================
@@ -3524,14 +3524,22 @@
tryToMatchAndCreateMulAccumulateReduction(VPReductionRecipe *Red,
};
VPValue *VecOp = Red->getVecOp();
+ VPValue *Mul = nullptr;
+ VPValue *Sub = nullptr;
VPValue *A, *B;
+ // Sub reductions could have a sub between the add reduction and vec op.
+ if (match(VecOp,
+ m_Binary<Instruction::Sub>(m_SpecificInt(0), m_VPValue(Mul))))
+ Sub = VecOp;
+ else
+ Mul = VecOp;
----------------
sdesmalen-arm wrote:
There is a lot of renaming going on, making this change a lot bigger than it
should be. If you would write:
```
VPValue *Sub = nullptr;
// Sub reductions could have a sub between the add reduction and vec op.
if (match(VecOp,
m_Binary<Instruction::Sub>(m_SpecificInt(0), m_VPValue(VecOp))))
Sub = Red->getVecOp();
```
Then you can do away with the changes that just rename `VecOp` -> `Mul` and
`Mul -> MulR`.
I mentioned this in a previous review, but this suggestion somehow got reverted
again.
https://github.com/llvm/llvm-project/pull/160154
_______________________________________________
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
