[Mesa-dev] [PATCH] r600g: Implement GL_ARB_texture_query_lod

[Matt Turner]

On Wed, Jul 23, 2014 at 10:19 AM, Marek Olšák <maraeo at gmail.com> wrote:
> I don't see your NAK and Ian gave me an ACK. Besides, what would the
> tests be good for if you couldn't do any regression testing because
> they would always fail due to precision issues?

Please don't top quote.

Ian reviewed before my comments. Since I'm the author of the tests had
serious doubts, I don't think his R-b is sufficient. The thread ended
with you saying you didn't have any explanation for why the hardware
didn't return the expected values -- that doesn't seem sufficient. The
patch also seems really suspect because it modifies only some of the
tests, and does stuff like loosening the tolerance of the nearest
filtering tests which should only return integral values.

We've had cases like this in the past that turned out to be driver
bugs that we wouldn't have discovered if we'd papered over the test
failures. (In particular a case where you wanted to loosen the
restrictions of an MSAA test, and after Paul NAK'd it you discovered
that the sample positions used by r600 were wrong, IIRC)

I'm okay loosening tolerances if we can come up with reasonable
explanations why, but that was definitely not the case here.