CPE Release Notes For RHEL 8.7
HPE Cray Programming Environment 24.03 Full Release Notes for RHEL 8.7
=========================================================
HPE Cray Compiling Environment (CCE) 17.0.1
===========================================
Release Date:
-------------
March 2024
Purpose:
--------
CCE 17.0.1 provides Fortran, C, and C++ compilers for
HPE Cray supercomputer and HPE Apollo 2000 Gen10Plus systems.
Key Changes and Support with CCE 17.0.1:
----------------------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
- ROCm 6.0 Suppport
- Added OMPT device tracing support to enable profiling
- Tuned support for MI300A
- SLES gcc/13 support
- Address sanitizer support on GPUs in Fortran
-----------
CCE 17.0.0
- LLVM 17 base (Merges up to November 6th 2023)
- Fortran
- Support offload of DO CONCURRENT iterations to accelerators with unified shared memory
- Starting with 17.0, support passing any Fortran contained procedures to C/C++
- Code with both 17.0 and pre-17.0 objects files must be linked with 17.0 ftn
- Added the following Fortran 2023 Features:
- US 01 & 02 Allow much longer statement lines and overall statement length
- US 03 SPLIT and TOKENIZE intrinsics
- US 04 Trig functions that work in degrees
- US 06 selected_logical_kind intrinsic function
- US 07 Additional named constants to specify kinds in ISO_FORTRAN_ENV
- US 10 The at edit descriptor
- US 11 Control over leading zeros in output of real values
- US 15 Simple procedures
- US 20 Reduction specifier for do concurrent
- UK 01 Extend the intrinsic procedure c_f_pointer to allow its pointer result to have specified lower bounds
- A public namelist group can have a private namelist group object.
- Added the following OpenMP 5.1 Features:
- 'error' directive
- 'omp_get_mapped_ptr' routine
- 'omp_display_env' routine
- 'thread_limit' clause on 'target' construct (C/C++ only)
- New initial device number value
- loop construct
- Augmented OpenMP 5.X 'declare mapper' to support all Fortran derived types
- Updated GPU parallelism mapping policy to address performance concerns in some user applications
- Updated GPU unified shared memory software architecture to take advantage of supported hardware
- Full tuned support for Nvidia Grace Hopper
- Initial support for AMD MI300
- Experimental support for Sanitizers on AMD GPUs (C & C++ Only)
User and Application Impact:
----------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
-----------
CCE-17.0.0
- For SLES/COS based systems, the linking GCC library has been changed to use the OS Distributed version available through SLES channels
- For SLES/COS based systems, shared libraries built by prior CCE versions must be rebuilt with CCE 17
- For RHEL based systems, we continue to use gcc-toolset provided by Red Hat, updating to gcc-toolset-12 as our baseline
- For RHEL based systems, shared libraries built by prior CCE versions do not need to be rebuilt
Issues or Bugs Resolved:
------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
CAST-34955 CCE17 Regression with OpenMP Offload and Fortran Abstract Type
CAST-34862 fortran openmp default map(first private:scalar) disables reduction function
CAST-34547 roc-obj doesn't work with executable built with cce
CAST-34528 fortran arrays dimensioned 0 with omp target offload -runtime error
CAST-34517 Cray compiler segfaults on ARCHER2 when default IPA enabled
CAST-33381 ONETEP reaises a CCE internal compiler error (segfault)
CAST-31166 Cray Fortran compiler bug - Finalization
-----------
CCE-17.0.0
CAST-34375 ftn openmp target offload gets Internal Compiler Error
CAST-34321 GPU-RDC flag breaks ROCm/5.7.0 + CCE/16.0.1 interoperability
CAST-34267 Cray Clang 16 compiler crash when on Kokkos sources
CAST-34235 internal functions of elemental function are not compiling
CAST-34208 internal subprogram to ELEMENTAL subprogram
CAST-34118 ftn bounds checking -R b causes internal compiler error
CAST-34114 ftn openmp target offload gets Internal Compiler Error
CAST-33882 CrayFTN prevent reproductible compilation
CAST-33799 CCE internal compiler error with miniweather application
CAST-33775 ftn-7206 warning in cce/16.0.0
CAST-33773 INTERNAL COMPILER ERROR with cce 16.0.0
CAST-33711 SEGV from coarray test
CAST-33524 Cray EX - compile error with cce16 and performance issue with cce15.0.1 when using declare target for nested subroutines
CAST-33363 CCE 15.0.1 Very slow to compile lib
CAST-33343 Crayftn Unlimited Polymorphic Assumed Rank Argument
CAST-33342 Crayftn error with pure procedures in submodule
CAST-33324 Fortran with ACC fails on AMD GPU
CAST-33319 incorrect string lengths result after string arrays are initialized with array constructors of different lengths
CAST-33306 `select type` statement fails to identify argument as a `character` type
CAST-32911 needs -O0 to avoid a ftn internal compiler error
CAST-32887 cray-ftn bug with renaming derived type in use statement
CAST-32694 Unified memory does not work with omp declare target directive. Fortran compiler
CAST-32531 ICE using Cray compiler
CAST-32312 CCE Fortran+OpenMP Runtime Error on Crusher (Invalid dope vector)
CAST-31898 Cray Fortran + OpenACC bug
CAST-31821 MFEM: local memory exceeded CCE 15.0.0, ROCm 5.4.0 -O0
CAST-31788 ftn openmp target link directive runtime errors
CAST-31763 Add consistent debug flag behavior across CCE C++ and CCE Fortran compilers
CAST-31636 cray cce compiler OMPT support
CAST-31447 ftn openmp target link directive runtime errors
CAST-31375 Segfault on subroutine exit after omp target exit data clause
CAST-31156 Several problems with Cray CCE compiler version 14.0.3 and GFS weather code
CAST-30542 Cray fortran compiler error on access to module variable for unified memory
CAST-28426 CrayFTN build time issue - demonstrator of pathological behaviour
CAST-22740 Cray Fortran fails to catch bad G format spec
Product and OS Dependencies:
----------------------------
This CCE release is supported on
- HPE Cray supercomputer systems running CSM with COS 2.5.X (SLES15 SP4) or COS 23.11 (SLES15 SP5)
- HPE Cray supercomputer systems running HPCM with COS 2.5.X (SLES15 SP4) or COS 23.11 (SLES15 SP5)
- HPE Cray supercomputer systems running HPCM with RHEL 8.7
- HPE Apollo 2000 Gen10Plus systems running RHEL 8.7
- AMD GPU OpenMP and OpenACC offloading support requires at least ROCm 6.0 (AMD GPU HIP offloading can still use ROCm 5.x)
Notes and Limitations:
----------------------
- The OpenACC type `acc_handle_kind` has been changed to be of size 8 bytes (instead of 4 bytes). This will require recompilation of any codes that use this type.
- Linking Fortran applications may fail with an error message of the form "hidden symbol `<SYMBOL>' in <LIB> is referenced by DSO".
If this error message is seen, it can usually be worked around by adding '-lgcc_s' to your link line. This is known to affect use
of craypat and cray-parallel-netcdf but may be seen without them as well.
- There is a known bug in the CUDA header `cuda_fp16.h` detailed in the upstream issue https://github.com/llvm/llvm-project/issues/60296.
Programs compiling with `-fopenmp` must use CUDA 12.1.1 or later to avoid this bug, or manually change their header as described in the issue.
