On Fri, Aug 15, 2014 at 11:19 AM, Brian Paul <[email protected]> wrote: > On 08/15/2014 08:52 AM, Ilia Mirkin wrote: >> >> Brian, >> >> You recently went on a campaign against setting .window_width/height >> in piglit tests, in order to fix issues on Windows I guess? I noticed >> that a bunch of failures on a NV17 card (GeForce4 MX 440) are due to >> the fact that the window sizes are now 160x160 (NPOT) and the tests >> use that in order to size their textures. >> >> What is the proper way to resolve this? Temporarily I've changed the >> default window width/height to 128, but that's not going to work more >> generally for the same reason you increased it to 160x160. Should I >> just do so on tests I identify as runnable on nouveau_vieux and that >> also have this issue? Perhaps increase to 256x256? Something else? >> >> You can find such tests with >> >> git grep -C2 piglit_width tests/fbo | grep glTexImage >> >> Not all of them run on nouveau_vieux, but some do. >> >> Thanks for any advice, > > > If a test like fbo-alphatest-formats.c creates a texture equal in size to > the window and the window size is not explicit and the test doesn't check if > NPOT textures are supported, I'd say the test is broken. I didn't see any > regressions when I removed the original window size because llvmpipe and > more recent nvidia hardware support NPOT textures. > > For these tests, the simplest thing is probably just setting the window size > to 256x256. Otherwise, we should probably create the texture with an > explicit POT size and check that the window is at least that size before > probing. See sample patch attached.
I think the path of least resistance may be to just increase the window size to 256x256 for those tests. I've set it to a POT size globally for now, will compare results and send a patch updating the tests that end up being executable with nouveau_vieux. (Although I guess even GeForce FX 5000 series doesn't have NPOT support, at least not with the nv30 driver.) _______________________________________________ Piglit mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/piglit
