Add vp9 high bit depth option

Add vp9 high bit depth option

Allow generating highbitdepth builds from the generate_gni.sh script.

Restrict the $DISABLE_AVX and --as=yasm settings to x86 builds because that is the only place they are required.

Committed: https://crrev.com/928af1031f8eeb768f259937534d79052ea3141b
Cr-Commit-Position: refs/heads/master@{#417610}

[Johann, 2016-09-07 19:27:42]

The CQ bit was checked by johannkoenig@google.com to run a CQ dry run

[commit-bot: I haz the power, 2016-09-07 19:28:26]

Dry run: CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-09-07 19:38:17]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2016-09-07 19:38:20]

Dry run: Try jobs failed on following builders:
  chromium_presubmit on master.tryserver.chromium.linux (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.linux/builders/chromium_presub...)

[Johann, 2016-09-07 20:38:44]

johannkoenig@google.com changed reviewers:
+ jzern@google.com

[Johann, 2016-09-07 20:38:45]

FYI - HBD seems OK from this aspect. The failure in chromium_presubmit is due to
the DO NOT SUBMIT in the commit message.

[jzern, 2016-09-07 23:27:37]

jzern@chromium.org changed reviewers:
+ jzern@chromium.org
- jzern@google.com

[jzern, 2016-09-07 23:27:37]

https://codereview.chromium.org/2319813002/diff/1/third_party/libvpx/generate...
File third_party/libvpx/generate_gni.sh (right):

https://codereview.chromium.org/2319813002/diff/1/third_party/libvpx/generate...
third_party/libvpx/generate_gni.sh:315: all_platforms="--enable-external-build
--enable-postproc --disable-install-srcs --enable-multi-res-encoding
--enable-temporal-denoising --disable-unit-tests --disable-install-docs
--disable-examples --enable-vp9-temporal-denoising --enable-vp9-postproc
--size-limit=16384x16384 $DISABLE_AVX --as=yasm $HIGHBD"
for now we only want to do this for x86 where we have comparable performance for
8-bit decodes between the two configs.

[Johann, 2016-09-08 21:46:01]

On 2016/09/07 23:27:37, jzern wrote:
>
https://codereview.chromium.org/2319813002/diff/1/third_party/libvpx/generate...
> File third_party/libvpx/generate_gni.sh (right):
> 
>
https://codereview.chromium.org/2319813002/diff/1/third_party/libvpx/generate...
> third_party/libvpx/generate_gni.sh:315: all_platforms="--enable-external-build
> --enable-postproc --disable-install-srcs --enable-multi-res-encoding
> --enable-temporal-denoising --disable-unit-tests --disable-install-docs
> --disable-examples --enable-vp9-temporal-denoising --enable-vp9-postproc
> --size-limit=16384x16384 $DISABLE_AVX --as=yasm $HIGHBD"
> for now we only want to do this for x86 where we have comparable performance
for
> 8-bit decodes between the two configs.

chrome binary for linux without hbd: 808485496
                           with hbd: 807609864
                         difference:    875632

Huh. That's super unexpected. We dropped almost 900k ...

The intermediates grew 2mb though:
14M	Release/obj/third_party/libvpx/
16M	Release-with-highBD/obj/third_party/libvpx/

objdump of with hdb does indeed show hbd functions:
00000000046248b0 l     F .text  0000000000000320              .hidden
vpx_highbd_convolve8_avg_horiz_sse2
0000000004625230 l     F .text  00000000000002c9              .hidden
vpx_highbd_convolve8_avg_sse2
0000000004624bd0 l     F .text  0000000000000383              .hidden
vpx_highbd_convolve8_avg_vert_sse2
0000000004624200 l     F .text  0000000000000320              .hidden
vpx_highbd_convolve8_horiz_sse2

Even just the object file is growing, but admittedly not very much:
3.7M	Release/obj/third_party/libvpx/libvpx.a
4.1M	Release-with-highBD/obj/third_party/libvpx/libvpx.a

but the overall binary shrinks:
-rwxr-x--- 1 johannkoenig eng 808485496 Sep  8 12:24 Release/chrome
-rwxr-x--- 1 johannkoenig eng 807609864 Sep  8 12:05 Release-with-highBD/chrome

[Johann, 2016-09-08 21:48:37]

Description was changed from

==========
DO NOT SUBMIT - vp9 high bit depth experiment

See if there are any obvious issues with the dry run.
==========

to

==========
Add vp9 high bit depth option

Allow generating highbitdepth builds from the generate_gni.sh script.

