[PATCH] D153544: [AMDGPU] Use V_FMA_MIX* more often

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp:5340
+  assert(MI.getOpcode() == TargetOpcode::G_FPTRUNC);
+  const auto &TLI = *MI.getMF()->getSubtarget().getTargetLowering();
+
----------------
I think CombinerHelper already has a TLI member


================
Comment at: llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp:5361
+      auto Tmp = MRI.createGenericVirtualRegister(MRI.getType(SrcReg));
+      B.buildInstr(PreferredFusedOpcode, {Tmp},
+                  {FMulMI->getOperand(1).getReg(),
----------------
I think just using the type here works


================
Comment at: llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp:5365
+                  B.buildConstant(MRI.getType(SrcReg), 0).getReg(0)});
+      B.buildFPTrunc(DstReg, Tmp);
+    };
----------------
Missing an erase?


================
Comment at: llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp:15838
+  bool HasFMA =
+      TLI.isFMAFasterThanFMulAndFAdd(DAG.getMachineFunction(), VT) &&
+      (!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FMA, VT));
----------------
Swap the order of these checks


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D153544/new/

https://reviews.llvm.org/D153544