[PATCH] [SLPVectorization] Vectorize Reduction Chain feeding into a 'return' statement

[Arnold Schwaighofer]

I see this as an intermediate step that will allow us to catch certain patterns. If performance tests show it is beneficial why not take it?

Enabling horizontal reductions is a second step that might subsume this logic.


> On Nov 12, 2014, at 8:33 AM, James Molloy <james at jamesmolloy.co.uk> wrote:
> 
> Hi,
> 
> This seems to me to be a sticking plaster over a bigger problem. This will fix some reductions, but the bigger problem of loops being unrolled and then not optimised still exists. I'd much prefer to see a solution to this problem than just adding more Elastoplast-logic to SLP.
> 
> James
> On Wed, 12 Nov 2014 at 16:24 hfinkel at anl.gov <mailto:hfinkel at anl.gov> <hfinkel at anl.gov <mailto:hfinkel at anl.gov>> wrote:
> ================
> Comment at: test/Transforms/SLPVectorizer/X86/return.ll:58
> @@ +57,3 @@
> +
> +attributes #0 = { nounwind readonly "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
> +
> ----------------
> Please remove unnecessary attributes.
> 
> ================
> Comment at: test/Transforms/SLPVectorizer/X86/return.ll:62
> @@ +61,3 @@
> +
> +!0 = metadata !{metadata !"clang version 3.6.0 (221134)"}
> +!1 = metadata !{metadata !2, metadata !2, i64 0}
> ----------------
> Please remove unnecessary metadata.
> 
> http://reviews.llvm.org/D6227 <http://reviews.llvm.org/D6227>
> 
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu <mailto:llvm-commits at cs.uiuc.edu>
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits <http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141112/462e9583/attachment.html>