piglit test 'spec.glsl-1_30.execution.range_analysis_fsat_of_nan' failure
Failure began with piglit=6eef9ccbcf7310ac84b3049aef5abc8bf7ffd17e:
Author: Ian Romanick <ian.d.romanick@intel.com>
Date: Mon Mar 8 13:55:35 2021 -0800
glsl-1.30: Test range analysis of saturate in the presence of NaN
This test is a bit more tricky than the `min` and `max` tests. The
biggest problem, as noted in comments in the test case, is that there
are multiple bugs present in Mesa's handling of saturate. Rhys noticed
the issue in range analysis. This is essentially the same as the issue
for `min` and `max`.
Implementing this test discovered a problem in an algebraic optimization
that strips saturate from some comparisons with zero.
Part-of: <https://gitlab.freedesktop.org/mesa/piglit/-/merge_requests/463>
Enabling this test exposed the bug per this and the previous commit. Posting bug for tracking purposes.
stdout shows wrong color being displayed (https://mesa-ci.01.org/mesa_master/builds/25332/results/4575047604):
/tmp/build_root/m64/lib/piglit/bin/shader_runner /tmp/build_root/m64/lib/piglit/tests/spec/glsl-1.30/execution/range_analysis_fsat_of_nan.shader_test -auto -fbo
Probe color at (0,0)
Expected: 0 255 0 255
Observed: 255 0 0 255
Test failure on line 71
Edited by ngcortes