Program/transformfeedback restore order in virtual context switch.

Program/transformfeedback restore order in virtual context switch.

The original order is wrong. We have to pause active transformfeedback
from previous context, restore the program for this context, and then
restore the active transformfeedback (possibly resume it) for this
context.

Also, add transformfeedback validation for UseProgram.

Also, clean up some outdated bindings (auto-generated by running
ui/gl/generate_bindings.py).

With this CL, a few random crashes on Linux in WebGL2 conformance tests
are gone.

BUG=429053
TEST=gpu_unittests,webgl2_conformance
[email protected]
CQ_INCLUDE_TRYBOTS=tryserver.chromium.linux:linux_optional_gpu_tests_rel;tryserver.chromium.mac:mac_optional_gpu_tests_rel;tryserver.chromium.win:win_optional_gpu_tests_rel

Committed: https://crrev.com/744d40e0fc0133373c70406068ddb3bb5e3e462c
Cr-Commit-Position: refs/heads/master@{#392712}

[Zhenyao Mo, 2016-05-10 17:41:18]

Description was changed from

==========
Program/transformfeedback restore order in virtual context switch.

The original order is wrong. We have to pause active transformfeedback
from previous context, restore the program for this context, and then
restore the active transformfeedback (possibly resume it) for this
context.

Also, add transformfeedback validation for UseProgram.

Also, clean up some outdated bindings (auto-generated by running
ui/gl/generate_bindings.py).

With this CL, a few random crashes on Linux in WebGL2 conformance tests
are gone.

BUG=429053
TEST=gpu_unittests,webgl2_conformance
[email protected]
==========

to

==========
Program/transformfeedback restore order in virtual context switch.

The original order is wrong. We have to pause active transformfeedback
from previous context, restore the program for this context, and then
restore the active transformfeedback (possibly resume it) for this
context.

Also, add transformfeedback validation for UseProgram.

Also, clean up some outdated bindings (auto-generated by running
ui/gl/generate_bindings.py).

With this CL, a few random crashes on Linux in WebGL2 conformance tests
are gone.

BUG=429053
TEST=gpu_unittests,webgl2_conformance
[email protected]
CQ_INCLUDE_TRYBOTS=tryserver.chromium.linux:linux_optional_gpu_tests_rel;tryserver.chromium.mac:mac_optional_gpu_tests_rel;tryserver.chromium.win:win_optional_gpu_tests_rel
==========

[Zhenyao Mo, 2016-05-10 17:45:21]

The CQ bit was checked by [email protected] to run a CQ dry run

[Zhenyao Mo, 2016-05-10 17:45:49]

piman: PTAL
kbr: FYI (feel free to review)

https://codereview.chromium.org/1958253005/diff/20001/ui/gl/gl_enums_implemen...
File ui/gl/gl_enums_implementation_autogen.h (left):

https://codereview.chromium.org/1958253005/diff/20001/ui/gl/gl_enums_implemen...
ui/gl/gl_enums_implementation_autogen.h:3008: },
This is leftover from a previous patch that removed this extension.  I happen to
step on it while debugging the virtual context switch crash. It's trivia so I
didn't exclude it from this CL.

[commit-bot: I haz the power, 2016-05-10 17:46:14]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1958253005/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1958253005/20001

[commit-bot: I haz the power, 2016-05-10 19:02:02]

The CQ bit was unchecked by [email protected]

[commit-bot: I haz the power, 2016-05-10 19:02:02]

Dry run: This issue passed the CQ dry run.

[piman, 2016-05-10 19:47:45]

lgtm

[Zhenyao Mo, 2016-05-10 20:51:08]

The CQ bit was checked by [email protected]

[commit-bot: I haz the power, 2016-05-10 20:52:06]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1958253005/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1958253005/20001

[commit-bot: I haz the power, 2016-05-10 20:56:32]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2016-05-10 20:58:53]

Description was changed from

==========
Program/transformfeedback restore order in virtual context switch.

The original order is wrong. We have to pause active transformfeedback
from previous context, restore the program for this context, and then
restore the active transformfeedback (possibly resume it) for this
context.

Also, add transformfeedback validation for UseProgram.

Also, clean up some outdated bindings (auto-generated by running
ui/gl/generate_bindings.py).

With this CL, a few random crashes on Linux in WebGL2 conformance tests
are gone.

