Commen=
t # 2
on bug 10263=
3
from Samuel Pitoiset
Thanks for reporting this.
I'm able to reproduce the hang with the following environment:
OpenGL renderer string: AMD Radeon (TM) RX 480 Graphics (POLARIS10 / DRM 3.=
19.0
/ 4.13.0-rc5-ARCH+, LLVM 6.0.0)
OpenGL core profile version string: 4.5 (Core Profile) Mesa 17.3.0-devel
(git-8e8c7c6703)
No VM faults are reported, but it appears that one shader is stucked because
some uniform variables are undefined and it loops indefinitely.
When replaying the trace, I get a bunch of errors like:
Mesa: User error: GL_INVALID_OPERATION in glUniform(location=3D18)
4867496: message: major api error 3: GL_INVALID_OPERATION in
glUniform(location=3D18)
4867496 @0 glUniform1ivARB(location =3D 18, count =3D 1, value =3D &=
;0)
4867496: warning: glGetError(glUniform1ivARB) =3D GL_INVALID_OPERATION
Basically, the app is trying to update a uniform variable with an invalid
location.
After looking at the trace, the app assumes that uniform locations are fixe=
d,
which is wrong (except if they are explicitly set inside the shader). Inste=
ad,
new locations are generated after each successful re-link, which is why
GetUniformLocation exists (and the app never calls it).
The GLSL compiler doesn't allocate locations for built-in uniform variables
like gl_ModelViewMatrix which explains the above errors. The attached (hack=
y)
patch fixes the hang but don't except this to be upstream, the application =
is
buggy.
Though, it would be doable to add new drirc option like
glsl_force_fixed_locations but I personally don't want to introduce new hac=
ks
in Mesa.
Closing.