[Bug 105301] The big SKQP bug

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=105301

--- Comment #21 from Kenneth Graunke <kenneth at whitecape.org> ---
Created attachment 142270
  --> https://bugs.freedesktop.org/attachment.cgi?id=142270&action=edit
Patch to detect bug in mixedtextblobs test

(In reply to Dongseong Hwang from comment #20)
> Thanks, Kenneth. Let me try to fix skqp based on your explanation tmr.
> 
> beside, I printf all gl calls in skia and glReadPixels causes gpu hangs.
> Does it make sense with your theory?

No, glReadPixels would not hit this case.  GL calls batch up GPU commands in a
buffer, which isn't submitted immediately.  glReadPixels will demand the
results of the earlier rendering, so it causes the earlier commands to be
submitted, so they actually execute.  (The batch buffer may also fill up at a
random point, causing us to submit it to the GPU.)  So, it appears to be at
fault, but it's actually innocent.  If you want to do that kind of debugging,
you should set always_flush_batch=true and INTEL_DEBUG=sync to avoid batching
and wait for rendering to finish (although this can hide many bugs).

I have attached a Mesa patch you can apply to make it assert fail in the
undefined case.  Then, the backtrace should point to the actual draw that's
going wrong.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20181030/75fbb103/attachment.html>