media: Increase max number of decode threads for VP9 decodes at higher resolutions.

media: Increase max number of decode threads for VP9 decodes at higher resolutions.

This improves decode performance of VP9 input encoded with tiling enabled.  Each 
tile must be at least 256 pixels wide. The number of tile columns are log2. So encoded
files can have 1, 2, 4, 8, 16 tile columns. E.g. 1920x1080 files can have up to 4
encoded tile columns. Tile rows do not matter for decode performance.

Specifically, any system with multiple cores and/or hyper threading will
benefit from this change.

On desktop 1080p decode speed is roughly 1.5-2x with this change. For example, 
on the low end of the results from the test set used to produce these numbers 
an average increase is from 82 frames/second to 163 frames/second.

This holds on Android devices as well, with the same clip on a Nexus 7 going
from 14 frames/second to 25. Chrome OS shows similar gains. This puts some 1080p 
in the playable range and gives 720p more clearance.

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=232932

[scherkus (not reviewing), 2013-10-31 20:02:03]

on what sort of computer does this provide a noticeable benefit?

[jzern, 2013-11-02 01:44:58]

On 2013/10/31 20:02:03, scherkus wrote:
> on what sort of computer does this provide a noticeable benefit?

Anything with multiple cores / ht. On desktop 1080p performance is roughly
1.5-2x, on the low end of my test set this means, e.g., 82 -> 163fps.
This holds on Android devices as well, with the same clip on a Nexus 7 going
from 14 -> 25. Chrome OS shows similar gains. This puts some 1080p in the
playable range and gives 720p more clearance.

[scherkus (not reviewing), 2013-11-04 20:30:54]

thanks for fixing up the CL description

one question

https://codereview.chromium.org/55213002/diff/70001/media/filters/vpx_video_d...
File media/filters/vpx_video_decoder.cc (right):

https://codereview.chromium.org/55213002/diff/70001/media/filters/vpx_video_d...
media/filters/vpx_video_decoder.cc:56: if (config.coded_size().width() >= 2048)
I don't know much about tiling, but do we need to take area (width * height)
into account?

To take an extreme example, what is the intended result if we're decoding a
2048x32 video?

[jzern, 2013-11-04 20:41:18]

On 2013/11/04 20:30:54, scherkus wrote:
> thanks for fixing up the CL description
> 
> one question
> 
>
https://codereview.chromium.org/55213002/diff/70001/media/filters/vpx_video_d...
> File media/filters/vpx_video_decoder.cc (right):
> 
>
https://codereview.chromium.org/55213002/diff/70001/media/filters/vpx_video_d...
> media/filters/vpx_video_decoder.cc:56: if (config.coded_size().width() >=
2048)
> I don't know much about tiling, but do we need to take area (width * height)
> into account?
> 
> To take an extreme example, what is the intended result if we're decoding a
> 2048x32 video?

We're dealing with column tiles in this case -- row tiles are not independent.
So the width is the only controlling factor, with a minimum tile width of 256.
So is the question is it slower in this case? I don't know.

[fgalligan1, 2013-11-04 20:50:57]

On 2013/11/04 20:41:18, jzern wrote:
> On 2013/11/04 20:30:54, scherkus wrote:
> > thanks for fixing up the CL description
> > 
> > one question
> > 
> >
>
https://codereview.chromium.org/55213002/diff/70001/media/filters/vpx_video_d...
> > File media/filters/vpx_video_decoder.cc (right):
> > 
> >
>
https://codereview.chromium.org/55213002/diff/70001/media/filters/vpx_video_d...
> > media/filters/vpx_video_decoder.cc:56: if (config.coded_size().width() >=
> 2048)
> > I don't know much about tiling, but do we need to take area (width * height)
> > into account?
> > 
> > To take an extreme example, what is the intended result if we're decoding a
> > 2048x32 video?
> 
> We're dealing with column tiles in this case -- row tiles are not independent.
> So the width is the only controlling factor, with a minimum tile width of 256.
> So is the question is it slower in this case? I don't know.

I updated the description with some more info on vp9 tiling.

[scherkus (not reviewing), 2013-11-04 20:53:14]

On 2013/11/04 20:50:57, fgalligan1 wrote:
> On 2013/11/04 20:41:18, jzern wrote:
> > On 2013/11/04 20:30:54, scherkus wrote:
> > > thanks for fixing up the CL description
> > > 
> > > one question
> > > 
> > >
> >
>
https://codereview.chromium.org/55213002/diff/70001/media/filters/vpx_video_d...
> > > File media/filters/vpx_video_decoder.cc (right):
> > > 
> > >
> >
>
https://codereview.chromium.org/55213002/diff/70001/media/filters/vpx_video_d...
> > > media/filters/vpx_video_decoder.cc:56: if (config.coded_size().width() >=
> > 2048)
> > > I don't know much about tiling, but do we need to take area (width *
height)
> > > into account?
> > > 
> > > To take an extreme example, what is the intended result if we're decoding
a
> > > 2048x32 video?
> > 
> > We're dealing with column tiles in this case -- row tiles are not
independent.
> > So the width is the only controlling factor, with a minimum tile width of
256.
> > So is the question is it slower in this case? I don't know.
> 
> I updated the description with some more info on vp9 tiling.

thanks frank

lgtm

[commit-bot: I haz the power, 2013-11-04 21:19:04]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/tomfinegan@chromium.org/55213002/70001

[commit-bot: I haz the power, 2013-11-05 01:48:40]

Retried try job too often on linux_rel for step(s) browser_tests
http://build.chromium.org/p/tryserver.chromium/buildstatus?builder=linux_rel&...

[commit-bot: I haz the power, 2013-11-05 02:03:17]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/tomfinegan@chromium.org/55213002/70001

[commit-bot: I haz the power, 2013-11-05 03:16:19]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/tomfinegan@chromium.org/55213002/70001

[commit-bot: I haz the power, 2013-11-05 06:48:32]

Change committed as 232932