[Mesa-dev] i965: 12 more depthstencil tests fixed

[Eric Anholt]

I hate this code.  I've written 5 branches trying to fix this problem; this is
one of two non-regressing branches, and it's the one I hate less.

At this point, the 1024-sized tests are all passing, and only 292/146 tests
are left failing.  I haven't debugged any of the remaining failures.  The
simulator complains about hiz-related 8x4 alignment issues on both the 1024
and 292 tests, but that's all I see from it.

It also is known broken for array/cube depthstencil textures and the rework to
fix that is kinda scary.  Or I think we can eat some performance in that case
by plumbing a nasty hack down through a few layers, but I really don't want to
go there.