Restrict the $DISABLE_AVX and --as=yasm settings to x86 builds because that is
the only place they are required.
==========

[jzern, 2016-09-09 02:32:16]

On 2016/09/08 21:46:01, Johann wrote:
> On 2016/09/07 23:27:37, jzern wrote:
> >
>
https://codereview.chromium.org/2319813002/diff/1/third_party/libvpx/generate...
> > File third_party/libvpx/generate_gni.sh (right):
> > 
> >
>
https://codereview.chromium.org/2319813002/diff/1/third_party/libvpx/generate...
> > third_party/libvpx/generate_gni.sh:315:
all_platforms="--enable-external-build
> > --enable-postproc --disable-install-srcs --enable-multi-res-encoding
> > --enable-temporal-denoising --disable-unit-tests --disable-install-docs
> > --disable-examples --enable-vp9-temporal-denoising --enable-vp9-postproc
> > --size-limit=16384x16384 $DISABLE_AVX --as=yasm $HIGHBD"
> > for now we only want to do this for x86 where we have comparable performance
> for
> > 8-bit decodes between the two configs.
> 
> chrome binary for linux without hbd: 808485496
>                            with hbd: 807609864
>                          difference:    875632
> 
> Huh. That's super unexpected. We dropped almost 900k ...
> 
> The intermediates grew 2mb though:
> 14M	Release/obj/third_party/libvpx/
> 16M	Release-with-highBD/obj/third_party/libvpx/
> 
> objdump of with hdb does indeed show hbd functions:
> 00000000046248b0 l     F .text  0000000000000320              .hidden
> vpx_highbd_convolve8_avg_horiz_sse2
> 0000000004625230 l     F .text  00000000000002c9              .hidden
> vpx_highbd_convolve8_avg_sse2
> 0000000004624bd0 l     F .text  0000000000000383              .hidden
> vpx_highbd_convolve8_avg_vert_sse2
> 0000000004624200 l     F .text  0000000000000320              .hidden
> vpx_highbd_convolve8_horiz_sse2
> 
> Even just the object file is growing, but admittedly not very much:
> 3.7M	Release/obj/third_party/libvpx/libvpx.a
> 4.1M	Release-with-highBD/obj/third_party/libvpx/libvpx.a
> 
> but the overall binary shrinks:
> -rwxr-x--- 1 johannkoenig eng 808485496 Sep  8 12:24 Release/chrome
> -rwxr-x--- 1 johannkoenig eng 807609864 Sep  8 12:05
Release-with-highBD/chrome

lgtm.

I'll take a look at the size too. As you said, that's unexpected and the
opposite of the behavior of vpxdec/vpxenc sizes.

[Johann, 2016-09-09 16:14:29]

The CQ bit was checked by johannkoenig@google.com

[commit-bot: I haz the power, 2016-09-09 16:14:51]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-09-09 17:06:50]

Description was changed from

==========
Add vp9 high bit depth option

Allow generating highbitdepth builds from the generate_gni.sh script.

Restrict the $DISABLE_AVX and --as=yasm settings to x86 builds because that is
the only place they are required.
==========

to

==========
Add vp9 high bit depth option

Allow generating highbitdepth builds from the generate_gni.sh script.

Restrict the $DISABLE_AVX and --as=yasm settings to x86 builds because that is
the only place they are required.
==========

[commit-bot: I haz the power, 2016-09-09 17:06:51]

Committed patchset #2 (id:10001)

[commit-bot: I haz the power, 2016-09-09 17:08:27]

Description was changed from

==========
Add vp9 high bit depth option

Allow generating highbitdepth builds from the generate_gni.sh script.

Restrict the $DISABLE_AVX and --as=yasm settings to x86 builds because that is
the only place they are required.
==========

to

==========
Add vp9 high bit depth option

Allow generating highbitdepth builds from the generate_gni.sh script.

Restrict the $DISABLE_AVX and --as=yasm settings to x86 builds because that is
the only place they are required.

Committed: https://crrev.com/928af1031f8eeb768f259937534d79052ea3141b
Cr-Commit-Position: refs/heads/master@{#417610}
==========

[commit-bot: I haz the power, 2016-09-09 17:08:28]

Patchset 2 (id:??) landed as
https://crrev.com/928af1031f8eeb768f259937534d79052ea3141b
Cr-Commit-Position: refs/heads/master@{#417610}