This is a workaround for the bug filed here: http://crbug.com/110343 It only happens for Intel ...
11 years, 8 months ago
(2012-08-17 00:13:50 UTC)
#1
This is a workaround for the bug filed here: http://crbug.com/110343
It only happens for Intel GPUs on the M21 branch (currently stable and beta).
That branch does not have the shader cache or program binary changes.
Vangelis posted repro steps in comment #62.
I don't know exactly why it works around the bug but it might have something to
do with a shader being destroyed on one context while a D3D pixel shader is
current on the device and happens to have the same code.
In the repro steps in comment #62, a tab is closed without being made visible.
That would make the tab's GL context current only for the purpose of destroying
resources like shaders. It wouldn't draw anything using that context so ANGLE
wouldn't change the device's current pixel shader, which is deferred until a
draw call.
I want to commit to trunk to make sure it doesn't have any downside in ...
11 years, 8 months ago
(2012-08-17 00:42:58 UTC)
#3
I want to commit to trunk to make sure it doesn't have any downside in Canary
and then merge it into the m21 branch if it's okay. m21 is stable so I have to
be careful.
I haven't seen any issues with the vertex shader.
I would like to remove it if we don't need it post M21. I need ...
11 years, 8 months ago
(2012-08-17 20:16:52 UTC)
#5
I would like to remove it if we don't need it post M21. I need to figure out why
the bug doesn't happen in M22. I tried disabling the shader cache and
get_program_binary locally and the bug didn't happen. I think I'll revert it in
the next week or two and just put it back in if the bug shows up again.
On 2012/08/17 20:16:52, apatrick1 wrote: > I would like to remove it if we don't ...
11 years, 5 months ago
(2012-11-21 22:56:02 UTC)
#6
On 2012/08/17 20:16:52, apatrick1 wrote:
> I would like to remove it if we don't need it post M21. I need to figure out
why
> the bug doesn't happen in M22. I tried disabling the shader cache and
> get_program_binary locally and the bug didn't happen. I think I'll revert it
in
> the next week or two and just put it back in if the bug shows up again.
Al: do you think we can remove this now?
It would certainly make some of the refactoring we're doing easier if we didn't
need to keep this hack around.
Issue 6460100: Reset shaders to null in makeCurrent.
(Closed)
Created 11 years, 8 months ago by apatrick1
Modified 11 years, 5 months ago
Reviewers: dgkoch
Base URL: http://angleproject.googlecode.com/svn/trunk/
Comments: 0