Paul.Zimmermann at loria.fr
Wed Mar 6 09:53:05 CET 2013
> We need a forum for both posting patches, and discussing the same
> patches. To me, it seems easiest to have both on the same list.
however maybe not all people on gmp-devel are interested by the detail of all
patches and the corresponding discussions?
Personnally I subscribed to gmp-devel to be informed about new ideas or new
algorithms being implemented in GMP. I expected a low rate of about one
message per day, which was the case in 2012. So far in March we have over 100
messages in less than 6 days! I'm sorry I cannot follow so many messages thus
if this continues I will have to unsubscribe.
Finally to increase the technical content on this list Alexander Kruppa
told me about the following preprint of Ernst Mayer: arxiv.org/abs/1303.0328.
On a Core 2, he claims a mpn_mod_1 running at 6 cycles per limb, and a
mpn_divrem_1 at 12.5 cycles per limb.
On my computer GMP 5.1.1 runs at about 12 and 24 cycles per limb respectively:
tarte% ./speed -s 1000,2000,5000,10000 -c mpn_mod_1.16031552068505892269 mpn_divrem_1.16031552068505892269
overhead 6.13 cycles, precision 10000 units of 5.00e-10 secs, CPU freq 2000.00 MHz
1000 #12682.00 24064.00
2000 #25024.00 48187.00
5000 #62084.00 119298.00
10000 #123786.00 238280.00
tarte% cat /proc/cpuinfo | grep "model name" | head -1
model name : Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz
More information about the gmp-devel