On 18 aug 2010, at 17.08, Luca Barbieri wrote:
I have a feeling that CL performance will not matter that much for
nvfx and r300, compared to nv50 and r600.
Sure.
The point is that if you can't use normalized coordinates at all on
PIPE_TEXTURE_RECT, you can't implement OpenCL well on nv50 and r600.
Hence, that should be allowed on OpenCL capable cards (and not on
others).
Well a prerequisite for OpenCL would be PIPE_CAP_ADVANCED_SAMPLERS and
with that cap the state tracker shouldn't use PIPE_TEXTURE_RECT for
GL_TEXTURE_RECT but instead normal 2D textures and then just use
unnormalized/normalized as it see fit. Or just make
PIPE_CAP_ADVANCED_SAMPLERS also mean that PIPE_TEXTURE_RECT can be
treated as normal 2D target. Either one is fine, but for now we don't
have to worry about this.
r300 having to emit a shader instruction unconditionally in this case
really isn't that big of a deal.
On nVidia cards, this doesn't matter, since they never prefer the
normalization opposite to the OpenGL-required one.
So if the maintainers of Radeon and software renderers, who might
care, don't, I suppose we could as well not introduce the feature,
even though I'm not totally convinced of this.
Cheers Jakob.
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