Documentation:
--------------
- S-5212 Cray Compiling Environment Release Overview (17.0)
- S-3901 Cray Fortran Reference Manual
- S-2179 Cray C and C++ Quick Reference
- Basic man pages: crayftn(1), craycc(1), crayCC(1), intro_openmp()
- Please see https://clang.llvm.org/docs/UsersManual.html or use the -help command line option for
more information on using Clang
Modulefile:
-----------
The following will load the modules necessary to use CCE:
module load PrgEnv-cray
The following will switch to x.y.z version of CCE:
module swap cce cce/x.y.z
Installation instructions:
--------------------------
rpm -ivh cce-17.0.1-202402021627.99c2eadd92a56-4.el8.x86_64.rpm.x86_64.rpm
The following script will set CCE version 17.0.1 default:
/opt/cray/pe/admin-pe/set_default_files/set_default_cce_17.0.1
License:
--------
Except for the third party components and software licensed by HPE through
proprietary agreements, components, files or programs contained within this
package or product are Copyright -2023 Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/cce/17.0.1/ATTRIBUTIONS_17.0.txt
Cray PE DL Plugin with Resiliency Support 22.06.1.2:
Release Date:
-------------
June 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 21.04.1:
o Updated DL framework support
o Introduction of resiliency support
o BFloat16 data type support
Supported Configurations:
-------------------------
Shasta/EX: CPU and AMD GPU support with PrgEnv-gnu
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10.2
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
For info regarding use of the new resiliency feature, see the examples
and README.md in ${CRAYPE_ML_PLUGIN_BASEDIR}/examples/resiliency
Sample Installation instructions:
---------------------------------
Note: Package must be installed or sym linked in the default /opt/cray/pe path.,
ie: use of the --prefix flag for rpm is not supported.
rpm -ivh craype-dl-plugin-ftr-22.06.1.2*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
Shasta/EX:
module load craype-dl-plugin-ftr/22.06.1.2
Certain components, files or programs contained within this package or
product are Copyright 2017-2022 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 21.02.1.3:
Release Date:
-------------
February 2021
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 20.10.1:
o Bug fixes for CUDA 11 support and pip install of include source dist
o Initial CUDA 11.1 support
o Updated DL framework support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC CLE 7: CPU and GPU support
Shasta: CPU-only support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.4
* PyTorch v1.7.1
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-21.02.1.3*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/Shasta:
module load craype-dl-plugin-py3/21.02.1.3
CS:
module load craype-dl-plugin-py3/openmpi/21.02.1.3
module load craype-dl-plugin-py3/mvapich/21.02.1.3
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.09.1:
Release Date:
-------------
September 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.08.1:
o Keras callbacks bug fixes
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.09.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.09.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.12.1:
Release Date:
-------------
December 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.09.1:
o TF v2.9 Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.9
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.12.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.12.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 23.09.1:
Release Date:
-------------
September 2023
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.12.1:
o ROCm 5.3.x support
o TF v2.11 Support
o PT v1.12 Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support for both AMD and NVIDIA
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.11
* PyTorch v1.12
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using TensorFlow, if you see an exception error related to apply_gradients() then
downgrade your keras installation version in your python environment.
pip install keras==2.10.0
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-23.09.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/23.09.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
HPE Cray DL Plugin 24.03.1:
Release Date:
-------------
March 2024
Purpose:
--------
The following changes have been made since HPE Cray DL Plugin 23.09.1:
o AMD MI300A APU support
o TensorFlow v2.15 support
o PyTorch v2.0.1 support
o ROCm 6.0.x support
o Cray-mpich 8.1.27 support
o GCC 12.2.0 support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support for both AMD and NVIDIA
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.15
* PyTorch v2.0.1
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using TensorFlow, if you see an exception error related to apply_gradients() then
downgrade your keras installation version in your python environment.
pip install keras==2.10.0
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-24.03.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/24.03.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
CrayPE Targets 1.13.0
================================================================================
Release Date:
--------------------------------------------------------------------------------
March, 2023
Purpose:
--------------------------------------------------------------------------------
Adds modulefiles for targets - craype-accel-amd-gfx942
Bugs fixed in this release:
--------------------------------------------------------------------------------
None
Dependencies:
--------------------------------------------------------------------------------
The CrayPE Targets 1.13.0 release is dependent on the following
software products:
CrayPE
set_default_3
lmod_scripts
Documentation:
--------------------------------------------------------------------------------
See manpages for cc, CC, ftn, intro_craype-api and pkg-config
See section 2.6 Using Targeting Modules of the Cray Programming Environment
User's Guide (S-2529-116)
See http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh craype-targets-ex-hpcm-1.13.0-202402062025.1d638354624fb-0.ex.hpcm.noarch.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_craype-targets_1.13.0-tcl
/opt/cray/pe/admin-pe/set_default_files/set_default_craype-targets_1.13.0-lua
Product description:
--------------------------------------------------------------------------------
CrayPE Targets contains the modules for CrayPE targets.
TCL Modules - craype-accel-host craype-accel-amd-gfx90a craype-accel-amd-gfx908 craype-accel-amd-gfx940 craype-accel-amd-gfx942 craype-accel-intel-max craype-accel-nvidia70 craype-accel-nvidia80 craype-accel-nvidia90 craype-hugepages2M craype-hugepages4M craype-hugepages8M craype-hugepages16M craype-hugepages32M craype-hugepages64M craype-hugepages128M craype-hugepages256M craype-hugepages512M craype-hugepages1G craype-hugepages2G craype-arm-grace craype-x86-milan craype-x86-milan-x craype-x86-rome craype-x86-spr craype-x86-spr-hbm craype-x86-trento craype-x86-genoa craype-network-none craype-network-ofi craype-network-ucx
Lmod Modules - craype-accel-host craype-accel-amd-gfx90a craype-accel-amd-gfx908 craype-accel-amd-gfx940 craype-accel-amd-gfx942 craype-accel-intel-max craype-accel-nvidia70 craype-accel-nvidia80 craype-accel-nvidia90 craype-hugepages2M craype-hugepages4M craype-hugepages8M craype-hugepages16M craype-hugepages32M craype-hugepages64M craype-hugepages128M craype-hugepages256M craype-hugepages512M craype-hugepages1G craype-hugepages2G craype-arm-grace craype-x86-milan craype-x86-milan-x craype-x86-rome craype-x86-spr craype-x86-spr-hbm craype-x86-trento craype-x86-genoa craype-network-none craype-network-ofi craype-network-ucx
Copyright 2013-2023 Hewlett Packard Enterprise Development LP
CDEUX driver (craype) 2.7.31.11 - 11
==============
Release Date:
--------------
March 2024
Purpose:
--------
The Cray Development Environment User eXperience driver (CDEUX) (craype)
provides a common set of drivers for all CPE programming environments,
notably the C/C++ (cc/CC) and Fortran (ftn) languages.
New CDEUX driver (craype) features for HPE Cray EX and Apollo systems:
- Support for CCE using distribution GCC toolchain installation.
- Improved support for AMD ROCm installations.
- Improved AARCH64 support.
Key Changes and Bugs Closed:
----------------------------
- CDEUX driver (craype) 2.7.31.11 has fixed issues regarding OpenMP
Product and OS Dependencies:
----------------------------
The CDEUX driver (craype) 2.7.31.11 release is supported on the following
HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Documentation:
---------------
See manpages for cc, CC, ftn, and intro_craype-api.
See the pkg-config manpage or
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Product description:
--------------------
CrayPE contains drivers, cc, CC, and ftn to compile for all CPE
Programming Environments.
Certain components, files or programs contained within this package or product
are Copyright 2013-2023 Hewlett Packard Enterprise Development LP.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright 2023 Hewlett Packard
Enterprise Development LP.
Craypkg-gen 1.3.32
==============
Release date
------------
February 2024
Purpose
-------
Bug fix release.
