Intel® MPI Library Developer Reference for Linux* OS

Download PDF
ID 768732
Date 6/24/2024
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents
Document Table of Contents x
Intel® MPI Library Developer Reference for Linux* OS
Intel® MPI Library Developer Reference for Linux* OS x
Introduction Command Reference Environment Variable Reference Miscellaneous Notices and Disclaimers
Introduction x
Introducing Intel® MPI Library Notational Conventions Related Information
Command Reference x
Compiler Commands mpirun mpiexec.hydra cpuinfo impi_info mpitune_fast
Compiler Commands x
Compilation Command Options
mpiexec.hydra x
Global Hydra Options Local Hydra Options gtool Options
Environment Variable Reference x
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
Tuning Environment Variables x
Autotuning
Autotuning x
I_MPI_TUNING_AUTO Family Environment Variables
Process Pinning x
Environment Variables for Process Pinning Interoperability with OpenMP* API
GPU Support x
GPU Pinning GPU Buffers Support
Environment Variables for Fabrics Control x
Communication Fabrics Control Shared Memory Control I_MPI_SHM I_MPI_SHM_CELL_FWD_SIZE I_MPI_SHM_CELL_BWD_SIZE I_MPI_SHM_CELL_EXT_SIZE I_MPI_SHM_CELL_FWD_NUM I_MPI_SHM_CELL_BWD_NUM I_MPI_SHM_CELL_EXT_NUM_TOTAL I_MPI_SHM_CELL_FWD_HOLD_NUM I_MPI_SHM_MCDRAM_LIMIT I_MPI_SHM_SEND_SPIN_COUNT I_MPI_SHM_RECV_SPIN_COUNT I_MPI_SHM_FILE_PREFIX_4K I_MPI_SHM_FILE_PREFIX_2M I_MPI_SHM_FILE_PREFIX_1G OFI*-capable Network Fabrics Control
Miscellaneous x
Java* Bindings for MPI-2 Routines
Intel® MPI Library Developer Reference for Linux* OS

Shared Memory Control

I_MPI_SHM

Select a shared memory transport to be used.

Syntax

I_MPI_SHM=<transport>

Arguments

<transport> Define a shared memory transport solution.
disable | no | off | 0 Do not use shared memory transport.
auto Select a shared memory transport solution automatically.
bdw_sse The shared memory transport solution tuned for Intel® microarchitecture code name Broadwell. The SSE4.2. instruction set is used.
bdw_avx2 The shared memory transport solution tuned for Intel® microarchitecture code name Broadwell. The AVX2 instruction set is used.
skx_sse The shared memory transport solution tuned for Intel® Xeon® processors based on Intel® microarchitecture code name Skylake. The CLFLUSHOPT and SSE4.2 instruction set is used.
skx_avx2 The shared memory transport solution tuned for Intel® Xeon® processors based on Intel® microarchitecture code name Skylake. The CLFLUSHOPT and AVX2 instruction set is used.
skx_avx512 The shared memory transport solution tuned for Intel® Xeon® processors based on Intel® microarchitecture code name Skylake. The CLFLUSHOPT and AVX512 instruction set is used.
clx_sse The shared memory transport solution tuned for Intel® Xeon® processors based on Intel® microarchitecture code name Cascade Lake. The CLFLUSHOPT and SSE4.2 instruction set is used.
clx_avx2 The shared memory transport solution tuned for Intel® Xeon® processors based on Intel® microarchitecture code name Cascade Lake. The CLFLUSHOPT and AVX2 instruction set is used.
clx_avx512 The shared memory transport solution tuned for Intel® Xeon® processors based on Intel® microarchitecture code name Cascade Lake. The CLFLUSHOPT and AVX512 instruction set is used.
clx-ap The shared memory transport solution tuned for Intel® Xeon® processors based on Intel® microarchitecture code name Cascade Lake Advanced Performance.
icx The shared memory transport solution tuned for Intel® Xeon® processors based on Intel® microarchitecture code name Ice Lake.

