Re: [eigen] assembling a sparse matrix from multiple threads

[ Thread Index | Date Index | More Archives ]


if you can guaranty that each column* is written by a single thread AND that you can predict a reasonable upper-bound for the number of non-zeros in each column*, then you can preallocate using reserve() and the subsequent writes within different columns will be thread safe.

* you can replace "column" by row if you use a row-majpr storage.


On Thu, Jul 5, 2018 at 3:20 PM Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx> wrote:
On 2018-07-05 14:37, Lorenzo Botti wrote:
> The point here is that I know that the code is thread safe, that is no more
> than one thread will try write in position row i col j.
> Accondingly performing the assembly in parallel should not required to lock
> the insertion operation, am I wrong?

That would be true for a dense matrix, but for a sparse matrix that
can't work (in general) since sometimes values are moved or memory gets

To repeat it more clearly: An Eigen::SparseMatrix is _not_ inherently
thread-safe (when values are inserted).


  Dr.-Ing. Christoph Hertzberg

  Besuchsadresse der Nebengeschäftsstelle:
  Robotics Innovation Center
  Robert-Hooke-Straße 5
  28359 Bremen, Germany

  Postadresse der Hauptgeschäftsstelle Standort Bremen:
  Robotics Innovation Center
  Robert-Hooke-Straße 1
  28359 Bremen, Germany

  Tel.:     +49 421 178 45-4021
  Zentrale: +49 421 178 45-0
  E-Mail:   christoph.hertzberg@xxxxxxx

  Weitere Informationen:
  Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
  Firmensitz: Trippstadter Straße 122, D-67663 Kaiserslautern
  Geschaeftsfuehrung: Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster
  (Vorsitzender) Dr. Walter Olthoff
  Vorsitzender des Aufsichtsrats: Prof. Dr. h.c. Hans A. Aukes
  Amtsgericht Kaiserslautern, HRB 2313
  Sitz der Gesellschaft: Kaiserslautern (HRB 2313)
  USt-Id.Nr.:    DE 148646973
  Steuernummer:  19/672/50006

Mail converted by MHonArc 2.6.19+