search

LBNL-AMO-MCTDHF / V1

LBNL-AMO-MCTDHF v1.36! . . . . .
https://commons.lbl.gov/display/csd/LBNL-AMO-MCTDHF
Other
12 stars 3 forks source link

OpenMP atom and diatom -- ALMOST THERE #24

Closed djhaxton closed 9 years ago

djhaxton commented 9 years ago

In preparation for tag v1.03, I tried getting all the OpenMP I could in the code, and went too far on MCTDH.SRC/COREPROJECT/coreproject.f90, used by both chmctdhf_atom and chmctdhf_diatom.

This was commit

https://github.com/LBNL-AMO-MCTDHF/V1/commit/e8f5cb7463124e943d268e9a512477f4bf938894

These failed OpenMP edits to coreproject.f90 were removed and the old version reinstated in commits

https://github.com/LBNL-AMO-MCTDHF/V1/commit/3001c3bf276db420ee10fa0ef2a1b97683f62446

https://github.com/LBNL-AMO-MCTDHF/V1/commit/f8898a31086c3c8cc82c32c37a843ac9ceccad94

There is a lot of speed to be gained for chmctdhf_atom and chmctdhf_diatom, in coreproject.f90, using OpenMP. My attempt is in

https://github.com/LBNL-AMO-MCTDHF/V1/blob/master/MCTDH.SRC/COREPROJECT/coreproject-OPENMP-attempt-before-v1.03.f90

This version just needs debugging. It's probably almost there! Just a bug or two. I've done most of the work. Go go power rangers.

Dan

djhaxton commented 9 years ago

FYI, "almost there" is a horrible overstatement. The things I attempted in call_twoe_matel, and even below in mult_ke for instance, are incorrect for reasons that I do not understand.

https://github.com/LBNL-AMO-MCTDHF/V1/blob/master/MCTDH.SRC/COREPROJECT/coreproject-OPENMP-attempt-before-v1.03.f90

djhaxton commented 9 years ago

This has dubious value. I am closing this issue, implementation of openmp for atom and diatom, to focus on more important things.