aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp')
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp33
1 files changed, 23 insertions, 10 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 9bdf822..b51d664 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -12074,22 +12074,32 @@ SDValue TargetLowering::expandPartialReduceMLA(SDNode *N,
EVT::getVectorVT(*DAG.getContext(), AccVT.getVectorElementType(),
MulOpVT.getVectorElementCount());
- unsigned ExtOpcLHS = N->getOpcode() == ISD::PARTIAL_REDUCE_UMLA
- ? ISD::ZERO_EXTEND
- : ISD::SIGN_EXTEND;
- unsigned ExtOpcRHS = N->getOpcode() == ISD::PARTIAL_REDUCE_SMLA
- ? ISD::SIGN_EXTEND
- : ISD::ZERO_EXTEND;
+ unsigned ExtOpcLHS, ExtOpcRHS;
+ switch (N->getOpcode()) {
+ default:
+ llvm_unreachable("Unexpected opcode");
+ case ISD::PARTIAL_REDUCE_UMLA:
+ ExtOpcLHS = ExtOpcRHS = ISD::ZERO_EXTEND;
+ break;
+ case ISD::PARTIAL_REDUCE_SMLA:
+ ExtOpcLHS = ExtOpcRHS = ISD::SIGN_EXTEND;
+ break;
+ case ISD::PARTIAL_REDUCE_FMLA:
+ ExtOpcLHS = ExtOpcRHS = ISD::FP_EXTEND;
+ break;
+ }
if (ExtMulOpVT != MulOpVT) {
MulLHS = DAG.getNode(ExtOpcLHS, DL, ExtMulOpVT, MulLHS);
MulRHS = DAG.getNode(ExtOpcRHS, DL, ExtMulOpVT, MulRHS);
}
SDValue Input = MulLHS;
- APInt ConstantOne;
- if (!ISD::isConstantSplatVector(MulRHS.getNode(), ConstantOne) ||
- !ConstantOne.isOne())
+ if (N->getOpcode() == ISD::PARTIAL_REDUCE_FMLA) {
+ if (!llvm::isOneOrOneSplatFP(MulRHS))
+ Input = DAG.getNode(ISD::FMUL, DL, ExtMulOpVT, MulLHS, MulRHS);
+ } else if (!llvm::isOneOrOneSplat(MulRHS)) {
Input = DAG.getNode(ISD::MUL, DL, ExtMulOpVT, MulLHS, MulRHS);
+ }
unsigned Stride = AccVT.getVectorMinNumElements();
unsigned ScaleFactor = MulOpVT.getVectorMinNumElements() / Stride;
@@ -12099,10 +12109,13 @@ SDValue TargetLowering::expandPartialReduceMLA(SDNode *N,
for (unsigned I = 0; I < ScaleFactor; I++)
Subvectors.push_back(DAG.getExtractSubvector(DL, AccVT, Input, I * Stride));
+ unsigned FlatNode =
+ N->getOpcode() == ISD::PARTIAL_REDUCE_FMLA ? ISD::FADD : ISD::ADD;
+
// Flatten the subvector tree
while (Subvectors.size() > 1) {
Subvectors.push_back(
- DAG.getNode(ISD::ADD, DL, AccVT, {Subvectors[0], Subvectors[1]}));
+ DAG.getNode(FlatNode, DL, AccVT, {Subvectors[0], Subvectors[1]}));
Subvectors.pop_front();
Subvectors.pop_front();
}