[PATCH] D97938: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y, (fma u, v, z))

[Mirko Brkusanin via Phabricator via llvm-commits]

mbrkusanin added inline comments.


================
Comment at: llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp:4464
+  if (FMA) {
+    MachineInstr *FMA2 = MRI.getVRegDef(FMA->getOperand(3).getReg());
+    Register X = FMA->getOperand(1).getReg();
----------------
This matches to fmul in pattern in comment above. Better to rename to FMul.


================
Comment at: llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp:4472-4475
+      Register TmpReg = MRI.createGenericVirtualRegister(DstTy);
+      Builder.buildInstr(PreferredFusedOpcode, {TmpReg}, {U, V, Z});
+      Builder.buildInstr(PreferredFusedOpcode, {MI.getOperand(0).getReg()},
+                         {X, Y, TmpReg});
----------------
Use B not Builder.

You can just pass DstTy for dst instead of making TmpReg and get it below with .getReg(0). Or at lease rename it to FMA or InnerFMA.



CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D97938/new/

https://reviews.llvm.org/D97938