More FMA folding opportunities.
llvm-svn: 225380
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 64cc1f5..c68b1e3 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -6878,7 +6878,6 @@
}
} // enable-unsafe-fp-math
-
// FADD -> FMA combines:
if ((Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath) &&
TLI.isFMAFasterThanFMulAndFAdd(VT) &&
@@ -6896,6 +6895,50 @@
(N1->hasOneUse() || TLI.enableAggressiveFMAFusion(VT)))
return DAG.getNode(ISD::FMA, SDLoc(N), VT,
N1.getOperand(0), N1.getOperand(1), N0);
+
+ // Remove FP_EXTEND when there is an opportunity to combine. This is
+ // legal here since extra precision is allowed.
+
+ // fold (fadd (fpext (fmul x, y)), z) -> (fma x, y, z)
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FMUL)
+ return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N00.getOperand(0), N00.getOperand(1), N1);
+ }
+
+ // fold (fadd x, (fpext (fmul y, z)), z) -> (fma y, z, x)
+ // Note: Commutes FADD operands.
+ if (N1.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N10 = N1.getOperand(0);
+ if (N10.getOpcode() == ISD::FMUL)
+ return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N10.getOperand(0), N10.getOperand(1), N0);
+ }
+ }
+
+ // More folding opportunities when target permits.
+ if (TLI.enableAggressiveFMAFusion(VT)) {
+
+ // fold (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y (fma u, v, z))
+ if (N0.getOpcode() == ISD::FMA &&
+ N0.getOperand(2).getOpcode() == ISD::FMUL)
+ return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N0.getOperand(0), N0.getOperand(1),
+ DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N0.getOperand(2).getOperand(0),
+ N0.getOperand(2).getOperand(1),
+ N1));
+
+ // fold (fadd x, (fma y, z, (fmul u, v)) -> (fma y, z (fma u, v, x))
+ if (N1->getOpcode() == ISD::FMA &&
+ N1.getOperand(2).getOpcode() == ISD::FMUL)
+ return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N1.getOperand(0), N1.getOperand(1),
+ DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N1.getOperand(2).getOperand(0),
+ N1.getOperand(2).getOperand(1),
+ N0));
}
return SDValue();
@@ -6989,6 +7032,95 @@
DAG.getNode(ISD::FNEG, dl, VT, N00), N01,
DAG.getNode(ISD::FNEG, dl, VT, N1));
}
+
+ // Remove FP_EXTEND when there is an opportunity to combine. This is
+ // legal here since extra precision is allowed.
+
+ // fold (fsub (fpext (fmul x, y)), z) -> (fma x, y, (fneg z))
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FMUL)
+ return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N00.getOperand(0),
+ N00.getOperand(1),
+ DAG.getNode(ISD::FNEG, SDLoc(N), VT, N1));
+ }
+
+ // fold (fsub x, (fpext (fmul y, z))) -> (fma (fneg y), z, x)
+ // Note: Commutes FSUB operands.
+ if (N1.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N10 = N1.getOperand(0);
+ if (N10.getOpcode() == ISD::FMUL)
+ return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ N10.getOperand(0)),
+ N10.getOperand(1),
+ N0);
+ }
+
+ // fold (fsub (fpext (fneg (fmul, x, y))), z)
+ // -> (fma (fneg x), y, (fneg z))
+ if (N0.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FNEG) {
+ SDValue N000 = N00.getOperand(0);
+ if (N000.getOpcode() == ISD::FMUL) {
+ return DAG.getNode(ISD::FMA, dl, VT,
+ DAG.getNode(ISD::FNEG, dl, VT,
+ N000.getOperand(0)),
+ N000.getOperand(1),
+ DAG.getNode(ISD::FNEG, dl, VT, N1));
+ }
+ }
+ }
+
+ // fold (fsub (fneg (fpext (fmul, x, y))), z)
+ // -> (fma (fneg x), y, (fneg z))
+ if (N0.getOpcode() == ISD::FNEG) {
+ SDValue N00 = N0.getOperand(0);
+ if (N00.getOpcode() == ISD::FP_EXTEND) {
+ SDValue N000 = N00.getOperand(0);
+ if (N000.getOpcode() == ISD::FMUL) {
+ return DAG.getNode(ISD::FMA, dl, VT,
+ DAG.getNode(ISD::FNEG, dl, VT,
+ N000.getOperand(0)),
+ N000.getOperand(1),
+ DAG.getNode(ISD::FNEG, dl, VT, N1));
+ }
+ }
+ }
+
+ // More folding opportunities when target permits.
+ if (TLI.enableAggressiveFMAFusion(VT)) {
+
+ // fold (fsub (fma x, y, (fmul u, v)), z)
+ // -> (fma x, y (fma u, v, (fneg z)))
+ if (N0.getOpcode() == ISD::FMA &&
+ N0.getOperand(2).getOpcode() == ISD::FMUL)
+ return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N0.getOperand(0), N0.getOperand(1),
+ DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ N0.getOperand(2).getOperand(0),
+ N0.getOperand(2).getOperand(1),
+ DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ N1)));
+
+ // fold (fsub x, (fma y, z, (fmul u, v)))
+ // -> (fma (fneg y), z, (fma (fneg u), v, x))
+ if (N1.getOpcode() == ISD::FMA &&
+ N1.getOperand(2).getOpcode() == ISD::FMUL) {
+ SDValue N20 = N1.getOperand(2).getOperand(0);
+ SDValue N21 = N1.getOperand(2).getOperand(1);
+ return DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ N1.getOperand(0)),
+ N1.getOperand(1),
+ DAG.getNode(ISD::FMA, SDLoc(N), VT,
+ DAG.getNode(ISD::FNEG, SDLoc(N), VT,
+ N20),
+ N21, N0));
+ }
+ }
}
return SDValue();