Paul Berry <stereotype...@gmail.com> writes: > Geometry shaders that run in "DUAL_INSTANCED" mode store their inputs > in vec4's. This means that when compiling gl_PointSize input > swizzling (a MOV instruction which uses a geometry shader input as > both source and destination), we need to do two things: > > - Set force_writemask_all to ensure that the MOV happens regardless of > which channels are enabled. > > - Set the source register region to <4;4,1> (instead of <0;4,1> to > satisfy register region restrictions.
This sure sounds like something you empirically found, but I'm confused. I would have assumed that DUAL_INSTANCED with an instance count of 1 meant that the channel enables the hardware gave you had the first 4 enabled and the second 4 disabled. And since the dst.width (and thus execsize) is 4, whether or not the second 4 are disabled wouldn't matter. In that case, why do you need the writemask forced, since just the 4 channels you care about will be affected, anyway? And, if dst.width == 4, then execsize == 4, and I'm confused what register region restriction is being honored by promoting the hstride to 4. Putting these fixups for a couple of weird cases in just MOV and ADD feels wrong to me, but maybe when I understand better what's going on it'll seem more natural. Patches except for 5, 6 are: Reviewed-by: Eric Anholt <e...@anholt.net>
pgpj74F1y1o5U.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