Bugs fixed in this release
--------------------------
- Preserve build numbers in ROCm versions
- Exclude hip paths in the ROCm pkg-config files for ROCm >= 5.7.0
- Change -D_HIP_PLATFORM_HCC_ to -D__HIP_PLATFORM_AMD__ in pkg-config files for ROCm >= 5.7.0
- Utilize Jinja2 templating for ROCm pkg-config file
- Update logic to generate modulefiles for new file tree in Intel OneAPI 2024.0
- Update the Nvidia HPC SDK compiler compatibility version for aarch64
- Remove non-existent CUPTI path from Nvidia HPC SDK modulefile
Dependencies
------------
The craypkg-gen 1.3.32 release is supported on the following HPE
Cray systems:
- Cray EX systems SLE 15.0 or later and RHEL 8.0 or later.
Driver support for integrating Third Party C, C++, and Fortran libraries
through .pc files using pkg-config is used in CrayPE 2.x and later.
Limitation
----------
- Library dependencies for static libraries are not added to the .pc files
for keywords Requires.private and Libs.private. A warning is issued by
the craypkg-gen tool to advise users to add this information to the
libraries .pc files.
- RPM limits packages to 4GB
Documentation
-------------
Man pages for craypkg-gen are found by executing `module load craypkg-gen`
and then `man craypkg-gen`. See
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Examples for creating modulefiles for Intel, PGI and Python are included in
the craypkg-gen ‘doc’ directory:
/opt/cray/craypkg-gen/1.3.32/doc/intel_example.txt
/opt/cray/craypkg-gen/1.3.32/doc/pgi_example.txt
/opt/cray/craypkg-gen/1.3.32/doc/python_example.txt
Example:
- As an example, the Intel 16.0.3.210 compiler was recently released.
After installing the compiler the administrator creates a modulefile for
this release by executing the following commands:
# module load craypkg-gen
# craypkg-gen -m /opt/intel/compilers_and_libraries_2016.3.210
This version of the Intel compiler is made default by executing the
command:
# /opt/admin-pe/set_default_craypkg/set_default_intel_16.0.3.210
Installation instructions
-------------------------
rpm -ivh craypkg-gen-1.3.32-1.3.32-1.el8.noarch.rpm
To change the product version to default after installation:
/opt/cray/pe/admin-pe/set_default_files/set_default_craypkg-gen_1.3.32
Product description
-------------------
The craypkg-gen 1.3.32 utility provides the system administrator
a tool to integrate third party software with the Cray software stack.
Craypkg-gen assists with integration by creating .pc files for C, C++, and
Fortran libraries, pkg-config enabled modulefiles and RPMs.
The workflow for using craypkg-gen is
1) Build the open source software
2) Create .pc files for libraries
3) Create pkg-config enabled modulefiles
4) Customize pkg-config and modulefiles if needed
Certain components, files or programs contained within this package or product
are Copyright 2013-2024 Hewlett Packard Enterprise Development LP.
Cray DSMML 0.3.0:
==========================
Release Date:
-------------
March, 2024
Product Information:
--------------------
DSMML is a stand-alone memory management library for maintaining distributed
shared symmetric memory heaps for top level PGAS languages and libraries like
Coarray Fortran, UPC, and OpenSHMEM. DSMML allows user libraries to create
multiple symmetric heaps and share information with other libraries. Through
DSMML, we could extract interoperability between PGAS programming models.
Announcements, release informations, supported environments, and backward
compatibility informations about this product can be viewed in the following
location: https://pe-cray.github.io/cray-dsmml/
Purpose:
--------
Cray DSMML 0.3.0 supports the following features:
- Create symmetric heap segments for being used by PGAS languages and
libraries
- Allows options to allocate, deallocate memory buffers from previously
created symmetric heap segments
- Enables sharing information about the symmetric heap segments between
different programming models
- Added support for shared symmetric heap (SSHEAP) that can be used to
provide enhanced SMP data transport in user-libraries like SHMEM, CAF, UPC
The following features are fixed in Cray DSMML 0.3.0 when compared
to Cray DSMML 0.2.2:
- Add support for the library for AARCH64 processor based platforms
- Internal bug fixes involving hugepage cleanups
Refer intro_dsmml(3) manpage for more information on the supported features
and syntax for different routines
Product and OS Dependencies:
----------------------------
The Cray DSMML 0.3.0 release is supported on the following
Cray systems:
* HPE Cray EX systems with CLE
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.3 |
----------------------------+---------------------+
One or more compilers:
* AOCC 2.2 or later
* CCE 9.1 or later
* GNU 9.1 or later
* Intel 19.0 or later
* Nvidia 20.7 or later
Documentation:
--------------
Use the Cray DSMML man pages for more information on the library
functions and use intro_dsmml(3) man page for general information.
Use https://pe-cray.github.io/whitepapers/ for access to different
Cray DSMML specific whitepapers
Announcements, release informations, supported environments, and backward
compatibility informations about this product can be viewed in the following
location: https://pe-cray.github.io/cray-dsmml/
Modulefile:
-----------
module load cray-dsmml/0.3.0
Certain components, files or programs contained within this package or
product have the following Copyright:
Copyright 2018-2021 Hewlett Packard Enterprise Development LP.
FFTW 3.3.10.7
============
Release Date:
-------------
March 2024
Purpose:
--------
This Cray FFTW 3.3.10.7 release is supported on Cray EX (formerly
Shasta) systems. FFTW is supported on the host CPU but not on the
accelerator of Cray systems.
The Cray FFTW 3.3.10.7 release provides the following:
- Compiler and dependency updates
See the Product and OS Dependencies section for details.
Product and OS Dependencies:
----------------------------
The Cray FFTW 3.3.10.7 release is supported on the following Cray systems:
Cray EX Systems with SLES 15 SP4 or later based OS
The FFTW 3.3.10.7 release requires the following software products:
craype 2.7.5 or later
MPT 8.0 or later
One or more of the following major compiler versions:
CCE 17
GCC 12, 13
AOCC 4.1
Intel 2024 or later
Notes and Limitations:
----------------------
Starting with cray-fftw/3.3.6.1 the fftw module has been renamed from fftw
to cray-fftw.
Documentation:
--------------
http://www.fftw.org/index.html#documentation
See the intro_fftw3 manual page for additional information.
Modulefile:
-----------
module load cray-fftw
Installation
------------
To install the rpm, execute:
rpm -ivh cray-fftw-3.3.10.7-202402071520.4d6d346567dfd-1.shasta.el8.x86_64.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_fftw_3.3.10.7
Certain components, files or programs contained within this package or
product are Copyright 2011-2024 Hewlett Packard Enterprise Development LP
# HDF5 1.12.2.11
## Release Date:
* March 2024
## Purpose:
This HDF5 release contains the following:
* Support for compilers on SLES x86_64: CCE 17, Native GCC 12.3.0, Intel-OneAPI 2023.2.0, AOCC 4.2, Nvidia 23.3
* Support for compilers on SLES aarch64: CCE 17, Native GCC 12.3.0, Nvidia 23.11
* Support for compilers on RHEL 8: CCE 17, GCC Toolset 12.2.1, Intel-OneAPI 2023.2.0, AOCC 4.1, Nvidia 23.3
## Product and OS Dependencies:
The HDF5 1.12.2.11 release is supported on the following systems, requiring the following software products:
* CrayPE 2.1.2 or later
* Cray MPT 8.1 or later
SLES x86_64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
SLES aarch64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Nvidia 23.11 or later
RHEL 8
* One or more compilers:
* CCE 17.0 or later
* GCC Toolset 12.2.1 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
## Notes and Limitations:
Users are encouraged to use the Cray compiler scripts (cc, ftn, CC) to include the HDF5 header files and link in the HDF5 libraries, rather than the HDF5 compiler scripts (h5cc, h5fc, h5c++) included in the release.
