Thread: [Lcms-user] New minor revision available on site
An ICC-based CMM for color management
Brought to you by:
mm2
From: <ma...@li...> - 2002-08-07 14:36:53
|
Hi, Regarding those problems recently reported by Armindo and Auke Nauta, I've put a slightly modified packages in site. It solves following problems: - Removed a spare 'printf' used by debug purposes and left by mistake. I wonder if this is the cause of the corruption in BCB and Delphi 6 - Solved a memory overwritting on precalculated CMYK -> Hexachrome transforms - Added 16 bit Gray+3 alpha format (a very weird one!) I wish to thank all people, specially James Burgess, that has given a hand in fixing those bugs. Regards, Martí Maria The little cms project http://www.littlecms.com ma...@li... |
From: ADS <tec...@wa...> - 2002-08-07 16:20:40
|
Marti, any plan to add mmx support and prefetching in lcms? thanks Armindo |
From: <ma...@li...> - 2002-08-08 08:45:40
|
Hi Armindo, > any plan to add mmx support and prefetching in lcms? Belive or not, I tried it. And found MMX is not suitable. lcms is doing all computation on 16 bits per sample, MMX has support for 8 bps. Of course a implementation of trilinear or tetrahedral interpolation using MMX will spedup things, but the price would be very hight. A portability nightmare, plus special handling of cmsFLAGS_NOTPRECALC (MMX and floating point doesn't mix very well) plus assembly coding... Too many penalties for a small speed gain. I look forward for very optimizing compilers that translates pure C to MMX that, no doubt, would be the best solution. As a sample, in ver 1.07 I did remove asm implementation of trilinear interpolation and left a C version only. Some compilers does generate *better* code that my hand-optimized asm. At that point I drop any asm support ;-) About prefetching.... If you mean caché, the cost of hash function is most times larger that the transform time. So a cache will take a lot of memory only to slow down some transforms. Sounds incredible but is quite true. Regards, Marti. ----- Original Message ----- From: "ADS" <tec...@wa...> To: <lcm...@li...> Sent: Wednesday, August 07, 2002 1:00 PM Subject: Re: [Lcms-user] New minor revision available on site > Marti, > > any plan to add mmx support and prefetching in lcms? > > thanks > > Armindo > > > > ------------------------------------------------------- > This sf.net email is sponsored by:ThinkGeek > Welcome to geek heaven. > http://thinkgeek.com/sf > _______________________________________________ > Lcms-user mailing list > Lcm...@li... > https://lists.sourceforge.net/lists/listinfo/lcms-user > > |
From: ADS <tec...@wa...> - 2002-08-08 09:20:59
|
Marti, just for information a few links (that contains Image processing Delphi/Kylix samples). http://esibert.developpez.com/delphi/prefetch/ http://esibert.developpez.com/delphi/mmx/mmx.htm http://esibert.developpez.com/delphi/simd/sse.htm Armindo |
From: James B. <jr...@pi...> - 2002-08-08 16:43:12
|
How about AltiVec then? :-) On Thursday, August 8, 2002, at 07:08 AM, Mart=ED Maria wrote: > Hi Armindo, > >> any plan to add mmx support and prefetching in lcms? > > Belive or not, I tried it. > > And found MMX is not suitable. lcms is doing all computation > on 16 bits per sample, MMX has support for 8 bps. > |
From: ADS <tec...@wa...> - 2002-08-08 16:54:50
|
>How about AltiVec then? What is AltiVec? |