[PATCH] D57059: [SLP] Initial support for the vectorization of the non-power-of-2 vectors.

Wed Dec 9 04:58:03 PST 2020

anton-afanasyev added a comment.

In D57059#2442286 <https://reviews.llvm.org/D57059#2442286>, @RKSimon wrote:

> AFAICT the only outstanding question is whether the compile time increase is acceptable?

I'd agree that geomean = 0.2% is acceptable for the change with such awesome performance impact, just noted that changed time compilation is significant in comparision with other changes. Generally it looks good to me apart from one minor unaddressed comment.

================
Comment at: llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp:3032
+                     ReuseShuffleIndicies);
+        LLVM_DEBUG(dbgs() << "SLP: Non-consecutive store.\n");
+        return;
----------------
vdmitrie wrote:
> "Non-consecutive" here is not the actual reason.
Well, "unsortable" or "unprocessable" term would be more precise. But why did we change `if (sortPtrAccesses)...` to opposite condition? This change just duplicate debug output, since we didn't differentiate it. Also I'd prefer to see the same `if-else` structure as for the load case.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57059/new/

https://reviews.llvm.org/D57059