## Documentation:
http://www.hdfgroup.org/HDF5/doc/index.html
## Modulefile:
`module load cray-hdf5`
OR
`module load cray-hdf5-parallel`
## Product description:
HDF5 is a data model, library, and file format for storing and managing data. It supports an unlimited variety of datatypes, and is designed for flexible and efficient I/O and for high volume and complex data. HDF5 is portable and is extensible, allowing applications to evolve in their use of HDF5. The HDF5 Technology suite includes tools and applications for managing, manipulating, viewing, and analyzing data in the HDF5 format.
## Installation:
`rpm -ivh cray-hdf5-1.12.2.11-*202402200749.acf05c8988d11-3.el8.x86_64.rpm`
The "`*`" in the install command represents compiler version combinations.
To make this the default version, execute:
`/opt/cray/pe/admin-pe/set_default_files/set_default_hdf5_1.12.2.11`
Certain components, files or programs contained within this package or product are Copyright 2019-2024 Hewlett Packard Enterprise Development LP.
Cray LibSci 24.03.0
=====================
Release Date:
-------------
March 2024
Purpose:
--------
Cray LibSci 24.03.0 provides scientific libraries for Cray
HPC systems. Cray LibSci is supported on the host CPU but
not on the accelerator of these systems.
The Cray LibSci 24.03.0 release provides the following:
* support for RHEL gcc-toolset-{10,11,12}
* support for Nvidia HPC SDK 23.11 on aarch64
Cray LibSci 24.03.0 includes the following versions of publicly
available libraries:
* LAPACK 3.10.1 - For further information, see
http://www.netlib.org/lapack
* ScaLAPACK 2.2.0 - (Scalable LAPACK) For further information, see
http://www.netlib.org/scalapack.
* QDWH 2.0.0 KSVD 1.0.0 - Polar decomposition and SVD packages, see
https://github.com/ecrc/qdwh
https://github.com/ecrc/ksvd
Key Changes and Bugs Closed:
----------------------------
Changes in Cray LibSci 24.03.0
* CAST-34956 - Lua module environment variable issue
Changed in Cray LibSci 23.12.5
* CAST-33777 - LibSci mutex contention issue
* Removed support for Intel KNL
Product and OS Dependencies:
----------------------------
The Cray LibSci 24.03.0 release requires the following:
* SLES 15 SP4/SP5 or RHEL 8.8
* cray-mpich/8.1.0 or later
Supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Cray XD systems with CLE
One or more of the following compiler major versions:
* CCE 17.x (SLES)
* CCE 16.x or later (RHEL)
* GCC 12.3 or later (SLES)
* gcc-toolset-10.3/11.2/12.2 (RHEL)
* AOCC 4.1.0 or later
* AMD ROCm 5.0.0 or later
* Intel 2023.2.0 or later
* Nvidia 23.11 or later (aarch64) and 23.3 or later (x86_64)
Notes and Limitations:
----------------------
OMP threaded versions:
The CrayPE 2.1.2 and later releases add support for link line generation
for the multi-threaded versions of the libsci library based on the OpenMP
flags the user specifies for each compiler:
* CCE by default links to the OpenMP LibSci library. CrayPE will link in
the serial version of LibSci when the CCE flag -hnoomp is used.
* GNU by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the GNU flag -fopenmp is used.
* INTEL by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the INTEL flag -qopenmp is used.
Stack size limit:
The 'ulimit -s unlimited' system stack size setting is required for
cray-libsci on all supported Cray platforms. This is typically set as a
system default by admin but may otherwise need to be set at runtime.
Non-default dynamic linking:
When using a non-default version of cray-libsci with dynamic linking
users should set at runtime and at linktime after loading the desired
cray-libsci module--or set the equivalent for their linux shell:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci man page for additional information.
See the csmlversion man page for information to display version information
on the currently loaded scientific libraries.
Modulefile:
-----------
module load cray-libsci
Installation instructions:
--------------------------
LibSci is now packaged into separate compiler specific RPMs to allow
rpmbuild to correctly include compiler dependencies.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
/opt/cray/libsci/24.03.0/ATTRIBUTIONS
Cray LibSci_ACC 24.03.1
=======================
Release Date:
-------------
March 2024
Purpose:
--------
The Cray LibSci_ACC 24.03.1 release provides accelerated versions of
scientific libraries for HPE Cray systems with accelerators.
Product and OS Dependencies:
----------------------------
Cray LibSci_ACC 24.03.1 requires the following platform:
HPE Cray EX system with AMD MI250X
HPE Cray EX system with AMD MI300A
HPE Cray EX system with Nvidia Hopper
Cray LibSci_ACC 24.03.1 supports the following OS:
RHEL 8.8 x86_64
SLES 15 SP4 x86_64
SLES 15 SP5 x86_64 & aarch64
Notes and Limitations:
----------------------
LAPACK workspace calculations from other implementations or hard-coded
values may not be compatible with libsci_acc. Workspaces reported from
a query to the corresponding libsci_acc function should be used.
The current supported usecase is a one-rank to one-gpu mapping for apps.
Performance improvements can be achieved in programs calling libsci_acc
subroutines by using pinned memory. See the intro_libsci_acc man page for
details.
Use of the aprun option "-cc none" or the srun option "--cpu_bind=none"
to disable core affinity is strongly suggested to maintain performance.
When using non-default cray-libsci_acc versions users should set equivalent
at runtime and linktime after loading the desired cray-libsci_acc module:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci_acc man page for additional information.
Modulefile:
-----------
module load cray-libsci_acc
Sample Installation instructions:
---------------------------------
RPM install command:
rpm -ivh cray-libsci-acc-24.03.1-*.##ARCH##.rpm
The "*" in the install command represents compiler version combinations.
To make this the default version:
##PREFIX##/admin-pe/set_default_files/set_default_libsci_acc_24.03.1
License:
--------
Except for the third party modules and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright -2023 Hewlett
Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in
this package are detailed in the file:
##PREFIX##/libsci_acc/24.03.1/ATTRIBUTIONS_libsci_acc_24.03.1.txt
Cray Lmod 8.7.32
================================================================================
Release Date:
--------------------------------------------------------------------------------
March, 2024
Purpose:
--------------------------------------------------------------------------------
Provide a Lmod 8.7.32 package for CPE customers
Bugs fixed in this release:
--------------------------------------------------------------------------------
N/A
Documentation:
--------------------------------------------------------------------------------
See CPE's Installation and User Guides for documentation.
Also see Lmod's official website: https://lmod.readthedocs.io
Dependencies:
--------------------------------------------------------------------------------
The Lmod Scripts 8.7.32 release is dependent on the following
software:
bc
sed
lua >= 5.3
lua-luaposix
lua-luafilesystem
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh cray-lmod-8.7.32-1.el8.x86_64.rpm
To make this the default version, update the symbolic "lmod" link
to the desired Lmod version directory.
Lmod Scripts 3.2.1
================================================================================
Release Date:
--------------------------------------------------------------------------------
December, 2023
Purpose:
--------------------------------------------------------------------------------
Updates package license.
Bugs fixed in this release:
--------------------------------------------------------------------------------
Documentation:
--------------------------------------------------------------------------------
See system Installation and User Guides for documentation
Dependencies:
--------------------------------------------------------------------------------
The Lmod Scripts 3.2.1 release is dependent on the following
software:
set_default_3
Lua
Lmod
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh lmod_scripts-3.2.1-202311141921.212030d47edcb-0.noarch.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_lmod_scripts_3.2.1
Product description:
--------------------------------------------------------------------------------
Contains Lua scripts which support the Cray implementation of Lmod
Copyright 2020-2023 Hewlett Packard Enterprise Development LP
modules 3.2.11.7
==================
Release Date:
-------------
Dec, 2023
Purpose:
--------
- Package dependency update
Documentation:
---------------
See man-pages for module and modulefile.
