Hi again,
At 07:52 PM 4/26/2001 +0000, Brian J. Beesley wrote:
>As opposed to 10.32 million with the 8087 code? If so, that's pretty
>good - I thought you might have lost more precision than that.
I haven't figured out the upper limits yet. I expect the limits to be roughly
what Ernst Mayer's program uses. As I recall, his limits are about 2% less
than prime95's.
> > 1.4GHz P4, new code: 0.048 sec.
>Not bad at all, especially if you still have PC600 memory!
>
>Intel could use some good
>publicity; I hope they reward you handsomely for this work, which
>surely must have some impact on sales.
I seriously doubt GIMPS benchmarks will have an impact on P4 sales!
>How many data passes per iteration? I think you may be getting very
>close to saturating PC600 memory throughput!
Not even close. I use two memory passes. A 512K FFT is 4MB. Two
reads, two writes, plus say 4MB of sin/cos data is 20MB. PC600 memory
can deliver 2.0GB/sec. Thus, 20MB / (2.0GB/sec) is 0.01 seconds.
When will this new code be ready? Patience please :) Above 8K, only
the 64K and 512K FFT have been coded. To do: more optimizing, better
prefetching, the rest of the FFT sizes, commonizing code to reduce
executable size, auxiliary code for ECM and P-1, reduce allocated memory,
find the FFT crossover points, and QA. Then I can take a little while to
add new features. In light of the clear benefits for P4 users, I'll make a
beta
available as soon as possible.
Now the bad news. I implemented my first optimization today. The one I
thought
would get me close to 0.40 seconds. Alas, I grossly overestimated the
benefits.
I'm now at 0.46 and worried about getting to 0.40.
Some have asked about Athlon optimizations. I'm not an expert on the
Athlon CPU. The only change I see to make is a different memory layout to
take advantage of its different cache layout. I suspect a best case
improvement
of 10%. That's a lot of work (for me) for a modest gain. AMD has committed to
implementing SSE2 in a future chip. Then AMD users will also benefit from
this new code.
Regards,
George
_________________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
