[PATCH] D29402: [SLP] Initial rework for min/max horizontal reduction vectorization, NFC.

Simon Pilgrim via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Sat Jun 24 09:32:41 PDT 2017


RKSimon added a comment.

What's happening with this? It seems almost ready to go.



================
Comment at: lib/Analysis/CostModel.cpp:196
+
+static bool matchPairwiseReductionAtLevel(Value *V, unsigned Level,
+                                          unsigned NumLevels) {
----------------
ABataev wrote:
> mkuper wrote:
> > Can this ever be a Value that's not an Instruction?
> > (Same question applies to getReductionOpcode())
> Of course no, only instructions can be used here. I'll convert these params to `Instruction *`.
Still not Instruction*


================
Comment at: lib/Analysis/CostModel.cpp:200
+
+static ReductionData getReductionData(Instruction *I) {
+  Value *L, *R;
----------------
Worth making this an Optional<> instead of relying on Opcode == 0?


https://reviews.llvm.org/D29402





More information about the llvm-commits mailing list