[Mesa-dev] [PATCH 1/3] mesa: Only end render-to-texture at bind time for drawbuffers.

[Eric Anholt]

On Fri, 10 Feb 2012 16:19:14 -0700, Brian Paul <brianp at vmware.com> wrote:
> On 02/10/2012 03:53 PM, Eric Anholt wrote:
> > If we're only starting for new draw buffers, why would we end for old
> > read buffers along with draw buffers?
> > ---
> >   src/mesa/main/fbobject.c |    7 ++-----
> >   1 files changed, 2 insertions(+), 5 deletions(-)
> >
> > diff --git a/src/mesa/main/fbobject.c b/src/mesa/main/fbobject.c
> > index 987d687..64c3393 100644
> > --- a/src/mesa/main/fbobject.c
> > +++ b/src/mesa/main/fbobject.c
> > @@ -1758,11 +1758,8 @@ _mesa_BindFramebufferEXT(GLenum target, GLuint framebuffer)
> >      if (bindDrawBuf) {
> >         FLUSH_VERTICES(ctx, _NEW_BUFFERS);
> >
> > -      /* check if old read/draw buffers were render-to-texture */
> > -      if (!bindReadBuf)
> > -         check_end_texture_render(ctx, oldReadFb);
> > -
> > -      if (oldDrawFb != oldReadFb)
> > +      /* check if old framebuffer had any texture attachments */
> > +      if (oldDrawFb)
> >            check_end_texture_render(ctx, oldDrawFb);
> >
> >         /* check if newly bound framebuffer has any texture attachments */
> 
> This looks OK to me, but it's one of those things that deserves a lot 
> of testing.
> 
> FWIW, I've never been totally happy with the 
> ctx->Driver.RenderTexture() and FinishRenderTexture() hooks.  It's 
> been hard to concisely define when RTT starts and stops.
> 
> Maybe we should just have a driver hook to create a renderbuffer that 
> wraps a texture slice when glFrameBufferTexture() is called.  Then, 
> leave it up to the driver to figure out the rest at draw-validation time.

This patch is not strictly needed for the rest of the series, so I'll be
fine with reverting it if things go weird.  It at least survived a
piglit run for me.

You may have seen the commit message of despair in 3/3.  I had an
alternate series that made RT/FRT actually occur only during drawing to
the buffer, but then I lost my gl_renderbuffers for attachments and
things went badly.  If we separated RT into RT and BeginRenderTexture
(and that operation was paired with FinishRenderTexture), it might be
useful.  But the only thing I would need to do in BRT is the gen4
workaround, which went fairly naturally into
brw_update_renderbuffer_surface().  So at least for my needs, I can
leave RT as being "called at least once to set up my
gl_renderbuffer". and the actual tracking of render-to-texture lives in
the driver.  I do rely on FinishRenderTexture currently to get the
flushing right, but it wouldn't be hard to live without.

-------------- 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/20120210/2017fa03/attachment.pgp>