[llvm] [CodeGen] [AMDGPU] Attempt DAGCombine for fmul with select to ldexp (PR #111109)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Thu Oct 31 15:01:40 PDT 2024
================
@@ -14476,6 +14477,66 @@ SDValue SITargetLowering::performFDivCombine(SDNode *N,
return SDValue();
}
+SDValue SITargetLowering::performFMulCombine(SDNode *N,
+ DAGCombinerInfo &DCI) const {
+ SelectionDAG &DAG = DCI.DAG;
+ EVT VT = N->getValueType(0);
+ EVT IntVT = VT.changeElementType(MVT::i32);
+
+ SDLoc SL(N);
+ SDValue LHS = N->getOperand(0);
+ SDValue RHS = N->getOperand(1);
+
+ SDNodeFlags Flags = N->getFlags();
+ SDNodeFlags LHSFlags = LHS->getFlags();
+
+ // It is cheaper to realize i32 inline constants as compared against
+ // as materializing f16 or f64 (or even non-inline f32) values,
+ // possible via ldexp usage, as shown below :
+ //
+ // Given : A = 2^a & B = 2^b ; where a and b are integers.
+ // fmul x, (select y, A, B) -> ldexp( x, (select i32 y, a, b) )
+ // fmul x, (select y, -A, -B) -> ldexp( (fneg x), (select i32 y, a, b) )
+ if (VT.getScalarType() == MVT::f64 || VT.getScalarType() == MVT::f32 ||
+ VT.getScalarType() == MVT::f16) {
+ if (RHS.hasOneUse() && RHS.getOpcode() == ISD::SELECT) {
+ const ConstantFPSDNode *TrueNode =
+ isConstOrConstSplatFP(RHS.getOperand(1));
+ const ConstantFPSDNode *FalseNode =
+ isConstOrConstSplatFP(RHS.getOperand(2));
+
+ bool AreNodesFP = TrueNode && FalseNode;
+ if (!AreNodesFP)
+ return SDValue();
----------------
arsenm wrote:
ping here
https://github.com/llvm/llvm-project/pull/111109
More information about the llvm-commits
mailing list