Environment Variable
Below a list of variables that can be exported on your environment to activate specific features of BerkeleyGW.
-
export BGW_NUM_THREADS_SORT=4
Set the number of OpenMP threads to be used in the sorting algorithm. -
export BGW_HDF5_WRITE_REDIST=1
In Full-Frequency calculations perform a data redistribution of the frequency epsilon over contiguous chunks of memory to improve HDF5 I/O performance paying an extra communication cost.- Activate using:
1
,T
,t
,Y
,y
- Deactivate using:
0
,F
,f
,N
,n
- Default: non active
- Activate using:
-
export BGW_WFN_HDF5_INDEPENDENT=1
Use MPIO independent instead of collective, depending on the HDF5 implementation and architecture can improve performance.- Activate using:
1
,T
,t
,Y
,y
- Deactivate using:
0
,F
,f
,N
,n
- Default: non active (
H5FD_MPIO_COLLECTIVE
)
- Activate using:
-
export BGW_HDF5_BANDS_COMM_STYLE=1
Allows to change the communication style when reading in blocks of bands fromWFN
files.0
Native HDF5 all ranks read from file1
Using MPI collectives (BCast) (default)2
Using MPI point-to-point (Send/Recv)
-
export BGW_SCALAPACK_MEM_RESCALE=0.5
Rescale the workspace memory in ScalaPACK diagonalization calls by the provided factor. -
export BGW_SCALAPACK_ALGO=1
Allows to select which diagonalization algorithm to use.-1
or0
PDSYEVX1
PDSYEVR2
PZHEEVD
-
export BGW_ELPA_NO_GPU=1
If ELPA is compiled with GPU support enable to switch to the CPU implementation.- Activate using:
1
,T
,t
,Y
,y
- Deactivate using:
0
,F
,f
,N
,n
- Default: non active (GPU version)
- Activate using: