[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Math-atlas-devel] Re: Athlon 1.2Ghz results == P4 1.5

>Thanks for this clarification, Peter!  Clint, I wonder if our results
>are consistent with what Intel knows.  You have some contacts with
>them, no?  Lastly, anyone timed Julian's kernel on a P4?  My apologies
>if I've missed a previous posting to this effect.

Just for fun, I tried it, and it didn't do too hot.  You know it's not going
to rock and roll, because on the P4, x87 code has 1/2 the theoretical peak
SSE2 enjoys.  It doesn't compete well with the x87 code either, which I
think is down to block size more than anything.  Julian's x87 code gets
very close to the best kernels on the PIII and PPRO (PIII has prefetch,
and PPRO has 8K cache that nb=30 is good for, PII is bad with no
prefetch and larger cache) . . .