picture_playback benchmarks dominate the bench timings. Is there a reason they have N=1000, or can ...
12 years, 6 months ago
(2012-06-26 13:51:44 UTC)
#1
picture_playback benchmarks dominate the bench timings.
Is there a reason they have N=1000, or can we dial them back to N=250 to get
their cost more in line with the other benchmarks? N=250 takes the iteration
time from c. 750ms to c. 150ms.
Running bench -repeat 100 -config GPU -match picture_ 4x on linux shows only
about 2% noise at N=250, which is relatively good.
On 2012/06/26 13:52:58, epoger wrote: > LGTM > > Looks like you set it to ...
12 years, 6 months ago
(2012-06-26 13:53:25 UTC)
#3
On 2012/06/26 13:52:58, epoger wrote:
> LGTM
>
> Looks like you set it to 200, not 250; either one is fine with me.
That would explain why I was seeing 5x speedup instead of 4x, wouldn't it?
On 2012/06/26 13:53:25, TomH wrote: > On 2012/06/26 13:52:58, epoger wrote: > > LGTM > ...
12 years, 6 months ago
(2012-06-26 14:24:02 UTC)
#4
On 2012/06/26 13:53:25, TomH wrote:
> On 2012/06/26 13:52:58, epoger wrote:
> > LGTM
> >
> > Looks like you set it to 200, not 250; either one is fine with me.
>
> That would explain why I was seeing 5x speedup instead of 4x, wouldn't it?
lgtm
Issue 6325063: Reduce repeat count for picture_playback benchmarks
(Closed)
Created 12 years, 6 months ago by TomH
Modified 12 years, 6 months ago
Reviewers: DerekS, epoger
Base URL: http://skia.googlecode.com/svn/trunk/
Comments: 0