rmcdhf -- in-memory versions #79
Conversation
New "in-memory" versions of rmcdhf90 and rmcdhf90_mpi to GRASP by G. Gaigalas and suggested by Charlotte F. Fischer. Qin Yi has collected some information about speed up of this new program: The memory version (10 iterations; -np=32) Block NCSF 1 4499 2 34553 3 76011 4 106505 5 60403 In memory On disk real 26m 34.763s real 275m 14.224s user 796m 28.961s (u/32 = 24.875) user 3366m 36.840s (u/32 = 105.2) sys 41m 32.089s (u/32 = 1.30) sys 3036m 50.741s (s/32 =94.9) So the in memory method is 4-10 times faster than on disk, although this is not a really large case and 32 processors is too many.
|
@mortenpi Great, thanks Morten! Update: alright, sorry for bugging you more - but how do I get the actual build at github to use the correct environment variables? The build of e.g. the new rmcdhf_mem fails for some reason. Did I mess something up with the new Makefiles? They look correct to me and things compile file locally. |
src/appl/rmcdhf90_mem/BUILDCONF.sh
Outdated
| read1_mem.f90 read1_mem_I.f90 | ||
| read2_mem.f90 read2_mem_I.f90 | ||
| read3_mem.f90 read3_mem_I.f90 | ||
| rmcdhf_mem_C.f90 |
There was a problem hiding this comment.
The error seems to be that the compiled rmcdhf_mem_C module is not found: https://github.com/compas/grasp/runs/4871824355?check_suite_focus=true#step:4:2081
I reckon it's because Make tries to compile stuff that depend on this module before actually compiling it. I suggest moving this line all the way to the top, so that it would get compiled first by Make.
As for why it works locally: did you try deleting all the build files? I guess you have the .mod file lying around from a previous compilation.
There was a problem hiding this comment.
Yup, you are right. Not sure how but somehow the .mod files were never cleaned out properly in my tests. Seems to be working now after some shuffling around. And good I learned the procedure of generating/updating the Makefiles with your genbuild scripts!
New version of
RMCDHF(serial and parallel) by G. Gaigalas from a suggestion by Charlotte F. Fischer.I based the changes on the current, newest versions of the codes.
See the commit messages for details and a test run.
I hope I got all the Make and CMake additions correct - seems to work for me, but please test it.
Additional modifications:
CLEVEL*2-->*4make cleannow excludes the bin/ and lib/ directories, a newmake cleanallcall includes these directories.Jon