search

ufz / ogs5

OpenGeoSys THM/C simulator version 5
https://www.opengeosys.org/ogs-5
Other
39 stars 97 forks source link

Updates for frontend2 #145

Closed wenqing closed 5 years ago

wenqing commented 5 years ago

as titled.

Besides, changed the name of the benchmark of MPI/hm_tri/hm_tri to MPI/HM/hm.

wenqing commented 5 years ago

@bilke Please take a look. Or you can send a PR for the changes in the modules.

wenqing commented 5 years ago

@bilke I dropped the part to rename envinf1 to frontend.

wenqing commented 5 years ago

@bilke Thanks. I will check the results of the benchmarks that use MKL and LIS.

bilke commented 5 years ago

@wenqing LIS is green, MKL has a failing benchmark.

wenqing commented 5 years ago

@bilke There is a runtime error MKL FATAL ERROR: Cannot load libmkl_lapack.so. libmkl_lapack.so is not included in that precompiled lib. On envinf1, it is in /opt/intel/mkl. I will try to fix this runtime error. If there is no solution for that, I will remove the benchmarks for MKL temporarily,

wenqing commented 5 years ago

MKL benchmarks are disabled in ctest because one MKL library is missed on the frontends of eve. These two benchmarks can only be solved by using a direct solver (like pardiso).

wenqing commented 5 years ago

Deactivated the benchmark of AKS/GAS_FLOW/Gravity/Gravity, which sporadically failed or succeeded in linear solver convergence in the code compiled with gcc7.

The benchmark uses the exponential fluid density model (type 14), which has a bug. With the bugfixing in the exponential fluid density model, the sporadically failure still exists. Valgrind gives no memory leakage in running this benchmark. The benchmark runs good with other density models.

wenqing commented 5 years ago

I am going to merge it soon.