Summary: | Optimize vclip for NEON in VectorMath | ||||||
---|---|---|---|---|---|---|---|
Product: | WebKit | Reporter: | Gabor Rapcsanyi <rgabor> | ||||
Component: | Web Audio | Assignee: | Nobody <webkit-unassigned> | ||||
Status: | RESOLVED FIXED | ||||||
Severity: | Normal | CC: | crogers, eric.carlson, feature-media-reviews, kdj.tikka, s.choi, webkit.review.bot | ||||
Priority: | P2 | ||||||
Version: | 528+ (Nightly build) | ||||||
Hardware: | Unspecified | ||||||
OS: | Unspecified | ||||||
Attachments: |
|
Description
Gabor Rapcsanyi
2012-10-30 02:00:00 PDT
Created attachment 171404 [details]
proposed patch
I have made independent tests on this function.
The vector size was 1000 with float numbers and I ran it 100 000 times.
The average results:
REF elapsed time is 1219 ms
NEON elapsed time is 427 ms
Comment on attachment 171404 [details] proposed patch Clearing flags on attachment: 171404 Committed r132890: <http://trac.webkit.org/changeset/132890> All reviewed patches have been landed. Closing bug. Hi all, Here on top of this patch i would like to submit more optimized ARM-NEON solution. Please let me know, is it OK to submit here ? or should i create new bug. Regards, kdj (In reply to comment #4) > Hi all, > > Here on top of this patch i would like to submit more optimized ARM-NEON solution. > > Please let me know, is it OK to submit here ? > or should i create new bug. > > Regards, > kdj Hello! That's good but this patch landed so you should make another bug for yours. Please CC me as well :) |