Generalize GPU rasterization device whitelist.

Generalize GPU rasterization device whitelist.

This change is based on the empirical data in the following
spreadsheets:
1. http://go/ganesh-test-spreadsheet -- showing the results of a manual test
pass various devices we have locally)
2. http://go/ganesh-previous-whitelist -- showing the list of devices we've
successfully whitelisted by machine-model-name in this file.  (This
patch replaces that cherry-picked list, as every device in it that
survived until now has Adreno 3xx.)

Conclusions:
1. All Adreno 3xx are stable, with high confidence.
2. Nvidia, Broadcom and Hisilicon are considered stable with medium
confidence (fine everywhere we've tested so far, but number of devices
tested is limited).
3. All PowerVR and Mali-628 have really bad performance even on simple text pages.
4. Some but not all Mali-400 and Adreno 2xx have serious bugs.  There's
no systematic criteria to distinguish the good from the bad, so none of
them are whitelisted here.  Might be worth revisiting these in the
future as their market share seems relatively high (either via targeted
whitelisting, or workarounds).

NOTRY=true
BUG=424970
Committed: https://crrev.com/58782a11f98bee0dac437ae7ebc045d4c4628613
Cr-Commit-Position: refs/heads/master@{#307526}

[aelias_OOO_until_Jul13, 2014-12-06 01:01:14]

aelias@chromium.org changed reviewers:
+ zmo@chromium.org

[aelias_OOO_until_Jul13, 2014-12-06 01:04:09]

PTAL.  I still need to try installing this on a device of each category to make
sure I didn't get the strings wrong, but I think this is what we should go with
for 41.

[Zhenyao Mo, 2014-12-06 01:09:52]

LGTM

[aelias_OOO_until_Jul13, 2014-12-06 01:57:43]

I removed PowerVR 544 as I just tried a Galaxy S4 GT-I9500 and the janks on
raster are huge (can be >1 second), even on a simple text page.  Unfortunately,
the correctness workaround we did for that one isn't sufficient to ship.

[Zhenyao Mo, 2014-12-06 15:05:29]

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
File gpu/config/software_rendering_list_json.cc (right):

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
gpu/config/software_rendering_list_json.cc:1069: "gl_renderer": "NVIDIA.*"
should this be gl_vendor instead?

[vangelis, 2014-12-08 06:12:46]

Alex, do the devices you meant to blacklist not support ES 3.0 or are not
available on L? If not, they will be picked up by the later clauses in the
blacklist entry.

[vangelis, 2014-12-08 06:12:58]

vangelis@google.com changed reviewers:
+ vangelis@google.com

[vangelis, 2014-12-08 06:12:58]

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
File gpu/config/software_rendering_list_json.cc (right):

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
gpu/config/software_rendering_list_json.cc:1052: "id": 96,
I would suggest using a different blacklist id for the new entry so that we
don't get it mixed up with the older one when we look at the UMA histograms.

[aelias_OOO_until_Jul13, 2014-12-08 23:13:57]

On 2014/12/08 at 06:12:46, vangelis wrote:
> Alex, do the devices you meant to blacklist not support ES 3.0 or are not
available on L? If not, they will be picked up by the later clauses in the
blacklist entry.

Right.  For example, for Adreno 3xx, that supports ES 3.0 but is deployed on
Android versions prior to K.

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
File gpu/config/software_rendering_list_json.cc (right):

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
gpu/config/software_rendering_list_json.cc:1052: "id": 96,
On 2014/12/08 06:12:58, vangelis wrote:
> I would suggest using a different blacklist id for the new entry so that we
> don't get it mixed up with the older one when we look at the UMA histograms.

Done.

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
gpu/config/software_rendering_list_json.cc:1069: "gl_renderer": "NVIDIA.*"
On 2014/12/06 15:05:28, Zhenyao Mo wrote:
> should this be gl_vendor instead?

I tried using gl_vendor in an patch set #1 but I found that rules checking only
gl_vendor match 100% of all devices (of that vendor or not); it looks like
there's a bug with that, maybe you could look into it?

Anyway, for the purposes of this change, all gl_renderer strings I've seen for
NVIDIA start with the company name, so it should be fine.  I also think
gl_renderer-based regexps will be easier to make more specific if we need to
narrow it down in the future.

[aelias_OOO_until_Jul13, 2014-12-09 00:25:04]

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
File gpu/config/software_rendering_list_json.cc (right):

https://codereview.chromium.org/760053003/diff/20001/gpu/config/software_rend...
gpu/config/software_rendering_list_json.cc:1052: "id": 96,
On 2014/12/08 23:13:57, aelias wrote:
> On 2014/12/08 06:12:58, vangelis wrote:
> > I would suggest using a different blacklist id for the new entry so that we
> > don't get it mixed up with the older one when we look at the UMA histograms.
> 
> Done.

Sorry, I had to revert back to the previous approach.  Ganesh starts enabled
universally and then entries in this file can add to the blacklist.  It's not
possible to have individual entries that remove from the blacklist; the
"whitelist" entries need to be in a single exception block.

[vangelis, 2014-12-09 19:31:28]

lgtm

[aelias_OOO_until_Jul13, 2014-12-09 19:55:34]

The CQ bit was checked by aelias@chromium.org

[commit-bot: I haz the power, 2014-12-09 19:57:36]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/760053003/60001

[commit-bot: I haz the power, 2014-12-09 19:58:37]

Committed patchset #4 (id:60001)

[commit-bot: I haz the power, 2014-12-09 19:59:33]

Patchset 4 (id:??) landed as
https://crrev.com/58782a11f98bee0dac437ae7ebc045d4c4628613
Cr-Commit-Position: refs/heads/master@{#307526}

[Zhenyao Mo, 2015-01-29 22:27:06]

LGTM

[aelias_OOO_until_Jul13, 2015-01-29 22:40:05]

This is an old CL that I linked in https://codereview.chromium.org/877343008/,
did you mean to approve that one?