multisample.fbo_4_samples.constancy_sample_coverage
multisample.fbo_4_samples.constancy_sample_coverage_inverted
multisample.fbo_4_samples.constancy_both
multisample.fbo_4_samples.constancy_both_inverted
  (skip multisample.fbo_8_samples)
multisample.fbo_max_samples.constancy_sample_coverage          (same as 4 samples)
multisample.fbo_max_samples.constancy_sample_coverage_inverted (^)
multisample.fbo_max_samples.constancy_both                     (^)
multisample.fbo_max_samples.constancy_both_inverted            (^)

These tests fail (slightly flakily) on Google Pixel (7.1.1). The tests fail less often when run in isolation from the rest of the tests in the file.

It seems likely that this is a driver bug that is exposed by Chrome's GPU scheduling.

It doesn't seem to be fixed by
* enabling/disabling virtualized contexts
* increasing command buffer size limits (to reduce the number of command buffer flushes) - but not sure if this had an effect
 

deqp/functional/gles3/multisample.html

multisample.fbo_4_samples.constancy_sample_coverage
multisample.fbo_4_samples.constancy_sample_coverage_inverted
multisample.fbo_4_samples.constancy_both
multisample.fbo_4_samples.constancy_both_inverted
  (skip multisample.fbo_8_samples)
multisample.fbo_max_samples.constancy_sample_coverage          (same as 4 samples)
multisample.fbo_max_samples.constancy_sample_coverage_inverted (^)
multisample.fbo_max_samples.constancy_both                     (^)
multisample.fbo_max_samples.constancy_both_inverted            (^)

These tests fail (slightly flakily) on Google Pixel (7.1.1). The tests fail less often when run in isolation from the rest of the tests in the file.

It seems likely that this is a driver bug that is exposed by Chrome's GPU scheduling.

It doesn't seem to be fixed by
* enabling/disabling virtualized contexts
* increasing command buffer size limits (to reduce the number of command buffer flushes) - but not sure if this had an effect