diff options
| -rw-r--r-- | llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp | 12 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.cpp | 5 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.h | 3 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 2 | ||||
| -rw-r--r-- | llvm/test/MC/X86/vinsertps_decode.s | 11 |
5 files changed, 26 insertions, 7 deletions
diff --git a/llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp b/llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp index 70c7127..9f8bc57 100644 --- a/llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp +++ b/llvm/lib/Target/X86/MCTargetDesc/X86InstComments.cpp @@ -1122,15 +1122,21 @@ bool llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS, case X86::VINSERTPSrri: case X86::VINSERTPSZrri: Src2Name = getRegName(MI->getOperand(2).getReg()); - [[fallthrough]]; + DestName = getRegName(MI->getOperand(0).getReg()); + Src1Name = getRegName(MI->getOperand(1).getReg()); + if (MI->getOperand(NumOperands - 1).isImm()) + DecodeINSERTPSMask(MI->getOperand(NumOperands - 1).getImm(), ShuffleMask, + /*SrcIsMem=*/false); + break; + case X86::INSERTPSrmi: case X86::VINSERTPSrmi: case X86::VINSERTPSZrmi: DestName = getRegName(MI->getOperand(0).getReg()); Src1Name = getRegName(MI->getOperand(1).getReg()); if (MI->getOperand(NumOperands - 1).isImm()) - DecodeINSERTPSMask(MI->getOperand(NumOperands - 1).getImm(), - ShuffleMask); + DecodeINSERTPSMask(MI->getOperand(NumOperands - 1).getImm(), ShuffleMask, + /*SrcIsMem=*/true); break; case X86::MOVLHPSrr: diff --git a/llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.cpp b/llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.cpp index 82f4460..933fd16 100644 --- a/llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.cpp +++ b/llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.cpp @@ -23,7 +23,8 @@ namespace llvm { -void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask) { +void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask, + bool SrcIsMem) { // Defaults the copying the dest value. ShuffleMask.push_back(0); ShuffleMask.push_back(1); @@ -33,7 +34,7 @@ void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask) { // Decode the immediate. unsigned ZMask = Imm & 15; unsigned CountD = (Imm >> 4) & 3; - unsigned CountS = (Imm >> 6) & 3; + unsigned CountS = SrcIsMem ? 0 : (Imm >> 6) & 3; // CountS selects which input element to use. unsigned InVal = 4 + CountS; diff --git a/llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.h b/llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.h index 4ef9959..b58e3a7 100644 --- a/llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.h +++ b/llvm/lib/Target/X86/MCTargetDesc/X86ShuffleDecode.h @@ -28,7 +28,8 @@ template <typename T> class SmallVectorImpl; enum { SM_SentinelUndef = -1, SM_SentinelZero = -2 }; /// Decode a 128-bit INSERTPS instruction as a v4f32 shuffle mask. -void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask); +void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask, + bool SrcIsMem); // Insert the bottom Len elements from a second source into a vector starting at // element Idx. diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 2b497f3..934654a 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -5372,7 +5372,7 @@ static bool getTargetShuffleMask(SDValue N, bool AllowSentinelZero, assert(N.getOperand(0).getValueType() == VT && "Unexpected value type"); assert(N.getOperand(1).getValueType() == VT && "Unexpected value type"); ImmN = N.getConstantOperandVal(N.getNumOperands() - 1); - DecodeINSERTPSMask(ImmN, Mask); + DecodeINSERTPSMask(ImmN, Mask, /*SrcIsMem=*/false); IsUnary = IsFakeUnary = N.getOperand(0) == N.getOperand(1); break; case X86ISD::EXTRQI: diff --git a/llvm/test/MC/X86/vinsertps_decode.s b/llvm/test/MC/X86/vinsertps_decode.s new file mode 100644 index 0000000..b200fb1 --- /dev/null +++ b/llvm/test/MC/X86/vinsertps_decode.s @@ -0,0 +1,11 @@ +# RUN: llvm-mc -triple x86_64-unknown-unknown %s | FileCheck %s + +.intel_syntax + +# CHECK: insertps $176, (%rax), %xmm2 # xmm2 = xmm2[0,1,2],mem[0] +# CHECK: vinsertps $176, (%rax), %xmm2, %xmm2 # xmm2 = xmm2[0,1,2],mem[0] +# CHECK: vinsertps $176, (%rax), %xmm29, %xmm0 # xmm0 = xmm29[0,1,2],mem[0] + +insertps xmm2, dword ptr [rax], 0x0B0 +vinsertps xmm2,xmm2,dword ptr [rax],0x0B0 +vinsertps xmm0,xmm29,dword ptr [rax],0x0B0 |