Chad Versace <chad.vers...@linux.intel.com> writes: > + /* Generate multisample configs. > + * > + * This loop breaks early, and hence is a no-op, on gen < 6. > + * > + * Multisample configs must follow the singlesample configs in order to > + * work around an X server bug present in 1.12. The X server chooses to > + * associate the first listed RGBA888-Z24S8 config, regardless of its > + * sample count, with the 32-bit depth visual used for compositing. > + * > + * Only doublebuffer configs with GLX_SWAP_UNDEFINED_OML behavior are > + * supported. Singlebuffer configs are not supported because that would > + * require that rendering be eventually written to the singlesample buffer > + * even if DRI2Flush is never called; yet we downsample to the > singlesample > + * buffer only on DRI2Flush. GLX_SWAP_COPY_OML is not supported because > we > + * have no tests for its interaction with MSAA. > + */
We actually need to remove our claiming of GLX_SWAP_COPY_OML in general, because pageflipping means that we don't actually support SWAP_COPY. We only do UNDEFINED. I'd say instead "singlebuffer configs are not supported because nobody wants them". I think all you need is (pessimistically) intel_downsample_for_dri2_flush in intel_flush_front() to make front buffer rendering actually work, and it's a problem that exists even in a doublebuffer config. I think this concludes my review. Great work! I'm excited to see this finally land.
pgpnRDmNu36bc.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
