On Mon, 15 Aug 2011 16:05:51 -0600, Brian Paul <brian.e.p...@gmail.com> wrote:
> On Mon, Aug 15, 2011 at 3:43 PM, Ian Romanick <i...@freedesktop.org> wrote:
> >> + /* Create/bind FBO/renderbuffer */
> >> + if (decompress->FBO == 0) {
> >> + _mesa_GenFramebuffersEXT(1, &decompress->FBO);
> >> + _mesa_GenRenderbuffersEXT(1, &decompress->RBO);
> >> + _mesa_BindFramebufferEXT(GL_FRAMEBUFFER_EXT, decompress->FBO);
> >> + _mesa_BindRenderbufferEXT(GL_RENDERBUFFER_EXT, decompress->RBO);
> >> + _mesa_FramebufferRenderbufferEXT(GL_FRAMEBUFFER_EXT,
> >> + GL_COLOR_ATTACHMENT0_EXT,
> >> + GL_RENDERBUFFER_EXT,
> >> + decompress->RBO);
> >> + }
> >> + else {
> >> + _mesa_BindFramebufferEXT(GL_FRAMEBUFFER_EXT, decompress->FBO);
> >> + }
> >> +
> >> + if (destDataType == GL_SIGNED_NORMALIZED) {
> >> + rbFormat = GL_RGBA32F;
> >> + } else {
> >> + rbFormat = GL_RGBA;
> >> + }
> >
> > This bit seems odd. Is GL_RGBA32F the right choice? Won't this fail on
> > hardware that can do SIGNED_NORMALIZED but not float rendering?
>
> That bit wasn't in my original patch series. I'm not sure if it's
> even legal to do a glGetTexImage() of a compressed texture and get
> back signed normalized values. I'd have to do some digging through
> specs.Yeah, we'd get in trouble here for !EXT_texture_float. This was the fix for EXT_texture_snorm regression in fbo-generatemipmap-formats of compressed formats once that changed to doing metaops+GetTexImage. Not sure how to handle it -- we need the signed values, but getting them through the pixel path as signed bytes was awful. Thoughts?
pgp1xx2ffSc2m.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
