summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/Target/AMDGPU/SIFoldOperands.cpp')
-rw-r--r--llvm/lib/Target/AMDGPU/SIFoldOperands.cpp15
1 files changed, 12 insertions, 3 deletions
diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
index 9361b25db14..4a82d3a5879 100644
--- a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -762,14 +762,23 @@ static bool tryFoldInst(const SIInstrInfo *TII,
Opc == AMDGPU::V_CNDMASK_B64_PSEUDO) {
const MachineOperand *Src0 = TII->getNamedOperand(*MI, AMDGPU::OpName::src0);
const MachineOperand *Src1 = TII->getNamedOperand(*MI, AMDGPU::OpName::src1);
- if (Src1->isIdenticalTo(*Src0)) {
+ int Src1ModIdx = AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src1_modifiers);
+ int Src0ModIdx = AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src0_modifiers);
+ if (Src1->isIdenticalTo(*Src0) &&
+ (Src1ModIdx == -1 || !MI->getOperand(Src1ModIdx).getImm()) &&
+ (Src0ModIdx == -1 || !MI->getOperand(Src0ModIdx).getImm())) {
LLVM_DEBUG(dbgs() << "Folded " << *MI << " into ");
+ auto &NewDesc =
+ TII->get(Src0->isReg() ? (unsigned)AMDGPU::COPY : getMovOpc(false));
int Src2Idx = AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src2);
if (Src2Idx != -1)
MI->RemoveOperand(Src2Idx);
MI->RemoveOperand(AMDGPU::getNamedOperandIdx(Opc, AMDGPU::OpName::src1));
- mutateCopyOp(*MI, TII->get(Src0->isReg() ? (unsigned)AMDGPU::COPY
- : getMovOpc(false)));
+ if (Src1ModIdx != -1)
+ MI->RemoveOperand(Src1ModIdx);
+ if (Src0ModIdx != -1)
+ MI->RemoveOperand(Src0ModIdx);
+ mutateCopyOp(*MI, NewDesc);
LLVM_DEBUG(dbgs() << *MI << '\n');
return true;
}
OpenPOWER on IntegriCloud