Home

Overview

Why MMX

MMX in-depth

Future Development

MMX Future Development

MMX technology brings more power to multimedia applications. It adds new data types, instructions and hardwares that can process data in parallel to speed up the process. In short, MMX is the RISC part for a CISC processor.  As we all have seen the changes in MMX, it's not a typical techincal change, rather it's a fundimental conceptural change. The fundamental difference between CISC processor and RISC process is really the pipline, the idea of excuting code in parallel. This is exactly what MMX has adapted. As for the future development, we would suspect more changes toward in this direction.  And those changes will be symblized by the following

A smaller but more concentrated instruction set. The old IA has too many instrcutions that are rarely used by most of the programers, meanwhile this huge IA increases the unnecessary complexity of the hardware organization and makes it unprofitable for untilizing the parallel excution.  To reduce instructions will allow all the hardware designer to charactlize it better, and then more paralleled excution can take place.
More and better parallel algorithm. As we can see, the improvement in performance is bought up by not only the MMX technology, but also the IA superscalar architecture. This means, hardware can only play so much (the rule of thumb) that we have to have a good software (complier) to operate it. Examples are like, RISC architecture's scoreboard, loop unrolling, etc).

Do you smell what we are cooking? Yes! Those changes are precisely the characteristics of a RISC processor.  The ground of this suspected change of ours is simply due to the fact that RISC processor does offer a much better data processing algorithm -- the pipline line and everything else that comes with it.  (The second reason is because we all have taken CMSC411 very seriously!!!!!) And it's almost impossible for CISC architecture to compete.

However, this change will never happen if Intel is the only one that pushes it. All the big software companies will have to take part of it. Things like a good OS that can be "smart enough" to utilize the new processor, etc.. All those thing add up, there is still a tremndous amount of work should be done (or should have been done).  Let's just hope that someday intel can produce something that is better than what they are producing now!!!

 

FINAL WORDS

At the end of project we'd like to take our time to thank Dr. Hugue for giving us this opportunity to do this project. If it's not required, changes are, we may never get this far on MMX.  This has been a great project and we all enjoyed it very much!!!

 

Copyright University of Maryland.
For problems or questions regarding this web contact [ProjectEmail].
Last updated: December 21, 1998.