lio-rc failed to run with cuda 8.0 and intel compiler v15

MALBECC / lio

Linear implementation of DFT calculations (CPU and GPU)

GNU General Public License v2.0

26 stars 18 forks source link

lio-rc failed to run with cuda 8.0 and intel compiler v15 #149

Closed librahoo closed 7 years ago

librahoo commented 7 years ago

After compiling successfully for lio (rc version) with cuda 8.0 and intel compiler version 15 by make -j time=1 cpu=0 cuda=1 intel=2 sm52=1 I tried to compile amber16 with -lio, no error reports, however, when I go to directory $AMBERHOME/test/qmmm_EXTERN/QMMM_Lio/water and tried to run the test: Run.water & Run.td.water, the program failed: Error in `../../../../bin/sander': free(): invalid next size (normal): 0x0000000017914130 Aborted (core dumped) possible FAILURE: file mulliken does not exist.

Any suggestions?

fedepedron commented 7 years ago

I believe the problem lies within the fact that many changes have been made to how LIO writes outputs (even input reading has been changed, but we keep backwards-compatibility with AMBER for now) .

Try to run the test in lio/testamber/badamber, changing the "qm_theory='PM3'" for qm_theory='extern'" lines in the *.in files.

-- EDIT --

Do the LIO tests in lio/tests run properly? Try for example 0_aguaGPU or 1_OxymolGPU.

librahoo commented 7 years ago

Dear Federico,

For test in lio/testamber/badamber, after changing the "qm_theory" to "Extern", it works well.

For the lio/test folder, after modifying sm20=1 to sm52=1 ( my GPU card & CUDA support sm52) by executing correr.sh in the subdirectory, the tests run properly for 0_aguaGPU and 1_OxymolGPU.

By the way, I want to try some QM/MM study for enzyme systems as mentioned in this paper: DOI: 10.1021/acs.jpcb.6b10130 Is there any similar tutorials available?

Best, Z.

fedepedron commented 7 years ago

Glad everything ran properly (by the way, I don't believe the flag specifying smXX is necessary, proper architecture should be automatically selected.).

Regarding the study you want to perform, regrettably we do not have tutorials on our side, but it's not different as performing QM/MM in AMBER with any other method. Just take these into account:

In your amber input file, remember to add qm_theory='extern' and the following lines &lio / (this is needed for AMBER to work, it needs the lio namelist although it can be empty)
If you want to delve deeper into LIO options (the defaults should be OK), create in your working directory a file named "lio.in" and containing the same "&lio /" namelist.

We will be writing a LIO manual (well, expanding the existing one) in the near future, but please do ask us if you need a specific setup.