4-channel vectored lookup uses SSE then 4+ channel OR multiple 4-channel vectored lookups could use AVX instructions if needed, intrinsics are similar.
On Sat, Aug 23, 2014 at 1:50 PM, Larry Gritz <[email protected]> wrote: > The OIIO TextureSystem looks up texture, for any number of channels. It's > not clear that anybody uses or needs that functionality. OSL, for example, > inherently does only 1, 3, and 4 channel lookups. > > If limiting TextureSystem lookups to 4 channels per call would result in > simplification and/or speed up, would anybody have objections to making > such a change? > > (For the odd application that needs a >4 channel lookup, it could always > do it with a loop of batches <= 4 and increasing 'firstchannel'. But I > suspect that needing >4 channels is very rare, enough to make it worth this > extra trouble in order to speed up the common case and simplify the code.) > > I'm in the middle of implementing SIMD speedups to texture lookups (for > example, using SSE for 3 & 4 channel lookups to process the channels all at > once), having good luck so far and wondering how far to push this. > > -- > Larry Gritz > [email protected] > > > > _______________________________________________ > Oiio-dev mailing list > [email protected] > http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org >
_______________________________________________ Oiio-dev mailing list [email protected] http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
