[Webkit-unassigned] [Bug 98913] Change transformation precision from double to float
bugzilla-daemon at webkit.org
bugzilla-daemon at webkit.org
Thu Oct 18 06:37:22 PDT 2012
https://bugs.webkit.org/show_bug.cgi?id=98913
--- Comment #15 from Gabor Rapcsanyi <rgabor at webkit.org> 2012-10-18 06:38:15 PST ---
(In reply to comment #14)
> (In reply to comment #12)
> > Created an attachment (id=168725)
--> (https://bugs.webkit.org/attachment.cgi?id=168725&action=review) [details] [details]
> > transform.html
> >
> > Performance test for transformations.
>
> Profiling this shows that it spends 12% of its time in cssText(), and about 1.5% in TransformationMatrix::multiply(), on desktop at least. So multiply() is far from being the bottleneck. dtoa() takes more time.
Well I did some more investigation in this area.
I measured the speed and memory gain with this transform.html on a Pandaboard.
As I wrote above there was no big speed differences between the float and double types even on this specific test. I changed the matrix multiplication to NEON code to see the improvement but it wasn't too much.
Speed test:
Double: ~6.03 fps
Float: ~6.32 fps
NEON: ~6.6 fps
I also did some memory measurements on this test with Smaps.
libWebCore memory usage:
Float: 14 540 kB
Double:14 580 kB
Difference: 40 kB save
Heap usage:
Float: 59 352 kB
Double:59 792 kB
Difference: 440 kb save
Sum memory usage:
Float: 95 864 kB
Double:96 368 kB
Difference: 504 kb save
As we can see the memory difference is not too much.
Summarize all of these results I think changing to float the doubles is not worth. There weren't too much performance gain even with the neonized matrix multiplication on a specific test. It would only worth because of the memory savings but it's not too much either.
So I'll close this bug with wontfix if there are no objections.
--
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.
More information about the webkit-unassigned
mailing list