Discussion related to cado-nfs

[Emmanuel Thomé <Emmanuel.Thome@inria.fr>]

On Thu, Oct 29, 2015 at 04:14:16PM -0400, ƦOB COASTN wrote:
> Hello,
> 
> Attached to this message are the results from RSA120 factorization and the
> intel build log for the host binaries.
> Please pardon the lengthy output.
> 
> The baseline RSA120 results for the k1om benchmark are outlined below:
> dual 6core Intel(R) Xeon(R) CPU E52603 && (8) Xeon PHI 5110P
> RSA120: 1.65 hours = >1 day   =   4.53089e+06/5940.38
> 
> If we compare to the baseline for RSA120 on the cado homepage:
> dual 8core Intel(R) Xeon(R) CPU E52650
> RSA120: 43.9 hours = ~2 days  =   ~158040
> Comparison:
> It is unfair to directly compare 10 processors to 2, but here are the
> results anyway.
> 158040 / 5941 = 26.6x Speed-up!
> We could argue that they both exist as one computation node.

right, you may have a point ;-)

> The host binaries for ICC build fails with the message below:
> Scanning dependencies of target las
> [ 37%] Building CXX object sieve/CMakeFiles/las.dir/las.cpp.o
> /home/cado-nfs/sieve/las.cpp(1784): error: argument of type "unsigned char
> *" is incompatible with parameter of type "const char *"
>         _mm_prefetch(((unsigned char *) it)+256, _MM_HINT_NTA);

That should not be hard to fix.

E.