEXT: vpinsrw $6, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x06] +; X86-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] +; X86-NEXT: vpinsrw $7, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x07] ; X86-NEXT: retl # encoding: [0xc3] ; ; X64-LABEL: test_mask_ucmp_d_128: ; X64: # %bb.0: -; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf] -; X64-NEXT: vpcmpeqd %xmm1, %xmm0, %k0 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0x76,0xc1] -; X64-NEXT: vpcmpltud %xmm1, %xmm0, %k5 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xe9,0x01] -; X64-NEXT: vpcmpleud %xmm1, %xmm0, %k3 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xd9,0x02] -; X64-NEXT: vpcmpneqd %xmm1, %xmm0, %k2 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1f,0xd1,0x04] -; X64-NEXT: vpcmpnltud %xmm1, %xmm0, %k4 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xe1,0x05] -; X64-NEXT: vpcmpnleud %xmm1, %xmm0, %k6 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x1e,0xf1,0x06] -; X64-NEXT: kshiftlw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x32,0xc9,0x0c] -; X64-NEXT: kshiftrw $12, %k1, %k1 # encoding: [0xc4,0xe3,0xf9,0x30,0xc9,0x0c] -; X64-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] -; X64-NEXT: kmovw %k0, %ecx # encoding: [0xc5,0xf8,0x93,0xc8] -; X64-NEXT: vmovd %ecx, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6e,0xc1] -; X64-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02] +; X64-NEXT: kmovw %edi, %k2 # encoding: [0xc5,0xf8,0x92,0xd7] +; X64-NEXT: vpcmpeqd %xmm1, %xmm0, %k0 {%k2} # encoding: [0x62,0xf1,0x7d,0x0a,0x76,0xc1] +; X64-NEXT: vpcmpltud %xmm1, %xmm0, %k1 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xc9,0x01] +; X64-NEXT: vpcmpleud %xmm1, %xmm0, %k3 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xd9,0x02] +; X64-NEXT: vpcmpneqd %xmm1, %xmm0, %k4 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1f,0xe1,0x04] +; X64-NEXT: vpcmpnltud %xmm1, %xmm0, %k5 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xe9,0x05] +; X64-NEXT: vpcmpnleud %xmm1, %xmm0, %k6 {%k2} # encoding: [0x62,0xf3,0x7d,0x0a,0x1e,0xf1,0x06] +; X64-NEXT: kshiftlw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x32,0xd2,0x0c] +; X64-NEXT: kshiftrw $12, %k2, %k2 # encoding: [0xc4,0xe3,0xf9,0x30,0xd2,0x0c] +; X64-NEXT: kmovw %k0, %eax # encoding: [0xc5,0xf8,0x93,0xc0] +; X64-NEXT: vpxor %xmm0, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xef,0xc0] +; X64-NEXT: vpinsrw $0, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x00] +; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] +; X64-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x01] ; X64-NEXT: kmovw %k3, %eax # encoding: [0xc5,0xf8,0x93,0xc3] -; X64-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04] -; X64-NEXT: xorl %eax, %eax # encoding: [0x31,0xc0] -; X64-NEXT: vpinsrb $6, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06] -; X64-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] -; X64-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x08] +; X64-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x02] ; X64-NEXT: kmovw %k4, %eax # encoding: [0xc5,0xf8,0x93,0xc4] -; X64-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0a] +; X64-NEXT: vpinsrw $4, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x04] +; X64-NEXT: kmovw %k5, %eax # encoding: [0xc5,0xf8,0x93,0xc5] +; X64-NEXT: vpinsrw $5, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x05] ; X64-NEXT: kmovw %k6, %eax # encoding: [0xc5,0xf8,0x93,0xc6] -; X64-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0c] -; X64-NEXT: kmovw %k1, %eax # encoding: [0xc5,0xf8,0x93,0xc1] -; X64-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x0e] +; X64-NEXT: vpinsrw $6, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x06] +; X64-NEXT: kmovw %k2, %eax # encoding: [0xc5,0xf8,0x93,0xc2] +; X64-NEXT: vpinsrw $7, %eax, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xc4,0xc0,0x07] ; X64-NEXT: retq # encoding: [0xc3] %res0 = call i8 @llvm.x86.avx512.mask.ucmp.d.128(<4 x i32> %a0, <4 x i32> %a1, i32 0, i8 %mask) %vec0 = insertelement <8 x i8> undef, i8 %res0, i32 0 -- 2.7.4