Archived

This topic is now archived and is closed to further replies.

LoreKeeper

Okay, 'nuf Talk

Recommended Posts

Hey - here''s the first release version of the math stuff I''ve been bugging you all about. Several versions of routines are available, a "reference" version high-level routine, an assembler version, and then 3DNow, SSE and a few SSE2 versions. Particularly the SSE2 MatrixMatrix multiply is extremely fast. Two types of matrices and vectors are supported: 3x3, 4x4, 3x1 and 4x1. Two downloadables: math.zip (433kb) [link]http://www.cs.sun.ac.za/~henri/math.zip[/link] contains two exes that do performance testing, OptimizationTest.exe uses the windows performance timer and times across 1 000 000 repeats of every math routine. RDTSCOptimization.exe uses the rdtsc timer to measure "exact" clock-cycle counts of every routine (note that cache effects and the like can make results look different each time). mathsrc.zip (33kb) [link]http://www.cs.sun.ac.za/~henri/mathsrc.zip[/link] contains the necessary source code to compile the above programs. All routines have been tested, but I''m sure a few bugs and errors have slipped in. Improvements are always possible though, so suggestions are welcome. Henri

Share this post


Link to post
Share on other sites
To clarify things then...
The Exes produce testresult text files that show the timings for the math routines - this can give you an idea of how much faster your programs could work and which routines are most suited for your needs (sometimes the basic assembler version is better then the 3DNow/SSE versions).

Also you are welcome to spot flaws/errors/possible-optimizations.

The Source zip contains the actual math libraries, and you are welcome to do whatever you like with them (though credit would be nice ).

Share this post


Link to post
Share on other sites