[PATCH] D11530: RFC: LoopEditor, a high-level loop transform toolkit

James Molloy via llvm-commits llvm-commits at lists.llvm.org
Thu Aug 13 00:21:23 PDT 2015


Hi Hal,

Sorry that I missed your initial comment.

I want as many operations as possible to be able to operate on entire loop
nests rather than leaf loops. Even vectorization, as a stretch goal (the
legalities of such a thing I leave to some other component; I'm talking
here about the pure mechanics).

A personal goal of mine is for this work to enable a simple unroll-and-jam
pass which would indeed require loop interleaving on a non-innermost loop.

If-conversion was something I hadn't thought of previously, but I think
it'd be really nice to break that away a bit from vectorization; have the
LoopVectorizer if-convert the scalar code, then widen/vectorize that
afterwards. I think that separates concerns a bit easier.

James

On Mon, 10 Aug 2015 at 04:51 hfinkel at anl.gov via llvm-commits <
llvm-commits at lists.llvm.org> wrote:

> hfinkel added a comment.
>
> I think this looks quite useful.
>
> One thing that is not obvious to me is what the LoopEditor will do for
> loops that are not inner loops. Are any of these operations inner-loop only?
>
> Also, how will if-conversion (during vectorization, etc.) work?
>
>
> Repository:
>   rL LLVM
>
> http://reviews.llvm.org/D11530
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150813/c2a0fc2b/attachment.html>


More information about the llvm-commits mailing list