On 2 February 2018 at 03:40, Roland Scheidegger <srol...@vmware.com> wrote: > I suppose this is a bit inefficient if the hw actually supports some > double ops? > I mean for uint->double you could just convert the low 8 bits and high > 24 bits (or different split) separately to float, convert both to double > and add them. (I think should work for int->double exactly the same > despite the sign bit.) > But in any case, for hw not supporting double ops at all this looks like > the right approach.
I've sent a patch that did this instead, definitely smaller instructions count. Consider this series dropped for now, Elie don't worry we'll get soft fp64 in there yet :-) Dave. _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
