[petsc-dev] PetscHashIJ scaling problem?

[Matthew Knepley]

I think I have tracked down the DMPlex performance issue to a problem with
PetscHashIJKL. I replaced it with PetscHashIJ in 2D, and I still have bad
behavior. I made a simple test and the growth of runtime is definitely
superlinear. The test is ex26 in knepley/fix-hash-scaling. I get

next *$:/PETSc3/petsc/petsc-dev$
/PETSc3/petsc/petsc-dev/arch-c-opencl-next/lib/ex26-obj/ex26 -N 100
-log_summary | grep "(sec):" | cut -c 23-32
6.067e-01
next *$:/PETSc3/petsc/petsc-dev$
/PETSc3/petsc/petsc-dev/arch-c-opencl-next/lib/ex26-obj/ex26 -N 110
-log_summary | grep "(sec):" | cut -c 23-32
1.491e+00
next *$:/PETSc3/petsc/petsc-dev$
/PETSc3/petsc/petsc-dev/arch-c-opencl-next/lib/ex26-obj/ex26 -N 120
-log_summary | grep "(sec):" | cut -c 23-32
2.913e+00
next *$:/PETSc3/petsc/petsc-dev$
/PETSc3/petsc/petsc-dev/arch-c-opencl-next/lib/ex26-obj/ex26 -N 130
-log_summary | grep "(sec):" | cut -c 23-32
4.892e+00
next *$:/PETSc3/petsc/petsc-dev$
/PETSc3/petsc/petsc-dev/arch-c-opencl-next/lib/ex26-obj/ex26 -N 140
-log_summary | grep "(sec):" | cut -c 23-32
6.812e+00
next *$:/PETSc3/petsc/petsc-dev$
/PETSc3/petsc/petsc-dev/arch-c-opencl-next/lib/ex26-obj/ex26 -N 150
-log_summary | grep "(sec):" | cut -c 23-32
9.777e+00

Who put this in?

  Thanks,

     Matt

-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20131108/0200b03a/attachment.html>