Edouard Gomez ([EMAIL PROTECTED]) wrote: > Edouard Gomez ([EMAIL PROTECTED]) wrote: > > Here is an updated patch... still no SIMD... might come later. > > Ok found some time to write the kernel of the computation in SSE
Ok i made great progress here, but i'm still not pushing the new patch because i'm not happy with the 3dnow functions yet. For the SSE needs i now precompute the same luminance table as for fixed point calculations but in float (for sure!). It saves me the div which is 32 cycles long on P4. I also added the luminance calculation to the already implemented RGB processing as a fourth component, so we get that value for free too (comparing cycles to compute RGB before and after). I had just to add the final mulps to use the mapped from LUT Y'/Y value and a few pshufps which should not impact performance that much per pixel. The C version has been tested a bit more and bugfixed for some cases where the precision might overflow. Once i have a good 3dnow version... i post the patch again, this time complete. -- Edouard Gomez _______________________________________________ Rawstudio-dev mailing list [email protected] http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-dev
