On Sat, 29 Oct 2011 11:43:37 -0600, Brian Paul <brian.e.p...@gmail.com> wrote: > On Fri, Oct 28, 2011 at 1:50 PM, Eric Anholt <e...@anholt.net> wrote: > > In all of piglit, only two tests hit it (reading to RGBA float, where > > GetRow would drop floats into place from R, RG, or RGB). > > So if an app does glReadPixels(format=GL_RGBA, type=GL_UNSIGNED_BYTE) > is it a slow path now? > > In the past, quite a few apps that used glReadPixels used that > format/type combo, or GL_RGB/GLubyte. I think it would be good to > keep those off a slow path that involves a lot of ubyte/float > conversion.
That fast path (or the memcpy one) basically hasn't been hit since _ClampReadColor came along and made transferOps always be set. I want to optimize out that clamping again, but these "fast" paths are totally overwhelmed by the cost of going through non-movntdqa or cached blit-based mappings on hardware drivers anyway. If I can get this landed and finish off my GL3 tasks (some of which are pending on this), I'm hoping to add Mesa code for movntdqa to get ~10x readpixels performance improvements on hardware drivers that can't do cached blits.
pgpZpS5Ky3gky.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
