[Mesa-dev] Concurrent piglit run on Skylake (was: [PATCH] i965/skl: Use an exec size of 8 to initialise the message header)

[Neil Roberts]

Ben Widawsky <ben at bwidawsk.net> writes:

> Does this happen to allow concurrent piglit to not die in a fire?

Hrm, I just tried running piglit without -1 on my wip/skylake branch
(without this patch) and it seems to work fine! I don't know what has
fixed it but these are the three patches on that branch:

d2e8cd7 i965/skl: Don't use ALL_SLICES_AT_EACH_LOD
8e17e0d i965/skl: Fix the qpitch value
41f2612 i965/skl: Align compressed textures to four times the block size

This is on top of Mesa 4deca1274c2 with the nightly kernel from the 10th
of April (ecec944313497a). If I run it without those patches then it
seems to hang around 16,000 tests in.

The second patch is fully reviewed but I don't want to land it without
first landing the first patch. That is waiting for a reply from my last
email. It would be really good to get that landed. I need to reply to
Ben's last message for the third patch.

http://lists.freedesktop.org/archives/mesa-dev/2015-March/079018.html

I should mention that I'm still filtering out some Piglit tests so I'm
actually running it like this:

PIGLIT_PLATFORM=gbm ./piglit-run.py tests/quick.py \
 results -v -x miplevel -x fbo-depth-array

Without the patches it seems to consistently hang here:

running: spec/glsl-4.30/execution/built-in-functions/cs-op-sub-mat2-float
pass: spec/ARB_depth_buffer_float/fbo-depthstencil-GL_DEPTH32F_STENCIL8-readpixels-24_8
running: spec/arb_gpu_shader_fp64/compiler/built-in-functions/op-mult-dmat2x3-double.geom
skip: spec/glsl-4.00/compiler/built-in-functions/op-div-dmat4x2-double.geom
running: spec/ARB_copy_image/arb_copy_image-formats --samples=8
skip: spec/glsl-4.30/execution/built-in-functions/cs-op-sub-mat2-float
running: spec/ARB_copy_image/arb_copy_image-formats --samples=4
skip: spec/arb_gpu_shader_fp64/compiler/built-in-functions/op-mult-dmat2x3-double.geom
running: spec/ARB_copy_image/arb_copy_image-formats --samples=2
[16062/27390] crash: 2, fail: 223, pass: 9591, skip: 6244, warn: 2 /\|/

Regards,
- Neil