Cray MPICH 8.1.29:
=======================================
Release Date:
-------------
February 21, 2024
Purpose:
--------
Cray MPICH 8.1.29 is based upon ANL MPICH 3.4a2 with support for libfabric
and is optimized for the Cray Programming Environment.
Major Differences Cray MPICH 8.1.29 from the XC Cray MPICH include:
- Uses the new ANL MPICH CH4 code path and libfabric for network
support.
- Does not support -default64 mode for Fortran
- Does not support C++ language bindings
New Cray MPICH features for HPE Cray EX and Apollo systems:
- Starting from the 8.1.26 release, Cray MPICH supports the Intel Sapphire Rapids CPU HBM
processor architecture.
- On systems with AMD GPUs, Cray MPICH 8.1.26 supports all ROCm
versions starting from ROCm 5.0, including the latest ROCm 5.5.0
release.
The Cray MPICH 8.1.25 release and prior versions of
Cray MPICH are only compatible with ROCm versions up to (and
including) the ROCm 5.4.0 release.
- Cray MPICH uses the libfabric "verbs;ofi_rxm" provider by default.
This is the supported and optimized OFI libfabric provider for
Slingshot-10 and Apollo systems.
- Cray MPICH offers support for multiple NICs per node. Starting with
version 8.0.8, by default Cray MPICH will use all available NICs on
a node. Several rank-to-NIC assignment policies are supported. For
details on choosing a policy for assigning ranks to NICS, or for
selecting a subset of available NICs, please see the following
environment variables documented in the mpi man page.
MPICH_OFI_NIC_VERBOSE
MPICH_OFI_NIC_POLICY
MPICH_OFI_NIC_MAPPING
MPICH_OFI_NUM_NICS
- Enhancements to the MPICH_OFI_NIC_POLICY NUMA mode have been added.
Starting with version 8.0.14, if the user selects the NUMA policy,
the NIC closest to the rank is selected. A NIC no longer needs to
reside in the same numa node as the rank. If multiple NICs are
assigned to the same numa node, the local ranks will round-robin
between them. Numa distances are analyzed to select the closest NIC.
- Cray MPICH supports creating a full connection grid during MPI_Init.
By default, OFI connections between ranks are set up on demand. This
allows for optimal performance while minimizing memory requirements.
However, for jobs requiring an all-to-all communication pattern, it
may be beneficial to create all OFI connections in a coordinated
manner at startup. See the MPICH_OFI_STARTUP_CONNECT description in
the mpi man page.
- Cray MPICH supports runtime switching to the UCX netmod starting
with version 8.0.14. To do this load the craype-network-ucx module
and module swap between Cray-MPICH and Cray-MPICH-UCX modules. For
more information including relevant environment variables reference
the intro_mpi man page with the Cray-MPICH-UCX module loaded.
- Lmod support for HPE Cray EX starting with Cray MPICH 8.0.16.
Key Changes and Bugs Closed:
----------------------------
Changes in Cray MPICH 8.1.29
New Features:
- Cray MPICH 8.1.29 for aarch64 is in early stage of support and
has some issues. Setting these variables will help circumvent most
known issues.
export MPICH_SMP_SINGLE_COPY_MODE=CMA
export MPICH_MALLOC_FALLBACK=1
Bugs Fixed:
- CAST-34406 - Improve performance in MPI Allreduce benchmark
- CAST-34872 - Fix seg-fault in extremely large allreduce calls
- CAST-34972 - Use dynamic CXI PID to avoid conflicts
- PE-45873 - Added infrastructure for CXI triggered operations non-blocking collectives
- PE-45874 - Add MPI_Ibcast() using CXI triggered operations
- PE-45876 - Add MPI_Igather() using CXI triggered operations
- PE-45877 - Add MPI_Iscatter() using CXI triggered operations
- PE-45878 - Add MPI_Ireduce() using CXI triggered operations
- PE-45879 - Add MPI_Iallreduce() using CXI triggered operations
- PE-49870 - Support MPI_LONG_DOUBLE with CCE for aarch64
- PE-51028 - Improved debugging to diagnose memcpy issues
- PE-51233 - Build libraries for the nvidia compiler on aarch64
- PE-51326 - Fix window creation using MPI_WIN_FLAVOR_SHARED to avoid running out of fi_mr resources
- PE-51365 - Fix bad initialization of hint striping_unit in MPI-IO
- PE-51445 - Support ROCM 6.0 with Cray-MPICH GTL
- PE-51470 - Update Dynamic Process Management documentation
- PE-51524 - Document PBS options for using Spawn on Slingshot
- PE-51738 - Bugfix for GPU NUMA nodes on MI300 GPUs
Product and OS Dependencies:
----------------------------
The Cray MPICH 8.1.29 release is supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.6 |
----------------------------+---------------------+
cray-pals | >= 1.0.6 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
One or more compilers:
* AMD ROCM 5.0 or later
* AOCC 4.1 or later
* CCE 16.0 or later
* GNU 10.3 or later
* Intel 2022.1 or later
* Nvidia 23.3 or later
Notes and Limitations:
----------------------
Limitations in Cray MPICH 8.1.29:
- Cray MPICH 8.1.29 can support only ~2040 simultaneous MPI
communicators. This limit is less the XC Cray MPICH limit of
~4090 simultaneous communicators. Cray intends to raise the
limit in a future release of Cray MPICH for Shasta to at least
the XC limit.
Documentation:
--------------
For more information see the intro_mpi man page.
Modulefile:
-----------
module load cray-mpich/8.1.29
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright -2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/mpich/8.1.29/ATTRIBUTIONS
Copyright -2023 Hewlett Packard Enterprise Development LP
# NetCDF 4.9.0.11
## Release Date:
* March 2024
## Purpose:
This NetCDF release contains the following:
* Support for compilers on SLES x86_64: CCE 17, Native GCC 12.3.0, Intel-OneAPI 2023.2.0, AOCC 4.2, Nvidia 23.3
* Support for compilers on SLES aarch64: CCE 17, Native GCC 12.3.0, Nvidia 23.11
* Support for compilers on RHEL 8: CCE 17, GCC Toolset 12.2.1, Intel-OneAPI 2023.2.0, AOCC 4.1, Nvidia 23.3
## Product and OS Dependencies:
The NetCDF 4.9.0.11 release is supported on the following systems, requiring the following software products:
* Cray HDF5 1.12.2.*
* CrayPE 2.1.2 or later
Cray MPT 8.1 or later
SLES x86_64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
SLES aarch64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Nvidia 23.11 or later
RHEL 8
* One or more compilers:
* CCE 17.0 or later
* GCC Toolset 12.2.1 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
## Notes and Limitations:
For certain F77 source files, when compiling with `gcc/10+`, it may be necessary to configure/compile with:
`export FCFLAGS="-fallow-argument-mismatch"`
`export FFLAGS="-fallow-argument-mismatch"`
This will turn mismatch errors between actual and dummy argument lists to warnings. See:
https://github.com/Unidata/netcdf-fortran/issues/212
https://gcc.gnu.org/gcc-10/changes.html
Unidata now packages Netcdf-4 and legacy Netcdf-3 separately. HPE has decided not to continue supplying the legacy Netcdf-3 package. Due to CCE changes a version of netcdf built with "-sreal64" is neither needed nor provided.
NetCDF is supported on the host CPU but not on the accelerator.
