This is an important topic, and you've raised a lot of very interesting questions!
Unfortunately, my last post exhausted my knowledge of the subject. I refer you to my
source, which is Ben Collins' post to debian-mentors a week or so ago:
Camm Maguire wrote:
> But fortunately, for
> all cases you mention above except the k6,k7,p2,p3, code compiled on
> one family (ev67, say) will run on ev4, though not optimally.
I assumes this means there's no alpha asm involved? If there were, you'd get illegal
instruction errors running some ev6+ code on ev5s, etc.
> In the case we're discussing here, code compiled on p3 will not run on p2,k6,k7.
I guess this means there is some "Intel"-specific asm involved.
In this case, has there been any mention of including some alpha asm in ATLAS?
Kazushige Goto's work in this area runs substantially faster (on the order of 20-50%)
than ccc-compiled ATLAS, and comes *extremely* close to the theoretical performance
of the alpha. It's all free, and available at
Welcome to the best software in the world today cafe!