This is far from Open-graphics but there is an improvement area in the design of new instructions.
The new instruction must be faster than a SW implementation, this is not so obvious if some memory access are involve. The use of a new instruction discard the need of the communication by a specialised "core". If the data packet is to small, it"s most of the time faster to compute it on the cpu. Using a new instruction on an existing CPU, enable the use of all the existing compiler suite which are the most complexe and long task to complete. This ease also the use of the hardware under an OS: no needs for a drivers or for ressource management, etc... This enable also the resuse of pre-existing SW satck. I think this is the easiest way to have more processing power without the cost in terme of complexity (out of order cpu) and power consumption. On generic computing, cpu intensive task are well defined : - idct on all codec processing (this include JPEG but also mpeg2, mpeg4, mp3,...) that's 90% of the cpu time - color conversion - image zooming - compression: gzip but also lzma/lzo scheme - crypto : AES, DSA,... - IP stack - string processing,... For memory intensive task, you 'only' need more bandwith :) So a single cycle instruction, using a new set of register or the old one, could be a good candidate. You could need a new registers set for AES that needs 256 bits chunks of data, for example. But then you need to modify the OS, for saving and restoring those register. openRISC could be a good base for changing a GPL cpu. Regards, Nicolas Boulay _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