## Documentation:
http://www.unidata.ucar.edu/software/netcdf/docs
## Modulefile:
`module load cray-netcdf`
OR
`module load cray-netcdf-hdf5parallel`
## Product Description:
NetCDF (network Common Data Form) is a set of interfaces for array-oriented data access and a freely-distributed collection of data access libraries for C, Fortran, C++, Java, and other languages. The netCDF libraries support a machine-independent format for representing scientific data. Together, the interfaces, libraries, and format support the creation, access, and sharing of scientific data.
## Installation:
`rpm -ivh cray-netcdf-4.9.0.11-*202402202022.484a8086ac4a4-2.el8.x86_64.rpm`
The "*" in the install command represents compiler version combinations.
To make this the default version, execute:
`/opt/cray/pe/admin-pe/set_default_files/set_default_netcdf_4.9.0.11`
Certain components, files or programs contained within this package or product are Copyright 2019-2024 Hewlett Packard Enterprise Development LP.
Cray PALS 1.3.2:
================
Release Date:
-------------
Nov 2023
Purpose:
--------
Cray Parallel Application Launch Service (PALS) is an application launcher for
Cray PE applications.
Key Changes:
------------
Add host2xname tool
Allow configuring the PMI port range
Add more examples in manpages
Add SLES 15 SP5 build
Add SLES 15 SP4 and SP5 aarch64 build
Add PMI and PMI2 as --pmi options for mpiexec
Periodically ping applications
Add the ability to allocate additional VNIs for special services like DAOS
Add mpiexec --disable-rdzv-get option to disable rendezvous gets in the NICs
Enable --cpu-bind and --gpu-bind=verbose to work with --rankfile
Add PALSD_FANOUT env var for palsd to set fanout tree width
Add ability for simultaneous mpiexecs to divide up PBS job slots
Add RHEL 8.8 build
Add support for AMD & NVIDIA GPU visibility
Change the wire format for PMIx data to make PMIx_Fence more efficient
Increase core file size limit on RHEL systems
Add retries to SPIRE auth token call
Install aprun and mpirun as symlinks
Use position-independent executables
Change F_CXI_COLL_* -> FI_CXI_COLL_* environment variables
Configure Instant On URL on Shasta
Move binaries and libraries to /opt/cray
Change pmi_example to print NIC distances for all ranks
Make PALS command's nodelists consistent
Set ATOM_SOCKET by default
Improve vnid request error reporting
Launch with PMIx support automatically fetch NIC information
Reduce default fanout to 32 nodes
Require PMIx version 4.2.4 or newer
Send the job-step name to vnid when creating a job-step VNI
Include pid in palsd log messages
Raise maximum number of ZMQ sockets to 4096
Munge auth no longer used for some internal RPCs with trusted connections
Increase default send timeout to 8s
Fix errors found with clang scan-build
Fix nohup mpiexec behavior
Improve cleanup when launch RPC fails
Implement the verbose option for GPU binding
Fix support for the cycle keyword with the mpiexec --rankfile option
Fix error checking for GPU binding options
Better handle mpiexec disconnects
Fix segfault on PMIx RPC for missing app
Fix spawn-related memory leaks
Fix misleading warning in line buffer mode
Remove SLES 15 SP3 build
Deprecate PALS_FANOUT and APRUN_FANOUT environment variables
Install cray-pals RPM
Documentation:
--------------
For more information see aprun, mpiexec, palscmd, palsctrl, palsig, and palstat man page.
Modulefile:
-----------
module load cray-pals
Installation instructions:
--------------------------
rpm -ihv cray-pals-<version>-<date>.el8.x86_64.rpm
rpm -ihv cray-palsd-<version>-<date>.el8.x86_64.rpm
Copyright 2022-2023 Hewlett Packard Enterprise Development LP
cray-papi
=================
Release date:
-------------
December 2023
Package:
--------
cray-papi-7.0.1.2-202310131612.70757f986edac-4 x86_64
Purpose:
--------
New version of papi 7.0.1.2 release 4
Documentation:
--------------
Overview: https://github.com/icl-utk-edu/papi
Web Site: https://icl.utk.edu/papi
Product description:
--------------------
PAPI aims to provide the tool designer and application engineer with a
consistent interface and methodology for use of the performance counter
hardware found in most major microprocessors. PAPI enables software
engineers to see, in near real time, the relation between software
performance and processor events.
Dependencies:
-------------
For a list of software used when validating this version of
PAPI on Cray and HPE systems, see the HPE Cray Programming
Environment release announcements.
Copyright 2015-2017,2019-2023 Hewlett Packard Enterprise Development LP
# Parallel NetCDF 1.12.3.11
## Release Date:
* March 2024
## Purpose:
This Parallel NetCDF release contains the following:
* Support for compilers on SLES x86_64: CCE 17, Native GCC 12.3.0, Intel-OneAPI 2023.2.0, AOCC 4.2, Nvidia 23.3
* Support for compilers on SLES aarch64: CCE 17, Native GCC 12.3.0, Nvidia 23.11
* Support for compilers on RHEL 8: CCE 17, GCC Toolset 12.2.1, Intel-OneAPI 2023.2.0, AOCC 4.1, Nvidia 23.3
See the Parallel NetCDF documentation for details and usage information.
## Product and OS Dependencies:
This Parallel NetCDF 1.12.3.11 release is supported on on the following systems, requiring the following software products:
* CrayPE 2.1.2 or later
* Cray MPT 8.1 or later
SLES x86_64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
SLES aarch64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Nvidia 23.11 or later
RHEL 8
* One or more compilers:
* CCE 17.0 or later
* GCC Toolset 12.2.1 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
## Notes and Limitations:
For certain F77 source files, when compiling with `gcc/10+`, it may be necessary to configure/compile with:
`export FCFLAGS="-fallow-argument-mismatch"`
`export FFLAGS="-fallow-argument-mismatch"`
This will turn mismatch errors between actual and dummy argument lists to warnings. See:
https://github.com/Unidata/netcdf-fortran/issues/212
https://gcc.gnu.org/gcc-10/changes.html
## Documentation:
https://parallel-netcdf.github.io/wiki/Documentation.html
## Modulefile:
`module load cray-parallel-netcdf`
## Product Description:
Parallel NetCDF is a library providing high-performance I/O while maintaining file-format compatibility with Unidata's NetCDF.
## Installation:
`rpm -ivh cray-parallel-netcdf-1.12.3.11-*202402200749.41338063d7353-1.el8.x86_64.rpm`
The "`*`" in the install command represents compiler version combinations.
To make this the default version, execute:
`/opt/cray/pe/admin-pe/set_default_files/set_default_parallel-netcdf_1.12.3.11`
Certain components, files or programs contained within this package or product are Copyright 2019-2024 Hewlett Packard Enterprise Development LP.
Perftools 23.12.0
===============
Release Date: December, 2023
Purpose:
========
This is a feature and bugfix release for the following systems:
- HPE Cray EX and HPE Cray Supercomputer Systems with HPCM
- HPE Cray EX and HPE Cray Supercomputer Systems with CSM
- HPE Apollo 2000 Gen 10 Plus Systems (x86)
Key enhancements or changes from the previous release:
======================================================
o Upgrade to DWARF 0.8.0
o Upgrade to PAPI 7.0.1.2
- Upstream bugfixes
- Upstream CUDA component version locked from prior release and renamed cuda_7011
- Upstream ROCm component version locked from release 7.0.0.1 and renamed rocm_7001
- Support for Nvidia Grace processor
o Add support for ROCm 5.7.0
o Add support for CUDA 12.2
o Add support for Nvidia H100 GPUs
o Update ROCm trace wrappers to support up to ROCm 5.7
o Add support for PAT_region_push/pop in full trace mode
o Add support for profiling Python 3.11
o Add a DSMML tracegroup
o Improved counter groups and derived metrics for AMD Genoa processors
o Add support for OMPT target callbacks with CCE's OpenMP runtime, enabled
via the CRAYPAT_OMP_TOOL environmnet variable (NOTE: this is primarily intented
for use with pat_run. When instrumenting executables with CCE, proprietary
tracepoints are used in favor of OMPT. To use OMPT instead, '-fno-instrument-openmp'
must be added to compilation.)
o Perftools and its libraries are now compiled using the OS native
gcc and no longer depend on cray-gcc
o Set PAT_RT_ACC_MODEL_NAME in the runtime environment to manually specify
the accelerator model name.
