hello all,
yesterday i had a computer graphics university lecture on gaming history and how in this example id software used interleaving on the pentium to calc their divisions for perspective correction for in fact no cost (only every 16 pixels, interpolating between them, that was another optimisation ;).
now i'm wondering if all newer cpus (pentium up) support interleaving (p4, amd athlon?) and furthermore, exactly which commands can be interleaved with which other commands? and how exactly? is it all floating point commands? what about mmx commands - they are using the fp registers...?

i'm really interested in this technique as it seems to present some serious optimisation potential and can find by far too less information about it, help and examples would be very appreciated! :D

thanks a lot in advance,
gx, buggy
Posted on 2003-03-25 03:07:41 by BugByter
www.agner.org has some nice documentation on pentium-style optimization. Interleaving MMX and FP code is definitely a no-no, and also be careful not to switch too often between MMX and FP code (might be less performance penalty on newer processors, some of the clever heads can probably tell you more ;-)).
Posted on 2003-03-26 02:07:07 by f0dder