I just pushed some fixes for the issue in this thread.  You will
typically notice somewhat different performance with multiple blocks per
process because a different partition is used.  In some special cases,
you can reproduce the partition through appropriate use of
-mat_partitioning_type.  For example

  $ mpirun -n 7 ./ex2 -pc_type asm -pc_asm_blocks 7
  Norm of error 0.000458307 iterations 7
  $ mpirun -n 1 ./ex2 -pc_type asm -pc_asm_blocks 7         # parmetis for me
  Norm of error 0.000180024 iterations 8
  $ mpirun -n 1 ./ex2 -pc_type asm -pc_asm_blocks 7 -mat_partitioning_type 
square
  Norm of error 0.000458307 iterations 7

Jed

Reply via email to