Visible to Intel only — GUID: GUID-2CFC12D9-0F88-4056-AE01-5EAD84F47CDB
Compilation Environment Variables
Hydra Environment Variables
I_MPI_ADJUST Family Environment Variables
Tuning Environment Variables
Process Pinning
GPU Support
NIC Pinning
Environment Variables for Fabrics Control
Environment Variables for Memory Policy Control
Environment Variables for Asynchronous Progress Control
Environment Variables for Multi-EP
Other Environment Variables
Visible to Intel only — GUID: GUID-2CFC12D9-0F88-4056-AE01-5EAD84F47CDB
OFI*-capable Network Fabrics Control
I_MPI_OFI_PROVIDER
Define the name of the OFI provider to load.
Syntax
I_MPI_OFI_PROVIDER=<name>
Arguments
| <name> | The name of the OFI provider to load |
Description
Set this environment variable to define the name of the OFI provider to load. If you do not specify this variable, the OFI library chooses the provider automatically. You can check all available providers by using the I_MPI_OFI_PROVIDER_DUMP environment variable. If you set the wrong name for an available provider, use FI_LOG_LEVEL=debug to get a hint to set the name correctly.
I_MPI_OFI_PROVIDER_DUMP
Control the capability of printing information about all OFI providers and their attributes from an OFI library.
Syntax
I_MPI_OFI_PROVIDER_DUMP=<arg>
Arguments
| <arg> | Binary indicator |
| enable | yes | on | 1 | Print the list of all OFI providers and their attributes from an OFI library |
| disable | no | off | 0 | No action. This is the default value |
Description
Set this environment variable to control the capability of printing information about all OFI providers and their attributes from an OFI library.
I_MPI_OFI_DRECV
Control the capability of the direct receive in the OFI fabric.
Syntax
I_MPI_OFI_DRECV=<arg>
Arguments
| <arg> | Binary indicator |
| enable | yes | on | 1 | Enable direct receive. This is the default value |
| disable | no | off | 0 | Disable direct receive |
Description
Use the direct receive capability to block MPI_Recv calls only. Before using the direct receive capability, ensure that you use it for single-threaded MPI applications and check if you have selected OFI as the network fabric by setting I_MPI_FABRICS=ofi.
I_MPI_OFI_MATCH_COMPLETE
Control the capability of the match complete in the OFI fabric for all providers if applicable.
Syntax
I_MPI_OFI_MATCH_COMPLETE=<arg>
Arguments
| <arg> | Binary indicator |
| enable | yes | on | 1 | Enable match complete if applicable. This is the default value |
| disable | no | off | 0 | Disable match complete |
I_MPI_OFI_LIBRARY_INTERNAL
Control the usage of libfabric* shipped with the Intel® MPI Library.
Syntax
I_MPI_OFI_LIBRARY_INTERNAL=<arg>
Arguments
| <arg> | Binary indicator |
| enable | yes | on | 1 | Use libfabric from the Intel MPI Library |
| disable | no | off | 0 | Do not use libfabric from the Intel MPI Library |
Description
Set this environment variable to disable or enable usage of libfabric from the Intel MPI Library. The variable must be set before sourcing the vars.sh script.
Example
$ export I_MPI_OFI_LIBRARY_INTERNAL=1 $ source <installdir> /env/vars.sh
Setting this variable is equivalent to passing the -ofi_internal option to the vars.sh script.
For more information, refer to the Intel® MPI Library Developer Guide, section Libfabric* Support.
I_MPI_OFI_TAG_DYNAMIC
Enable dynamic tag partitioning.
Syntax
I_MPI_OFI_TAG_DYNAMIC=<arg>
Arguments
| <arg> | Binary indicator |
| enable | yes | on | 1 | Enable automatic OFI tag partitioning |
| disable | no | off | 0 | Use static OFI tag layout. This is the default value |
Description
Set this environment variable to enable dynamic OFI Netmod tag partitioning based on the run configuration. You can use it to get larger MPI tag space or to improve scalability in large-scale runs.
Parent topic: Environment Variables for Fabrics Control