Bugs Fixed:
===========
o Unexpected callers appearing in the calltree when using perftools-lite-loops
o An inappropriate assertion failure when MPI_Get_count returns MPI_UNDEFINED
o A runtime failure when incompatible versions of HIP and pat_run are used
o An issue preventing lambda and template function resolution using Perftools
with Caliper
o Segfaults when managing a hipStream with singletons
o Segfaults when tracing OpenACC codes
o Program hangs at dlopen using pat_run on certain Python codes
o Program hangs using the '-gheap' trace option on RHEL systems
o Incorrect number of PEs/Node reported when using PBS and PALS
o Incorrect launching of Apprentice2 when using reveal launcher
Notes:
======
o This release of Perftools only supports ROCm 5.7.0; SLES 15 support is limited to SP5
o This release of Perftools only supports CUDA 12.0 and later
Known Issues:
=============
o Fortran applications compiled against CCE 16 may fail during instrumentation
with the following error message: "hidden symbol `<SYMBOL>' in <LIB> is
referenced by DSO". To work around this error specify the following during
the pat_build invocation '-Dlink-instr=-lgcc_s'.
o When using OMPT with CCE's openmp runtime, to and from device transfers and
kernel executions originating from a single construct are treated as separate
events. For example, 'target map(tofrom:)' will accumulate 3 events each
time it is encountered.
o Nvidia device drivers do not correctly populate system files used by Perftools
for the Grace-Hopper superchip, impacting driver versions up to: 545.23.06.
Reports generated from applications run on these devices may list the accelerator
device as "Unknown". Additionally, usage of accelerator performance counter groups
is not supported when the device is unresolved. To work around this issue set the
following runtime environment variable prior to running the experiment
'PAT_RT_ACC_MODEL_NAME="GH200 120GB"'.
o Tracing experiments with accelerator performance counters and all sampling experiments
may produce unexpected behavior on Grace-Hopper superchips. At this time, there is
no work around this issue and the capabilities should be treated as experimental.
o Sampling experiments on aarch64 systems lead to fatal errors if
PAT_RT_CALLSTACK_MODE="unwind" is used. The setting is disabled for these situations
and PAT_RT_CALLSTACK_MODE="frames" is used instead.
o The Intel OneAPI Fortran compiler does not correctly produce the ELF .debug_pubnames or
.debug_names section in object files when the "-debug pubnames" option is specified
(as required when the perftools module is loaded). This prevents pat_build from collecting
the necessary information that supports tracing user-defined entry points. Specify the
"-Ddebug-names=n" option to pat_build so pat_build can collect the information from the
ELF .debug_info section.
o The Intel OneAPI Fortran compiler does not accurately represent formal parameters to a
subprogram in some cases. This results in the generation of incorrect intercepts for
user-defined entry points when pat_build -u is specified and results in seg faults when
the instrumented program is executed. Specify the "-Dtrace-force-deref=y" to pat_build
to account for the missing formal parameter information.
Deprecations and Removals:
==========================
o Remove support for PGI programming Environment
o Remove support for following trace groups: craymem, hbw, gni, jemalloc, memkind
Dependencies:
=============
o A PDF reader (such as evince, acroread or okular) is required to use the
app2 online help.
o For a list of software used when validating this version of
Perftools on Cray and HPE systems, see the HPE Cray Programming Environment
release announcements.
Installation instructions:
==========================
For Apollo 80 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 80 (ARM) System (S-8013).
For Apollo 2000 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 2000 Gen10 Plus (x86) System (S-8012).
For HPE Cray EX (Shasta Architecture) systems, refer to the Cray Asynchronous
Installer Guide (S-8003).
Installation of app2 remote client (+ server) on Mac systems:
-------------------------------------------------------------
Apprentice2Installer-23.12.0.dmg
RevealInstaller-23.12.0.dmg
The Cray Apprentice2 and Reveal installers for Mac are included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Mac OS El Capitan through Mojave. Double click on installer
to begin installation. The installer will walk you through the
process for your system.
Installation of app2 remote client (+ server) on Windows 10 systems:
--------------------------------------------------------------------
Apprentice2Installer-23.12.0.exe
The Cray Apprentice2 installer for Windows is included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Windows 10. Double click on installer to begin installation.
The installer will walk you through the process for your system.
Documentation:
==============
See the following man pages:
app2, grid_order, intro_craypat, pat_build, pat_help, pat_info, pat_opts,
pat_report, pat_run, pat_view, reveal
intro_papi
perftools-base, perftools-lite, perftools-preload
accpc, cray_cassini, cray_pm, hwpc, nwpc, papi_counters, uncore
PAPIlicnotices
Search for perftools on https://support.hpe.com to access documentation
on Cray Performance Measurement and Analysis Tools
License:
========
Except for the third party modules discussed below and software licensed
by HPE through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
$CRAYPAT_ROOT/ATTRIBUTIONS_perftools.txt
Cray PMI 6.1.14
==============================================================================
Release Date:
--------------
February 12, 2024
Purpose:
--------
The Cray Process Manager Interface Library provides the interface between the
application launcher and other communication libraries such as MPICH and SHMEM.
Cray PMI 6.1.14 is optimized for the Cray Programming Environment.
New Cray PMI features for HPE Cray EX and Apollo systems:
Key Changes and Bugs Closed:
----------------------------
Starting with Cray-PMI 6.1.0 the cray-pmi and cray-pmi-lib modules have been
combined into a single cray-pmi module. The cray-pmi-lib module will no longer
be installed by new cray-pmi rpms. When the cray-pmi module is loaded, the
pmi libraries will be linked with as-needed flags.
Bug fixes new to Cray PMI 6.1.14 include:
- CAST-34500 - Include unversioned pmi library symlinks in the CPE ld cache
- PE-51739 - Create modern PMI process map code
Product and OS Dependencies:
----------------------------
The Cray PMI 6.1.14 release is supported on the following systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product Dependencies:
----------------------------------------+
cray-pals | >= 1.0.6 |
-----------------------+----------------+
slurm | >= 20.02 |
-----------------------+----------------+
Notes and Limitations:
----------------------
Documentation:
--------------
For more information see the intro_pmi man page.
Modulefile:
-----------
module load cray-pmi/6.1.14
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright 2020-2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/pmi/6.1.14/ATTRIBUTIONS
cray-python 3.11.7
====================
Release Date
------------
March 2024
Purpose
-------
Cray Python is an implementation of the Python programming language for the Cray
Programming Environment. The numpy and scipy modules are configured to call Cray
Libsci routines. The mpi4py module is configured to call Cray MPICH routines.
Cray Python is designed to run Python codes on the compute nodes of an HPE
supercomputer. HPE does not make changes to the Python source or any of its
libraries nor does it plan to make changes in future releases.
