Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that instead of a fence to synchronize between the decoder context and GPU context. This has saved up to 0.2W in my testing (out of 5W), though the number is not entirely consistent.

Chrome will create the texture itself and share it into ANGLE (rather than vice versa) so it can choose whether or not to use a keyed mutex. This helps in particular with D3D9Ex, which is incompatible with keyed mutexes.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do anything yet.

BUG=337606
Committed: https://crrev.com/ca5b9de50edf9f8d00cbfd43d593691d259818dc
Cr-Commit-Position: refs/heads/master@{#371998}

[jbauman, 2015-12-30 02:26:54]

Description was changed from

==========
Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create
the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that
instead of a fence to synchronize between the decoder context and GPU context.
This has saved up to 0.2W in my testing (out of 5W), though the number is not
entirely consistent.

Keyed mutexes are not compatible with D3D9Ex, so in that case Chrome will create
the texture in D3D9 and share it into ANGLE.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do
anything yet.

BUG=
==========

to

==========
Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create
the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that
instead of a fence to synchronize between the decoder context and GPU context.
This has saved up to 0.2W in my testing (out of 5W), though the number is not
entirely consistent.

Keyed mutexes are not compatible with D3D9Ex, so in that case Chrome will create
the texture in D3D9 and share it into ANGLE.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do
anything yet.

BUG=337606
==========

[jbauman, 2015-12-30 02:27:10]

jbauman@chromium.org changed reviewers:
+ ananta@chromium.org

[jbauman, 2016-01-20 22:40:21]

ananta: ping

[ananta, 2016-01-22 00:40:19]

https://codereview.chromium.org/1553603002/diff/20001/content/common/gpu/medi...
File content/common/gpu/media/dxva_video_decode_accelerator_win.cc (right):

https://codereview.chromium.org/1553603002/diff/20001/content/common/gpu/medi...
content/common/gpu/media/dxva_video_decode_accelerator_win.cc:431: if
(decoder.d3d11_device_) {
Please update the patch description to reflect that with this patch we are
creating the texture and sharing it with ANGLE instead of the other way around.

It would be great if you could move the texture creation into a separate
function to make this easier to read.

[jbauman, 2016-01-22 23:56:21]

Description was changed from

==========
Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create
the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that
instead of a fence to synchronize between the decoder context and GPU context.
This has saved up to 0.2W in my testing (out of 5W), though the number is not
entirely consistent.

Keyed mutexes are not compatible with D3D9Ex, so in that case Chrome will create
the texture in D3D9 and share it into ANGLE.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do
anything yet.

BUG=337606
==========

to

==========
Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create
the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that
instead of a fence to synchronize between the decoder context and GPU context.
This has saved up to 0.2W in my testing (out of 5W), though the number is not
entirely consistent.

Chrome will create the texture itself and share it into ANGLE (rather than vice
versa) so it can choose whether or not to use a keyed mutex. This helps in
particular with D3D9Ex, which is incompatible with keyed mutexes.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do
anything yet.

BUG=337606
==========

[jbauman, 2016-01-26 00:04:06]

PTAL.

On 2016/01/22 00:40:19, ananta wrote:
>
https://codereview.chromium.org/1553603002/diff/20001/content/common/gpu/medi...
> File content/common/gpu/media/dxva_video_decode_accelerator_win.cc (right):
> 
>
https://codereview.chromium.org/1553603002/diff/20001/content/common/gpu/medi...
> content/common/gpu/media/dxva_video_decode_accelerator_win.cc:431: if
> (decoder.d3d11_device_) {
> Please update the patch description to reflect that with this patch we are
> creating the texture and sharing it with ANGLE instead of the other way
around.
> 
Added

> It would be great if you could move the texture creation into a separate
> function to make this easier to read.

Ok, added DXVAVideoDecodeAccelerator::DXVAPictureBuffer::InitializeTexture

[ananta, 2016-01-26 23:09:16]

https://codereview.chromium.org/1553603002/diff/40001/content/common/gpu/medi...
File content/common/gpu/media/dxva_video_decode_accelerator_win.cc (right):

https://codereview.chromium.org/1553603002/diff/40001/content/common/gpu/medi...
content/common/gpu/media/dxva_video_decode_accelerator_win.cc:558: return false;
Should we be logging errors here?

https://codereview.chromium.org/1553603002/diff/40001/content/common/gpu/medi...
content/common/gpu/media/dxva_video_decode_accelerator_win.cc:642: if (result !=
S_OK)
Ditto for logging errors

[jbauman, 2016-01-27 22:52:54]

PTAL. Added logging.


