[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

This revision was automatically updated to reflect the committed changes. Closed by commit rL372082: [Clang] Pragma vectorize_width() implies vectorize(enable) (authored by SjoerdMeijer, committed by ). Herald added a project: LLVM. Herald added a subscriber: llvm-commits. Changed prior to commit

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

SjoerdMeijer updated this revision to Diff 220446. SjoerdMeijer added a comment. Just uploading new diff for completeness; I only had to change a test-case, and thus thought that committing this is okay. Many thanks again for reviewing and helping with the discussions! CHANGES SINCE LAST ACTIO

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

fhahn accepted this revision. fhahn added a comment. This revision is now accepted and ready to land. LGTM CHANGES SINCE LAST ACTION https://reviews.llvm.org/D66290/new/ https://reviews.llvm.org/D66290 ___ cfe-commits mailing list cfe-commits@lis

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

SjoerdMeijer updated this revision to Diff 217383. SjoerdMeijer edited the summary of this revision. SjoerdMeijer added a parent revision: D66796: [clang] Loop pragma vectorize(disable). SjoerdMeijer added a comment. Stripped out the functional change related to vectorize(disable). CHANGES SINC

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

SjoerdMeijer added a comment. > I think it would be slightly better to split off the change to disable > vectorization via llvm.loop.vectorize.enable=false instead of width=1. This is now D66796 . I will now start stripping it out from this patch. CHANGES SINC

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

SjoerdMeijer added a comment. Thanks, and sorry for the delay. Back in the office now, and I am addressing this: > I think it would be slightly better to split off the change to disable > vectorization via llvm.loop.vectorize.enable=false instead of width=1. Yep, I agree CHANGES SINCE LAST A

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

fhahn added a comment. I think it would be slightly better to split off the change to disable vectorization via `llvm.loop.vectorize.enable=false` instead of width=1. This changes the behaviour from "disable vectorization, but allow interleaving in the vectoriser" to "disable the vectoriser". I

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

SjoerdMeijer added a comment. Thanks for looking again! Good catch, feedback addressed. (forgot to add this message when I uploaded the new diff) CHANGES SINCE LAST ACTION https://reviews.llvm.org/D66290/new/ https://reviews.llvm.org/D66290 ___

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

SjoerdMeijer updated this revision to Diff 215388. SjoerdMeijer edited the summary of this revision. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D66290/new/ https://reviews.llvm.org/D66290 Files: clang/lib/CodeGen/CGLoopInfo.cpp clang/test/CodeGenCXX/pragma-loop-predicate.cpp clan

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

fhahn added a comment. > As also pointed out in the discussion on the cfe dev list, this is probably a > bit > of a silly combination: > > > vectorize(enable) vectorize_width(1) > > > but it could still mean that the vectorizer interleaves. So, with this > simplification, disabl

[PATCH] D66290: [clang] Pragma vectorize_width() implies vectorize(enable)

SjoerdMeijer created this revision. SjoerdMeijer added reviewers: Meinersbur, fhahn, hsaito, dorit. Specifying the vectorization width was supposed to implicitly enable vectorization, except that it wasn't really doing this. It was only setting the `vectorize.width` metadata, but not `vectorize.en