Re: [Piglit] [PATCH] all.py: add run_concurrent=False where needed

2019-01-22 Thread Ilia Mirkin
Huh, interesting. There might be more going on than meets the eye. config.requires_displayed_window = true; I don't know much about accumulators, so ... perhaps they don't work with fbo the way one might think they do. masked-clear appears to have the same config.requires_displayed_windo

Re: [Piglit] [PATCH] all.py: add run_concurrent=False where needed

2019-01-22 Thread Arthur Huillet
If I remember correctly, these tests ignore -fbo when passed. At any rate, they render to a real window. If you have two of these windows at the same time because they are run concurrently, you run into the pixel ownership problem which is present everywhere and potentially extra painful on the

Re: [Piglit] [PATCH] all.py: add run_concurrent=False where needed

2019-01-22 Thread Ilia Mirkin
run_concurrent=False implies that the test will be run with "-fbo". Looking at clear-accum.c, I don't see anything that will fail to work when run with an fbo. If piglit_probe_rect_rgb doesn't work with -fbo, then there are a LOT more tests that will have this problem. -ilia On Tue, Jan 22, 201

Re: [Piglit] [PATCH] all.py: add run_concurrent=False where needed

2019-01-22 Thread Arthur Huillet
Ping. Pixel ownership test makes it impossible for Piglit to blindly assume that it can read back all the pixels of its backbuffer, so tests that don't use an FBO can't be run concurrently unless it can be ensured that the windows do not overlap. One reason this seems to only show up on NVIDIA d

[Piglit] [PATCH] all.py: add run_concurrent=False where needed

2018-11-14 Thread arthur . huillet
From: Arthur Huillet Commit 57537d45b75218438716506594e16b91dade968f removed run_concurrent=False from a bunch of tests, saying there was no reason for them to be marked as such. This is wrong for at least two tests, which require a displayed window (cannot render to an FBO). As such, when run c