On 08/31/2010 02:57 PM, Ian Romanick wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
We'd still like to release Mesa 7.9 at the end of September. That's
four weeks from now. As soon as the glsl2-loops branch lands (I'm
expecting this to be this week), all of the major development on the new
compiler will be complete. There's still a fair amount of bug fixing to
be done, but that feels like the stabilization phase leading up to a
release.
Looking at the result of a bugzilla search for "glsl" I see a few
categories of bugs:
- Old bugs. Everything before #29044 was filed before the glsl2 work.
This accounts for almost half of the bugs (35 of 72).
- Bugs that only occur on softpipe or llvmpipe. Since these bugs
aren't reproducible in swrast, i965, or r300c, my first guess is that
the problem is either in the Gallium code or the Mesa IR to TGSI
translation layer. Someone familiar with that code is going to have to
do *some* leg work here. It may not be (as was the case with reads from
shader outputs), but I don't have the information to determine that.
I'm trying to look at some of these in my spare time. But it's really
easy to run the gallium softpipe driver, as Dave said.
- Additional regressions in applications that weren't completely
correct with the old compiler. As far as I'm aware, the regressions in
the Humus and Unigen demos are all added failures.
- Lack of noise support. This is really only an issue for swrast. As
far as I'm aware, none of the hardware drivers actually support the
noise opcode generated by the old compiler. Our plan here is to
implement noise in GLSL. There are a couple issues with that,
especially on old chips like r300 and i915, and it doesn't seem like a
high priority.
- A small number of assertion failures. I think #29573 and #29687 are
the only ones left. I believe Eric is working on these.
Looking at this, it seems like it won't take too much work to have
things in much, much better shape with the new compiler than they were
in with the old compiler. Even if it's not quite perfect, and it won't
be, getting a release out is a good thing.
So... what are our collective criteria for a 7.9 release to happen?
Ian, I'm still getting up to speed on the new compiler, but it looks
like a full 4-element temp/const vector is allocated for each GLSL
float. The previous compiler would try to pack four floats into a
single vector whenever possible. This could dramatically reduced the
number of vector temps/consts needed.
I'm worried that some apps that use large shaders will start failing.
-Brian
_______________________________________________
mesa-dev mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
