Paul Berry <stereotype...@gmail.com> writes: > On 21 October 2013 18:34, Eric Anholt <e...@anholt.net> wrote: >> Another option: How about instead of that assert in brw_eu_emit.c, we >> just smash the vstride to be width * hstride? We know the vstride >> doesn't matter, because you're only using execsize components, so let's >> just not bother our brw_eu.c callers with this little problem. >> > > I'm not really a fan of this--it feels like it's perpetuating the "<8;8,1> > instead of <4;4,1>" lie by introducing even more code in brw_eu_emit.c to > fix things up when we use a bogus <8;8,1> region. Also, if we do this, > there's no convenient place to put the assert(force_writemask_all), which I > think is really valuable. > > I would rather stick with the fixup at the top of > generate_vec4_instruction() for now, and revisit the decision when we get > rid of guess_execution_size().
This is totally fine with me.
pgp2CTLH0yEYy.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