BUG=429053
TEST=gpu_unittests,webgl2_conformance
[email protected]
CQ_INCLUDE_TRYBOTS=tryserver.chromium.linux:linux_optional_gpu_tests_rel;tryserver.chromium.mac:mac_optional_gpu_tests_rel;tryserver.chromium.win:win_optional_gpu_tests_rel
==========

to

==========
Program/transformfeedback restore order in virtual context switch.

The original order is wrong. We have to pause active transformfeedback
from previous context, restore the program for this context, and then
restore the active transformfeedback (possibly resume it) for this
context.

Also, add transformfeedback validation for UseProgram.

Also, clean up some outdated bindings (auto-generated by running
ui/gl/generate_bindings.py).

With this CL, a few random crashes on Linux in WebGL2 conformance tests
are gone.

BUG=429053
TEST=gpu_unittests,webgl2_conformance
[email protected]
CQ_INCLUDE_TRYBOTS=tryserver.chromium.linux:linux_optional_gpu_tests_rel;tryserver.chromium.mac:mac_optional_gpu_tests_rel;tryserver.chromium.win:win_optional_gpu_tests_rel

Committed: https://crrev.com/744d40e0fc0133373c70406068ddb3bb5e3e462c
Cr-Commit-Position: refs/heads/master@{#392712}
==========

[commit-bot: I haz the power, 2016-05-10 20:58:54]

Patchset 2 (id:??) landed as
https://crrev.com/744d40e0fc0133373c70406068ddb3bb5e3e462c
Cr-Commit-Position: refs/heads/master@{#392712}

[Ken Russell (switch to Gerrit), 2016-05-10 21:15:51]

[email protected] changed reviewers:
+ [email protected]

[Ken Russell (switch to Gerrit), 2016-05-10 21:15:51]

LGTM after the fact if this has been tested. One question.

https://codereview.chromium.org/1958253005/diff/20001/gpu/command_buffer/serv...
File gpu/command_buffer/service/context_state.cc (right):

https://codereview.chromium.org/1958253005/diff/20001/gpu/command_buffer/serv...
gpu/command_buffer/service/context_state.cc:300: feature_info_->IsES3Capable());
Could you explain under what conditions restore_transform_feedback_bindings
might be false? I don't understand why it would be OK to skip it in any
situation where feature_info_->IsES3Capable() returns true.

[Zhenyao Mo, 2016-05-10 21:19:19]

https://codereview.chromium.org/1958253005/diff/20001/gpu/command_buffer/serv...
File gpu/command_buffer/service/context_state.cc (right):

https://codereview.chromium.org/1958253005/diff/20001/gpu/command_buffer/serv...
gpu/command_buffer/service/context_state.cc:300: feature_info_->IsES3Capable());
On 2016/05/10 21:15:51, Ken Russell wrote:
> Could you explain under what conditions restore_transform_feedback_bindings
> might be false? I don't understand why it would be OK to skip it in any
> situation where feature_info_->IsES3Capable() returns true.

Good question.  This function is also called to simply restore the current
program. Say in a chromium extension function, we switch to a private program to
do some work, and after that, we want to restore the current program.  In that
use case, we don't really care about transformfeedbacks because we didn't touch
them at all.  So providing this option is for that use case.

[Ken Russell (switch to Gerrit), 2016-05-10 22:39:37]

https://codereview.chromium.org/1958253005/diff/20001/gpu/command_buffer/serv...
File gpu/command_buffer/service/context_state.cc (right):

https://codereview.chromium.org/1958253005/diff/20001/gpu/command_buffer/serv...
gpu/command_buffer/service/context_state.cc:300: feature_info_->IsES3Capable());
On 2016/05/10 21:19:19, Zhenyao Mo wrote:
> On 2016/05/10 21:15:51, Ken Russell wrote:
> > Could you explain under what conditions restore_transform_feedback_bindings
> > might be false? I don't understand why it would be OK to skip it in any
> > situation where feature_info_->IsES3Capable() returns true.
> 
> Good question.  This function is also called to simply restore the current
> program. Say in a chromium extension function, we switch to a private program
to
> do some work, and after that, we want to restore the current program.  In that
> use case, we don't really care about transformfeedbacks because we didn't
touch
> them at all.  So providing this option is for that use case.

OK, as you see fit. I have a hard time understanding whether skipping the
restoration would be correct in any situation. It seems to me that it would
always be required to pause / resume any transform feedback before switching
programs, even in the same context.