In the past few days I have seen intermittent failures for tests in this category:
piglit.spec.arb_texture_multisample.texelfetch fs sampler2dmsarray
/tmp/build_root/m64/lib/piglit/bin/texelFetch fs sampler2DMSArray 4 98x1x9-98x129x9 -auto -fbo
piglit: debug: Requested an OpenGL 3.1 Forward-Compatible Core Context, and received a matching 4.2 context
Probe at (500,16)
Expected: -0.000000 0.333333 0.625000 0.500000
Observed: 0.098039 0.098039 0.098039 1.000000
Also, piglit.spec.khr_texture_compression_astc.array-gl is newly intermittent:
/tmp/build_root/m64/lib/piglit/bin/khr_compressed_astc-array_gl -auto -fbo
Probe at (0,1)
Expected: 0.156863 0.101961 0.129412 1.000000
Observed: 0.909804 0.831373 0.784314 1.000000
Because these failures are intermittent, I'll need to do some work to bisect.
Dave Airlie suggests these may be blorp related. I'm disabling them in the CI until this is resolved.
first recorded sighting on Intel's CI:
piglit=e43a23e ext_image_dma_buf_import: Set tolerance to 565
mesa=0c02d70 i965: Don't use fast copy blit in case of logical operations other than GL_COPY
It took me half a day to bisect it to Anuj's series, ending in:
i965: Don't use fast copy blit in case of logical operations other than GL_COPY
XY_FAST_COPY_BLT command doesn't have a field for raster operation. So, fall
back to using XY_SRC_COPY_BLT to handle those cases.
Fixes piglit test gl-1.1-xor-copypixels when fast copy blit is enabled
for all tiling formats.
Signed-off-by: Anuj Phogat <firstname.lastname@example.org>
Reviewed-by: Matt Turner <email@example.com>
The failures are not easy to reproduce, and it can take several *full* runs to see them. I'm confident that it is from these commits, because I looked at Anuj's pre-commit developer builds, and also found the regressions in those results.
I don't see anything in the commit 0c02d70 which can cause this failure. Fast copy blit is not yet enabled on skylake. So, this commit should cause no functional change in any test.
I just re-tested. For SKLGT2, 0/7 piglit runs produced one of these failures for mesa c52e92ec3a37c9ab3fb35132e62e1ddf6a770c27
With mesa 0c02d7002d6c005b4c1fe997b5ef5916978dd183, 5/7 piglit runs produced between one and three test failures.
reportedly fixed by df210ff