[Mesa-dev] RFC: ctx->Driver.Map/UnmapTextureImage() hooks

[Eric Anholt]

On Wed, 27 Jul 2011 20:52:00 -0600, Brian Paul <brian.e.paul at gmail.com> wrote:
> On Wed, Jul 27, 2011 at 7:29 PM, Eric Anholt <eric at anholt.net> wrote:
> > On Sat, 23 Jul 2011 11:58:22 -0600, Brian Paul <brian.e.paul at gmail.com> wrote:
> >> On Sat, Jul 23, 2011 at 9:14 AM, Eric Anholt <eric at anholt.net> wrote:
> >> > On Fri, 22 Jul 2011 14:06:48 -0600, Brian Paul <brianp at vmware.com> wrote:
> >> >> On 07/22/2011 01:32 PM, Eric Anholt wrote:
> >> >> > On Thu, 23 Jun 2011 19:08:51 -0600, Brian Paul<brianp at vmware.com>  wrote:
> >> >> >>
> >> >> >> I'd like to overhaul the part of Mesa related to texture memory
> >> >> >> reading/writing.
> >> >> >
> >> >> > OK, I'm taking a look at map-texture-image-v4.  I like what I'm seeing
> >> >> > overall, I just want to be sure that this isn't something that gets
> >> >> > squash-merged.  There's going to be breakage, and I want to be able to
> >> >> > bisect into it.
> >> >> >
> >> >> > In the metaops code, please use glBufferData instead of
> >> >> > glBufferSubData.  If you BufferSubData, I have to block on the GPU if it
> >> >> > was using that buffer already.
> >> >>
> >> >> It looks like we'd have to change that in several other places too.
> >> >> Can we do that change later?
> >> >>
> >> >>
> >> >> > In the comments for void (*MapTextureImage), please note what the units
> >> >> > of rowStride are.  I see that's present in swrast later, but I think the
> >> >> > mtypes.h and dd.h files are used for reference a lot (I do, certainly).
> >> >>
> >> >> Will do.  The parameter comments in s_texture.c are out of date too.
> >> >>
> >> >>
> >> >> > c029312ad62039904818a8b1765c6bcdf50044df is huge, and it doesn't even
> >> >> > build.  Ouch.  I think there's some room for splitting some of this up
> >> >> > so that we can get a nice series.
> >> >>
> >> >> Where's the build breakage?  I don't remember that.
> >> >>
> >> >> This was originally a long series of sometimes ugly WIP patches.  At
> >> >> one point I had a git mishap and trashed some of the intermediate
> >> >> patches.  I agree that splitting up this commit would be good, but it
> >> >> would be a lot of work that I don't really have time for.
> >> >>
> >> >> It would be great if you could do a full piglit run with the branch
> >> >> and check for i965/i915 regressions.  I'm not aware of any with swrast
> >> >> or gallium.  I'd help diagnose any regressions.
> >> >
> >> > The piglit run was in bad shape and then hung the GPU, something that
> >> > piglit hasn't done for me in a long time.
> >> >
> >> > I think I'm going to need to split up the commits to make progress.
> >>
> >> OK, I'll try to find some time to do a piglit run on my i945 and see
> >> what's up (I don't have a i965 handy).  I may not get to it for a few
> >> days though.
> >
> > I'm cutting up your patches into a reasonable series so that we can
> > actually review and bisect the code.  So far I've run my mti-tested
> > branch on snb, and mti is your code I'm rebasing onto it as I cut chunks
> > out.
> 
> Thanks for doing that.
> 
> I found that the map-texture-image-v4 branch is missing some stuff
> that I lost during my git accident.  I think I'm nearly done restoring
> it.  I'll push it to the branch asap, maybe tonight.

New mti-tested branch is up.  Splitting up commits has paid off for at
least 3 regressions due to driver bugs so far that would have been awful
to work out otherwise.  I'd like to land these bits before we try
tackling the rest of this mess (particularly the move of fields to
swrast didn't look like it was going to work for hardware driver
fallbacks, the way it was done).  I'd also love to see some serious
review of this code.

mti-tested is not regressing on intel.

radeon is down to 1 regression on the very tip commit now that I got
some hardware -- gen-compressed-teximage fails because the GetTexImage()
of the meta-decompressed texture returns garbage.

softpipe is thoroughly regressed, probably because I missed a required
change there.  Just noticed it tonight, will look into it next.

I expect nouveau is rather broken for compressed textures, because I
didn't catch it up to things I noticed while working on other drivers.
I'm hoping I have some hardware that works (is gf 7600/7800 the right
kind for this driver?), but I might not get to it before siggraph.

My DRI1 system is so slow and buggy that I'm having a hard time testing
it.  This branch shouldn't break anything there as far as I can see, but
I'll try again on testing.  Also, we could delete a bunch of dri1 driver
code by implementing mapteximage in them instead.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20110803/7cb03708/attachment.pgp>