[Mesa-dev] [PATCH 00/18] More t_dd_dmatmp janitor work and misc other

[Ian Romanick]

This is the last of the t_dd_dmatmp janitor work that I plan to do.

The first patch is a change suggested by Ville while I was reviewing
some of his patches to t_dd_dmatmp earlier this year.  I re-found
Ville's patches while I was searching my inbox for reviews to my
previous t_dd_dmatmp series.

The next big chunk is Ville's series.  I have reviewed most of these.  I
am resending them because they depend on the first patch, and I have
modified some of the patches to account for other changes in
t_dd_dmatmp.  At the end of that bit is a couple refactors of my own.

There are some patches from Ville's original series unrelated to
t_dd_dmatmp that I have already pushed with my R-b.  The patch "i915:
Fix t_vb_rendertmp.h's provoking vertex handywork" still needs to go
back out, but I want to understand it more myself first. :)

These are followed by two patches to the radeon driver.  These mimic
changes that Ville made to the i915 driver.

Finally, there are a few additional cleanups to t_dd_dmatmp.  Two of
these could have gone at the front, but, basically, I was too lazy to
rebase and fix them up.  Also, "t_dd_dmatmp: Require HAVE_POINTS" would
have to be at the end anyway.

I ran piglit on a Radeon 7500 (radeon_dri.so) at several points through
the branch.  There were no regressions at any of the tested commits, but
there are a lot of tests on Radeon that crash.  Most of the crashes are
attributable to a single assertion failure.

    radeon_fbo.c:454: radeon_unmap_renderbuffer: Assertion `ok' failed.

I submitted https://bugs.freedesktop.org/show_bug.cgi?id=92184 for this
issue.

I also ran piglit on a G33 (i915_dri.so) at pretty much every commit,
and there were no regressions there either.  I did noticed that
tex3d-maxsize sometimes reports "incomplete."  This only seems to occur
during full piglit runs, so this could be a concurrency issue.