On 2016/01/26 23:09:16, ananta wrote:
>
https://codereview.chromium.org/1553603002/diff/40001/content/common/gpu/medi...
> File content/common/gpu/media/dxva_video_decode_accelerator_win.cc (right):
> 
>
https://codereview.chromium.org/1553603002/diff/40001/content/common/gpu/medi...
> content/common/gpu/media/dxva_video_decode_accelerator_win.cc:558: return
false;
> Should we be logging errors here?
> 
>
https://codereview.chromium.org/1553603002/diff/40001/content/common/gpu/medi...
> content/common/gpu/media/dxva_video_decode_accelerator_win.cc:642: if (result
!=
> S_OK)
> Ditto for logging errors

[ananta, 2016-01-28 02:32:56]

lgtm % nit

https://codereview.chromium.org/1553603002/diff/60001/content/common/gpu/medi...
File content/common/gpu/media/dxva_video_decode_accelerator_win.cc (right):

https://codereview.chromium.org/1553603002/diff/60001/content/common/gpu/medi...
content/common/gpu/media/dxva_video_decode_accelerator_win.cc:1304:
use_keyed_mutex_ =
Please initialize this member in the ctor.

[jbauman, 2016-01-28 03:01:04]

The CQ bit was checked by jbauman@chromium.org

[jbauman, 2016-01-28 03:01:04]

The patchset sent to the CQ was uploaded after l-g-t-m from ananta@chromium.org
Link to the patchset: https://codereview.chromium.org/1553603002/#ps80001
(title: " ")

[commit-bot: I haz the power, 2016-01-28 03:04:15]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1553603002/80001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1553603002/80001

[commit-bot: I haz the power, 2016-01-28 03:49:02]

Description was changed from

==========
Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create
the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that
instead of a fence to synchronize between the decoder context and GPU context.
This has saved up to 0.2W in my testing (out of 5W), though the number is not
entirely consistent.

Chrome will create the texture itself and share it into ANGLE (rather than vice
versa) so it can choose whether or not to use a keyed mutex. This helps in
particular with D3D9Ex, which is incompatible with keyed mutexes.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do
anything yet.

BUG=337606
==========

to

==========
Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create
the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that
instead of a fence to synchronize between the decoder context and GPU context.
This has saved up to 0.2W in my testing (out of 5W), though the number is not
entirely consistent.

Chrome will create the texture itself and share it into ANGLE (rather than vice
versa) so it can choose whether or not to use a keyed mutex. This helps in
particular with D3D9Ex, which is incompatible with keyed mutexes.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do
anything yet.

BUG=337606
==========

[commit-bot: I haz the power, 2016-01-28 03:49:03]

Committed patchset #5 (id:80001)

[commit-bot: I haz the power, 2016-01-28 03:50:51]

Description was changed from

==========
Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create
the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that
instead of a fence to synchronize between the decoder context and GPU context.
This has saved up to 0.2W in my testing (out of 5W), though the number is not
entirely consistent.

Chrome will create the texture itself and share it into ANGLE (rather than vice
versa) so it can choose whether or not to use a keyed mutex. This helps in
particular with D3D9Ex, which is incompatible with keyed mutexes.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do
anything yet.

BUG=337606
==========

to

==========
Use IDXGIKeyedMutex to synchronize between DXVA decoder and main context

When EGL_ANGLE_keyed_mutex is available, DXVAVideoDecodeAccelerator will create
the picture buffer with D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX and use that
instead of a fence to synchronize between the decoder context and GPU context.
This has saved up to 0.2W in my testing (out of 5W), though the number is not
entirely consistent.

Chrome will create the texture itself and share it into ANGLE (rather than vice
versa) so it can choose whether or not to use a keyed mutex. This helps in
particular with D3D9Ex, which is incompatible with keyed mutexes.

EGL_ANGLE_keyed_mutex is not yet enabled in ANGLE for chrome, so this won't do
anything yet.

BUG=337606

Committed: https://crrev.com/ca5b9de50edf9f8d00cbfd43d593691d259818dc
Cr-Commit-Position: refs/heads/master@{#371998}
==========

[commit-bot: I haz the power, 2016-01-28 03:50:52]

Patchset 5 (id:??) landed as
https://crrev.com/ca5b9de50edf9f8d00cbfd43d593691d259818dc
Cr-Commit-Position: refs/heads/master@{#371998}