Description

Set this environment variable to select a specific shared memory transport solution.

Automatically selected transports:

  • icx for Intel® Xeon® processors based on Intel® microarchitecture code name Ice Lake
  • clx-ap for Intel® Xeon® processors based on Intel® microarchitecture code name Cascade Lake Advanced Performance
  • bdw_avx2 for Intel® microarchitecture code name Haswell, Broadwell and Skylake
  • skx_avx2 for Intel® Xeon® processors based on Intel® microarchitecture code name Skylake
  • ckx_avx2 for Intel® Xeon® processors based on Intel® microarchitecture code name Cascade Lake
  • knl_mcdram for Intel® microarchitecture code name Knights Landing and Knights Mill
  • bdw_sse for all other platforms

The value of I_MPI_SHM depends on the value of I_MPI_FABRICS as follows: if I_MPI_FABRICS is ofi, I_MPI_SHM is disabled. If I_MPI_FABRICS is shm:ofi,  I_MPI_SHM defaults to auto or takes the specified value.

I_MPI_SHM_CELL_FWD_SIZE

Change the size of a shared memory forward cell.

Syntax

I_MPI_SHM_CELL_FWD_SIZE=<nbytes>

Arguments

<nbytes> The size of a shared memory forward cell in bytes
> 0 The default <nbytes> value depends on the transport used and should normally range from 64K to 1024K.

Description

Forward cells are in-cache message buffer cells used for sending small amounts of data. Lower values are recommended. Set this environment variable to define the size of a forward cell in the shared memory transport.

I_MPI_SHM_CELL_BWD_SIZE

Change the size of a shared memory backward cell.

Syntax

I_MPI_SHM_CELL_BWD_SIZE=<nbytes>

Arguments

<nbytes> The size of a shared memory backward cell in bytes
> 0 The default <nbytes> value depends on the transport used and should normally range from 64K to 1024K.

Description

Backward cells are out-of-cache message buffer cells used for sending large amounts of data. Higher values are recommended. Set this environment variable to define the size of a backwrad cell in the shared memory transport.

I_MPI_SHM_CELL_EXT_SIZE

Change the size of a shared memory extended cell.

Syntax

I_MPI_SHM_CELL_EXT_SIZE=<nbytes>

Arguments

<nbytes> The size of a shared memory extended cell in bytes
> 0 The default <nbytes> value depends on the transport used and should normally range from 64K to 1024K.

Description

Extended cells are used in the imbalanced applications when forward and backward cells are run out. An extended cell does not have a specific owner - it is shared between all ranks on the computing node. Set this environment variable to define the size of an extended cell in the shared memory transport.

I_MPI_SHM_CELL_FWD_NUM

Change the number of forward cells in the shared memory transport (per rank).

Syntax

I_MPI_SHM_CELL_FWD_NUM=<num>

Arguments

<num>  The number of shared memory forward cells
> 0 The default value depends on the transport used and should normally range from 4 to 16.

Description

Set this environment variable to define the number of forward cells in the shared memory transport.

I_MPI_SHM_CELL_BWD_NUM

Change the number of backward cells in the shared memory transport (per rank).

Syntax

I_MPI_SHM_CELL_BWD_NUM=<num>

Arguments

<num> The number of shared memory backward cells
> 0 The default value depends on the transport used and should normally range from 4 to 64.

Description

Set this environment variable to define the number of backward cells in the shared memory transport.

I_MPI_SHM_CELL_EXT_NUM_TOTAL

Change the total number of extended cells in the shared memory transport.

Syntax

I_MPI_SHM_CELL_EXT_NUM_TOTAL=<num>

Arguments

<num> The number of shared memory backward cells
> 0 The default value depends on the transport used and should normally range from 2K to 8K.

Description

Set this environment variable to define the number of extended cells in the shared memory transport.

NOTE:

This is not “per rank” number, it is total number of extended cells on the computing node.

I_MPI_SHM_CELL_FWD_HOLD_NUM

Change the number of hold forward cells in the shared memory transport (per rank).

