ITensor / ITensorParallel.jl

Parallel tools for ITensors.jl.
MIT License
21 stars 3 forks source link

Serialized collectives #13

Closed b-kloss closed 1 year ago

b-kloss commented 1 year ago

This implements a serialized MPI collective routine _gather to be used in MPISum reduction across nodes via _allreduce. The prior implementation via MPI.Allreduce of the data of an ITensor required same index and block-ordering across nodes, making it failure-prone and is thus removed. Synchronization of linkindices across processes during a DMRG run is taken care of by reimplementation of position! for MPISum. Closes #11 .

codecov-commenter commented 1 year ago

Codecov Report

Merging #13 (18cb966) into main (5dae5b6) will not change coverage. The diff coverage is 0.00%.

@@          Coverage Diff          @@
##            main     #13   +/-   ##
=====================================
  Coverage   0.00%   0.00%           
=====================================
  Files          6       7    +1     
  Lines        178     228   +50     
=====================================
- Misses       178     228   +50     
Impacted Files Coverage Δ
src/mpi_aux.jl 0.00% <0.00%> (ø)
src/mpi_projmposum.jl 0.00% <0.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.