[Mesa-dev] [PATCH] st/mesa: Finalize texture on render-to-texture.

[Michel Dänzer]

On Don, 2011-09-15 at 13:01 -0600, Brian Paul wrote:
> On 09/15/2011 10:08 AM, Michel Dänzer wrote:
> > On Don, 2011-09-15 at 09:01 -0600, Brian Paul wrote:
> >> On 09/15/2011 04:55 AM, Michel Dänzer wrote:
> >>> On Mit, 2011-09-14 at 11:44 -0600, Brian Paul wrote:
> >>>> On 09/14/2011 11:16 AM, Michel Dänzer wrote:
> >>>>> From: Michel Dänzer<michel.daenzer at amd.com>
> >>>>>
> >>>>> This makes sure that stObj->pt exists and is up to date.
> >>>>>
> >>>>> Fixes https://bugs.freedesktop.org/show_bug.cgi?id=39193 .
> >>>>>
> >>>>> Signed-off-by: Michel Dänzer<michel.daenzer at amd.com>
> >>>
> >>> [...]
> >>>
> >>>> Reviewed-by: Brian Paul<brianp at vmware.com>
> >>>
> >>> Thanks, Brian.
> >>>
> >>>> I assume there's no piglit regressions.
> >>>
> >>> There are no regressions per se with quick-driver.tests, but there's a
> >>> change for the fbo-incomplete-texture-03 test: Without this patch, it
> >>> fails with "FBO clear didn't work". With this patch, it passes without
> >>> --enable-debug or with GALLIUM_ABORT_ON_ASSERT=0, but there are some
> >>> assertion failures:
> >>>
> >>> state_tracker/st_texture.c:369:st_texture_image_copy: Assertion `u_minify(src->width0, srcLevel) == width' failed.
> >>> state_tracker/st_texture.c:370:st_texture_image_copy: Assertion `u_minify(src->height0, srcLevel) == height' failed.
> >>>
> >>> Not sure if this is a new problem introduced by this patch, or an old
> >>> one uncovered by it, but I'm leaning towards the latter.
> >>
> >> Yeah, I'm looking into it.  I have a patch that fixes the assertion
> >> but it needs testing...
> >
> > I'd be happy to give that a spin, but I guess it's okay to push this fix
> > in the meantime?
> 
> I guess I'd like to fix up the assertion too or we'll probably get 
> some new bug reports.
> 
> Here's my patch that seems to fix the issue:
> 
> diff --git a/src/mesa/state_tracker/st_cb_texture.c 
> b/src/mesa/state_tracker/st_cb_texture.c
> index eab02fb..538a7a3 100644
> --- a/src/mesa/state_tracker/st_cb_texture.c
> +++ b/src/mesa/state_tracker/st_cb_texture.c
> @@ -1679,7 +1679,7 @@ copy_image_data_to_texture(struct st_context *st,
>         assert(dstImage->Depth == stImage->base.Depth);
>      }
> 
> -   if (stImage->pt) {
> +   if (stImage->pt && stObj->base._Complete) {
>         /* Copy potentially with the blitter:
>          */
>         st_texture_image_copy(st->pipe,
> 
> 
> I did a quick run of piglit texture tests and didn't see any 
> regressions but I'd like to test more.
> 
> The idea with this change is that if the texture is not complete 
> (mismatched image sizes) we can't draw with it anyway so skip copying 
> the texture data into the texture buffer.
> 
> Maybe you can run some tests with this too...

This change seems to make all fbo-generatemipmap-formats tests fail.


-- 
Earthling Michel Dänzer           |                   http://www.amd.com
Libre software enthusiast         |          Debian, X and DRI developer