[llvm] Generate `kmov` for masking integers (PR #120593)
Phoebe Wang via llvm-commits
llvm-commits at lists.llvm.org
Thu Dec 19 18:54:16 PST 2024
================
@@ -4975,12 +4975,70 @@ bool X86DAGToDAGISel::tryVPTESTM(SDNode *Root, SDValue Setcc,
return tryFoldBroadcast(Root, P, L, Base, Scale, Index, Disp, Segment);
};
+ auto canUseKMOV = [&]() {
+ if (Src0.getOpcode() != X86ISD::VBROADCAST)
+ return false;
+
+ if (Src1.getOpcode() != ISD::LOAD ||
+ Src1.getOperand(1).getOpcode() != X86ISD::Wrapper ||
+ Src1.getOperand(1).getOperand(0).getOpcode() != ISD::TargetConstantPool)
+ return false;
+
+ const auto *ConstPool =
+ dyn_cast<ConstantPoolSDNode>(Src1.getOperand(1).getOperand(0));
+ if (!ConstPool)
+ return false;
+
+ const auto *ConstVec = ConstPool->getConstVal();
+ const auto *ConstVecType = dyn_cast<FixedVectorType>(ConstVec->getType());
+ if (!ConstVecType)
+ return false;
+
+ for (unsigned i = 0, e = ConstVecType->getNumElements(), k = 1; i != e;
+ ++i, k *= 2) {
+ const auto *Element = ConstVec->getAggregateElement(i);
+ if (llvm::isa<llvm::UndefValue>(Element)) {
+ for (unsigned j = i + 1; j != e; ++j) {
+ if (!llvm::isa<llvm::UndefValue>(ConstVec->getAggregateElement(j)))
+ return false;
+ }
+ return i != 0;
+ }
+
+ if (Element->getUniqueInteger() != k) {
+ return false;
+ }
+ }
+
+ return true;
+ };
+
// We can only fold loads if the sources are unique.
bool CanFoldLoads = Src0 != Src1;
bool FoldedLoad = false;
SDValue Tmp0, Tmp1, Tmp2, Tmp3, Tmp4;
+ SDLoc dl(Root);
+ bool IsTestN = CC == ISD::SETEQ;
+ MachineSDNode *CNode;
+ MVT ResVT = Setcc.getSimpleValueType();
if (CanFoldLoads) {
+ if (canUseKMOV()) {
+ auto Op = Src0.getOperand(0);
+ if (Op.getSimpleValueType() == MVT::i8) {
+ Op = SDValue(CurDAG->getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Op));
+ }
----------------
phoebewang wrote:
Remove parentheses
https://github.com/llvm/llvm-project/pull/120593
More information about the llvm-commits
mailing list