Author: Matt Taylor
Date: 22:45:16 12/23/02
Go up one level in this thread
Sigh... I've wasted -way- too much time debugging these silly routines. I now have 2 gripes with MMX: 1. No 64-bit arithmatic (add/sub/neg) 2. No 64-bit compare (pcmpeqq specifically) (For the record, SSE 2 adds 64-bit add/subtract to MMX.) Anyway, my emulation of such missing things is buggy. I copied all the code into a test project, and I'm going to debug them all before I post any more results. I've got 14 different functions now, and the variety of data is quite interesting. My hand-optimized version of the latest routine you posted isn't 14% faster, but it is faster than bsf/VC-optimized table routine. Also, some of my MMX routines are really poor; I haven't worked enough with MMX to fully understand how to optimize it. I'm familiar with the instruction set, but I've only done 1 other routine (antialiasing routine) in MMX. No updates tonight, and tomorrow I've got a lot to get done. I've got to pick up my car from the shop. Wednesday is Christmas, and I'll be spending the whole day with family. If I can work out all the bugs, I may be able to update the data on Thursday. I'll want to update after that, too, assuming someone helps me with my MMX. -Matt
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.