[Bug 101648] Adding a single-iteration do-while loop in a GLSL shader leads to a different image

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=101648

--- Comment #5 from Hugues Evrard <h.evrard at imperial.ac.uk> ---
Could the comparative code coverage between the different runs help to locate
the bug root cause?

The report covergage.html is available in the bug file archive, it is also
temporary directly accessible at this URL -- yet it is a 35MB web page, so it
may freeze your browser for a some seconds before being displayed:
https://www.doc.ic.ac.uk/~hevrard/hidden_glfuzz/coverage_mesa-17-1-3_18May_130_noliterals_colorgrid_modulo_inv_variant_2/index.html

This code coverage compares how many times each line of Mesa source code is
executed when rendering each shader. It highlights when line execution counters
are different (yellow), and especially when some lines are executed only during
the rendering of one of the two shaders (red). Only files with different
counters are reported, and they are ranked to display the source files where
there is the biggest difference first. The line counters indicates first the
number of hits when rendering the *original* shader, then the ones for the
*variant* shader.

We can see for instance that the top three source files where line execution
differs are:
src/intel/compiler/brw_cfg.cpp
src/intel/brw_predicated_break.cpp
src/compiler/nir_loop_analyze.c

My lack of knowledge of Mesa internals makes it harder for me to feel what's
going on, but for you Mesa developers, the red lines in these files may hint at
the bug root cause?

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20170703/ef697ee0/attachment.html>