Syntax

I_MPI_SHM_CELL_FWD_HOLD_NUM=<num>

Arguments

<num> The number of shared memory hold forward cells
> 0 The default value depends on the transport used and must be less than I_MPI_SHM_CELL_FWD_NUM.

Description

Set this environment variable to define the number of forward cells in the shared memory transport a rank can hold at the same time. Recommended values are powers of two in the range between 1 and 8.

I_MPI_SHM_MCDRAM_LIMIT

Change the size of the shared memory bound to the multi-channel DRAM (MCDRAM) (size per rank).

Syntax

I_MPI_SHM_MCDRAM_LIMIT=<nbytes>

Arguments

<nbytes> The size of the shared memory bound to MCDRAM per rank
1048576 This is the default value.

Description

Set this environment variable to define how much MCDRAM memory per rank is allowed for the shared memory transport. This variable takes effect with I_MPI_SHM=knl_mcdram only.

I_MPI_SHM_SEND_SPIN_COUNT

Control the spin count value for the shared memory transport for sending messages.

Syntax

I_MPI_SHM_SEND_SPIN_COUNT=<count>

Arguments

<count> Define the spin count value. A typical value range is between 1 and 1000.

Description

If the recipient ingress buffer is full, the sender may be blocked until this spin count value is reached. It has no effect when sending small messages.

I_MPI_SHM_RECV_SPIN_COUNT

Control the spin count value for the shared memory transport for receiving messages.

Syntax

I_MPI_SHM_RECV_SPIN_COUNT=<count>

Arguments

<count> Define the spin count value. A typical value range is between 1 and 1000000.

Description

If the receive is non-blocking, this spin count is used only for safe reorder of expected and unexpected messages.  It has no effect on receiving small messages.

I_MPI_SHM_FILE_PREFIX_4K

Change the mount point of the 4 KB pages size file system (tmpfs) where the shared memory files are created.

Syntax

I_MPI_SHM_FILE_PREFIX_4K=<path>

Arguments

<path> Define the path to the existed mount point of the 4 KB pages size file system (tmpfs). By default, the path is not set.

Description

Set this environment variable to define a new path to the shared memory files. By default, the shared memory files are created at /dev/shm/.

This variable affects shared memory transport buffers and RMA windows.

Example 

I_MPI_SHM_FILE_PREFIX_4K=/dev/shm/intel/

I_MPI_SHM_FILE_PREFIX_2M

Change the mount point of the 2 MB pages size file system (hugetlbfs) where the shared memory files are created.

Syntax

I_MPI_SHM_FILE_PREFIX_2M=<path>

Arguments

<path> Define the path to the existed mount point of the 2 MB pages size file system (hugetlbfs). By default, the path is not set.

Description

Set this environment variable to enable 2 MB huge pages on the Intel MPI Library.

The variable affects shared memory transport buffers. It may affect RMA windows as well if the windows size is greater than or equal to 2 MB.

Example 

I_MPI_SHM_FILE_PREFIX_2M=/dev/hugepages
NOTE:

The root privileges are required to configure the huge pages subsystem. Contact your system administrator to obtain permission.

I_MPI_SHM_FILE_PREFIX_1G

Change the mount point of the 1 GB pages size file system (hugetlbfs) where the shared memory files are created.

Syntax

I_MPI_SHM_FILE_PREFIX_1G=<path>

Arguments

<path> Define the path to the existed mount point of the 1 GB pages size file system (hugetlbfs). By default, the path is not set.

Description

Set this environment variable to enable 1 GB huge pages on the Intel MPI Library.

The variable affects shared memory transport buffers. It may affect RMA windows as well if the windows size is greater than or equal to 1 GB.

Example 

I_MPI_SHM_FILE_PREFIX_1G=/dev/hugepages1G
NOTE:

The root privileges are required to configure the huge pages subsystem. Contact your system administrator to obtain permission.

Parent topic: Environment Variables for Fabrics Control
Level Two Title