On Wednesday 22 December 2004 21:38, Gerhard Fuernkranz wrote: > > I can confirm that the speed of LCMS has noticeably improved in one of > the last few versions (I don't remember in which one), but I don't know > your performance requirements. >
If I cache the transform, lcms is fast enough for real time painting in cmyk mode on my Pentium 4M 3Ghz laptop. I convert the cmyk data using lcms to rgb in chunks of 64x64 pixels. Not that I wouldn't say no to extra speed: all the milliseconds I spend on display are milliseconds I don't have available for cool brush effects. > If you need even faster transforms, I can recommend Argyll's IMDI > routines (http://www.argyllcms.com - Do you remember? Marti has > forwarded Graeme's announcement with the new URL today). Particularly if > you can live with 8bpp input, then IMDI is _very fast_ (I measured circa > 14 Mpixels/second for a 3D -> 3D transform with 8bpp source and 8 or > 16bpp destination, on an AMD Athlon XP 2600+ with 2080MHz. This > corresponds to circa 18 frames/second @ 1024x768). I'm going to make a not of this as a possible optimization for those cases where it'll fit in Krita. > > And I suspect, that a non-portable implementation based on Intel's/AMD's > MMX/3DNow/SSE2 instruction sets could be likely even faster. > > I think PS also uses a trick to excite the impression of an instant > update: It updates the screen nearly instantly at low resolution, and > then continues in the background to refine the screen display up to full > resolution (please correct me, if I'm wrong). That's what I can see happening on my old G3 powerbook: you paint, it shows a ragged line, and then it suddenly smoothes out. -- Boudewijn Rempt | "Geef mij maar zuurtjes." http://www.valdyas.org/fading/index.cgi
pgpm94YUeWeLZ.pgp
Description: PGP signature
