Kernel: (drm-intel-nightly) git-ec45c7
Bug detailed description:
It’s not a regression, firstly test ogl-samples.
- build: https://github.com/g-truc/ogl-samples
Black instead of blue&texture. Please see error.png attached.
Created attachment 96098 [details]
render error png for gl-320-fbo-multisample-integer
Looking at the source code of the test I see it is missing the mig/mag filters for the texture used to paint the framebuffer. Adding these filters after creating the texture fixes the problem so this looks like a bug in the test code to me.
Created attachment 96708 [details] [review]
Patch for the test
This is a patch for the ogl-samples test adding the min/mag filters to the texture. Apply the patch to ogl-samples, then rebuild and re-run the test. Should render proper results now.
I'll try to reach the developers and let them know about this.
According to the ogl-samples developers this is a bug in Mesa:
"Sampler states don't affect rendering to a texture of blitting a texture.
The bug here is that the drivers consider the texture incomplete."
Actually the problem is not the blitting operation but the rendering to the main framebuffer with the blitting results. Setting the filters after the blit and before rendering to the framebuffer does fix the problem too so this may still be a bug in test code.
Tested today with a Haswell GPU, with 10.3.2 (system), master (git-b43bbfa), and the one used when the bug was reported (git-eaf9aff).
Then I moved to check the test suite itself, and it started to work (on all those mesa versions) after this commit:
So the bug was fixed on the test side, as suggested on comment 5.
I think that this bug can be closed.
(In reply to Alejandro Piñeiro (freenode IRC: apinheiro) from comment #6)
> Tested today with a Haswell GPU, with 10.3.2 (system), master (git-b43bbfa),
> and the one used when the bug was reported (git-eaf9aff).
Final sentence missing: "And worked on all of them."