[Libre-soc-dev] Hardware-accelerated specialized instructions
programmerjake at gmail.com
Fri Dec 11 00:41:22 GMT 2020
On Thu, Dec 10, 2020, 16:30 Cole Poirier <colepoirier at gmail.com> wrote:
> On Thu, Dec 10, 2020 at 4:25 PM Jacob Lifshay <programmerjake at gmail.com>
> > On Thu, Dec 10, 2020, 16:14 Luke Kenneth Casson Leighton <lkcl at lkcl.net>
> > wrote:
> > > you make a hardware triangle fill unit like Jeff did, it's not going
> > > to do multiplication, is it?
> > >
> > it might, since the per-pixel function is just a vec4 dot product
> > by a reciprocal and a pile of mul-adds.
> So the hypothetical pixel instruction in this case would use a mul
> unit, an add unit, and a specialized reciprocal unit?
for the dot product, we'd probably want a 32x32->64 mul-add since integers
are needed to avoid holes between polygons and 32-bit integers are slightly
too small to hold the full product.
So, maybe we do need the widening mul anyway.
More information about the Libre-soc-dev