thofma/Hecke.jl

Is it possible to parallelize class_group(K, grh=false)

Closed this issue · 8 comments

alexjbest commented

I'd like to compute some class groups without GRH, the built in progress estimator says around 20h, can this be parallelized easily?

Also I was quite surprised that class_group uses GRH by default with no indication of this in the doc that I could see

thofma commented

Where would you have liked to see it mentioned?

alexjbest commented

In

help?> class_group(OK)
  class_group(O::NfOrd; bound = -1, method = 3, redo = false, large = 1000) -> GrpAbFinGen, Map

  Returns a group A and a map f from A to the set of ideals of O. The inverse of the map is the projection onto the group of ideals modulo the group of
  principal ideals. redo allows to trigger a re-computation, thus avoiding the cache. bound, when given, is the bound for the factor base.

would be good I think! (I assume that would show up in the online doc then)

thofma commented

OK. I am on it!

I will check whether we can parallelize it easily.

  • 👍1
thofma commented

There are some improvements in the latest version (]up should give version 0.22.4). If you start with julia -j$n, the class group proof will use $n threads. I am not super happy with it, but it might do the trick. One can still enable verbose printing as before. Let me know if it crashes. Try to copy the backtrace if it does.

alexjbest commented

Hi Tommy, this is great thank you! I had to use julia -t$n I think but that seemed to work at least at the start.
After I left it running over the weekend it appears 6 of my 8 threads have died somehow, and are not restarted. I wasn't able to catch the reason. but it seems they failed after 10h on average. The remaining two are still going strong 60h later. I'll start a log and see if I can work out why the others crashed.

alexjbest commented

It also appears that they continue running if I interrupt the original!

alexjbest commented

I saw

[1413135] signal (11.1): Segmentation fault
in expression starting at none:0
fmpz_addmul_ui at /workspace/srcdir/flint2/fmpz/addmul_ui.c:23
unknown function (ip: 0x7ffffffe)
Allocations: 320061411284 (Pool: 320061402222; Big: 9062); GC: 504759
Segmentation fault (core dumped)

when I ended my julia process, I'll restart it now and try and keep a closer eye on it

thofma commented

Hm, seems that it is still not thread safe. I will try and see if I can reproduce it.