Optimizations for Computing Factorization [cado-nfs]

P__Robert — Fri, 13 Nov 2015 16:17:04 GMT

Hello,

I would like to determine the requirements for factorization of RSA120 using k1om offload:

227010481295437363334259960947493668895875336466084780038173258247009162675779735389791151574049166747880487470296548479

Target application: cado-nfs
Generic C builds possible

https://lists.gforge.inria.fr/pipermail/cado-nfs-discuss/2015-November/000527.html

Required libraries: gmp

Baseline Native Benchmark Complete

Currently, it is possible to build the latest git version of cado-nfs using icc for k1om.

Python3 cross build for k1om also required to execute the boss-worker model and multiple tasks throughout the factor operation.

Edit:

I am in need of some support if anyone is interested in checking out out this experiment.

python3-mic [this is totally butchered] : https://github.com/mancoast/Python-3.3.3 ;

cado-nfs-mic : https://github.com/mancoast/cado-nfs/tree/intel64-dev

The software runs in multiple stages, and some of these stages run in parallel across all resources.

This is done utilizing a Python3 based Boss-Worker model.

Mutliple binaries used for the tasks include:

tasks.polyselect.polyselect1.polyselect.execpath=$SRC/build/mic
tasks.polyselect.polyselect2.polyselect_ropt.execpath=$SRC/build/mic
tasks.sieve.sieving.las.execpath=$SRC/build/mic
tasks.sieve.factorbase.makefb.execpath=$SRC/build/xeon
tasks.sieve.freerel.freerel.execpath=$SRC/build/xeon
tasks.filter.execpath=$SRC/build/xeon
tasks.linalg.execpath=$SRC/build/xeon
tasks.sqrt.execpath=$SRC/build/xeon

The polynomial selection and sieving all occur on the MIC!

The linear algebra and square root tasks require MIC support as they require much time.

The other tasks are small? Any thoughts?

Hello,

P__Robert — Thu, 07 Jan 2016 20:36:00 GMT

Hello,

The system consistently factors 120 decimal digit semi-prime integers in under 2 hours.

EDIT: RSA-140 Trial #1: 18.18 hrs

After increasing the problem size to 155 decimal digit semi-prime integers, I am seeing a lot of red CPU utilization across all MIC's

My micsmc looks like this:

Can someone please explain system utilization versus user utilization?

I need to setup vTune with the cado-nfs source.

Will vTune support the CMAKE build system?

I ran vtune without the source files, and do not understand how to use the results.

End goal is to see what exact lines of C/C++ code are critical to determine what is causing all the system utilization [red].

Thanks!

topic Optimizations for Computing Factorization [cado-nfs] in Software Archive

Optimizations for Computing Factorization [cado-nfs]

Hello,