Allen Akin wrote: > On Wed, Aug 23, 2006 at 02:58:44AM +0200, Roland Scheidegger wrote: > | > I tried this, but glean gives me just failures (result attached), do > | > I need something else to enable? I have up-to-date mesa-cvs and drm. > | No that should be all needed. What is the output of glean itself? Maybe > | it's possible to get information out of that weird results file but I > | know I can't. > > That's to be expected -- the results file isn't intended for humans to > read. As you said, the right place to look is in the output of the run. > (In theory it should be possible to regenerate the output from the > contents of the results file, but currently not all the tests record > enough information to do that.) > > Depending on the test, running with increased verbosity (the "-v" > option) might also help. > > | And, what card is this? When I run it on my radeon 9000 (rv250), I get > | only 2 (repeated for all visuals) failures: > | FAILURE: > | Program: LIT test 1 > | Expected color: 1, 0.65, 0.443, 1 > | Observed color: 1, 0.65098, 0.431373, 1 > | FAILURE: > | Program: LIT test 2 (degenerate case: 0 ^ 0 -> 1) > | Expected color: 1, 0.65, 1, 1 > | Observed color: 1, 0.65098, 0, 1 > | > | Both are expected, r200 (and r300) apparently can't handle the > | degenerate case (that's really nitpicking), > > I didn't write that test, and I haven't looked into the logic behind it, > but it's possible that the OpenGL spec makes some guarantee about the > results in that case. Need to do a little research. > > | whereas the failure in LIT test 1 is actually not a failure at all > | (glean expects an approximation but with too tight tolerance, the value > | is actually "too good"). > > This is one of those situations in which the tolerance can't reliably be > based on the depth of the color channel, but might need to be based in > part on some internal precision that there's no way to query. > > That said, it looks to me like the main problem is that the expected > value in the test isn't accurate enough -- it should be about 0.430 > rather than 0.443. The tolerance itself looks reasonable. Since I'm > not familiar with the test I might be missing something, though.
Yes, it should be 0.430 - I'll fix that. As for 0 ^ 0, it is a corner case, but the ARB_v_p spec explicitly says that 1 is the expected result. I only ran the tests with Mesa's software renderer and an NVIDIA card. It could be that we need to loosen up the tolerance for other OpenGLs. -Brian ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 _______________________________________________ Mesa3d-dev mailing list Mesa3d-dev@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