The cray-python 3.11.7 release contains
- python-3.11.7
- numpy-1.24.4
- scipy-1.10.1
- mpi4py-3.1.4
- dask-2023.6.1
Product and OS Dependencies
---------------------------
The cray-python 3.11.7 release is supported on
- Cray EX systems running SLES 15 or RHEL 8
Documentation
-------------
https://www.python.org/doc
Changelog
---------
https://docs.python.org/release/3.11.7/whatsnew/changelog.html
Modulefile
----------
module load cray-python/3.11.7
Installation
------------
rpm -ihv cray-python-3.11.7-20240208204348.1cdbf56-1.el8.x86_64.rpm
To make this the default version, execute
[PREFIX]/admin-pe/set_default_files/set_default_python_3.11.7
Certain components, files, or programs contained within this package are
© Copyright 2021-2024 Hewlett Packard Enterprise Development LP
cray-R 4.3.2
============
Release Date
------------
March 2024
Purpose
-------
The cray-R 4.3.2 release.
cray-R is configured to call cray-libsci routines.
Product and OS Dependencies
---------------------------
The cray-R 4.3.2 release is supported on
- Cray EX systems running SLES 15 or RHEL 8
Documentation
-------------
https://cran.r-project.org/manuals.html
Modulefile
----------
module load cray-R/4.3.2
Installation
------------
rpm -ihv cray-R-4.3.2-20240208204434.1115d98-1.el8.x86_64.rpm
To make this the default version, execute:
[PREFIX]/admin-pe/set_default_files/set_default_R_4.3.2
Certain components, files or programs contained within this package or product are
Copyright 2018-2024 Hewlett Packard Enterprise Development LP
Cray OpenSHMEMX 11.7.1:
===============================
Release Date:
-------------
Mar, 2024
Purpose:
--------
OpenSHMEM is a Partitioned Global Address Space (PGAS) library interface
specification, which is the culmination of a standardization effort among
many implementers and users of SHMEM programming model. SHMEM has a long
history as a parallel programming model on HPE Cray supercomputer systems.
For the past two decades SHMEM library implementation in HPE Cray systems
evolved through different generations. Cray OpenSHMEMX is a proprietary,
OpenSHMEM specification compliant SHMEM implementation for HPE Cray
product line.
Key Changes:
------------
Major differences in Cray OpenSHMEMX 11.7.1 from
Cray OpenSHMEMX version 11.7.0 includes the following:
- Add support for logging HPE Slingshot NIC counter values
- Add support for using the library on platforms with AARCH64
processors
- Improve performance of shmem_quiet memory ordering operation
- Improve performance for allreduce collective operations when all
PEs are involved in the communication operations
Bugs Closed:
------------
The following bugs are fixed as part of the Cray OpenSHMEMX 11.7.1
release:
- Fix C11 generic APIs for shmem_sync operations
- Fix library usage with GNU and Intel compilers
Product and OS Dependencies:
----------------------------
The Cray OpenSHMEMX 11.7.1 release is supported on the following
Cray systems:
* HPE Cray EX systems with CLE
Product and OS Dependencies by network type:
--------------------------------------------------+
| HPE Slingshot |
----------------------------+---------------------+
craype | >= 2.7.3 |
----------------------------+---------------------+
cray-pals | >= 0.3.5 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
cray-dsmml | >= 0.2.2 |
----------------------------+---------------------+
One or more compilers:
* AOCC 2.2 or later
* CCE 9.1 or later
* GNU 9.1 or later
* Intel 19.0 or later
* Nvidia 20.7 or later
Notes and Limitations:
----------------------
Announcements, release information, supported environments, limitations,
usage models, and backward compatibility informations about this product
can be viewed in the following location:
https://pe-cray.github.io/cray-openshmemx/
Documentation:
--------------
Use the Cray OpenSHMEMX man pages for more information on the library
functions and use intro_shmem(3) man page for general information.
Initial support for Cray OpenSHMEMX man pages are derived from OpenSHMEM
standards specification document
Use https://pe-cray.github.io/whitepapers/ for access to different
Cray OpenSHMEMX specific whitepapers
Modulefile:
-----------
module load cray-openshmemx/11.7.1
License:
--------
Components, files, and programs contained within this package or
product are Copyright Hewlett Packard Enterprise Development LP
CPE Release Notes For RHEL 8.8
HPE Cray Programming Environment 24.03 Full Release Notes for RHEL 8.8
=========================================================
HPE Cray Compiling Environment (CCE) 17.0.1
===========================================
Release Date:
-------------
March 2024
Purpose:
--------
CCE 17.0.1 provides Fortran, C, and C++ compilers for
HPE Cray supercomputer and HPE Apollo 2000 Gen10Plus systems.
Key Changes and Support with CCE 17.0.1:
----------------------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
- ROCm 6.0 Suppport
- Added OMPT device tracing support to enable profiling
- Tuned support for MI300A
- SLES gcc/13 support
- Address sanitizer support on GPUs in Fortran
-----------
CCE 17.0.0
- LLVM 17 base (Merges up to November 6th 2023)
- Fortran
- Support offload of DO CONCURRENT iterations to accelerators with unified shared memory
- Starting with 17.0, support passing any Fortran contained procedures to C/C++
- Code with both 17.0 and pre-17.0 objects files must be linked with 17.0 ftn
- Added the following Fortran 2023 Features:
- US 01 & 02 Allow much longer statement lines and overall statement length
- US 03 SPLIT and TOKENIZE intrinsics
- US 04 Trig functions that work in degrees
- US 06 selected_logical_kind intrinsic function
- US 07 Additional named constants to specify kinds in ISO_FORTRAN_ENV
- US 10 The at edit descriptor
- US 11 Control over leading zeros in output of real values
- US 15 Simple procedures
- US 20 Reduction specifier for do concurrent
- UK 01 Extend the intrinsic procedure c_f_pointer to allow its pointer result to have specified lower bounds
- A public namelist group can have a private namelist group object.
- Added the following OpenMP 5.1 Features:
- 'error' directive
- 'omp_get_mapped_ptr' routine
- 'omp_display_env' routine
- 'thread_limit' clause on 'target' construct (C/C++ only)
- New initial device number value
- loop construct
- Augmented OpenMP 5.X 'declare mapper' to support all Fortran derived types
- Updated GPU parallelism mapping policy to address performance concerns in some user applications
- Updated GPU unified shared memory software architecture to take advantage of supported hardware
- Full tuned support for Nvidia Grace Hopper
- Initial support for AMD MI300
- Experimental support for Sanitizers on AMD GPUs (C & C++ Only)
User and Application Impact:
----------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
-----------
CCE-17.0.0
- For SLES/COS based systems, the linking GCC library has been changed to use the OS Distributed version available through SLES channels
- For SLES/COS based systems, shared libraries built by prior CCE versions must be rebuilt with CCE 17
- For RHEL based systems, we continue to use gcc-toolset provided by Red Hat, updating to gcc-toolset-12 as our baseline
- For RHEL based systems, shared libraries built by prior CCE versions do not need to be rebuilt
Issues or Bugs Resolved:
------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
CAST-34955 CCE17 Regression with OpenMP Offload and Fortran Abstract Type
CAST-34862 fortran openmp default map(first private:scalar) disables reduction function
CAST-34547 roc-obj doesn't work with executable built with cce
CAST-34528 fortran arrays dimensioned 0 with omp target offload -runtime error
CAST-34517 Cray compiler segfaults on ARCHER2 when default IPA enabled
CAST-33381 ONETEP reaises a CCE internal compiler error (segfault)
CAST-31166 Cray Fortran compiler bug - Finalization
-----------
CCE-17.0.0
CAST-34375 ftn openmp target offload gets Internal Compiler Error
CAST-34321 GPU-RDC flag breaks ROCm/5.7.0 + CCE/16.0.1 interoperability
CAST-34267 Cray Clang 16 compiler crash when on Kokkos sources
CAST-34235 internal functions of elemental function are not compiling
CAST-34208 internal subprogram to ELEMENTAL subprogram
CAST-34118 ftn bounds checking -R b causes internal compiler error
CAST-34114 ftn openmp target offload gets Internal Compiler Error
CAST-33882 CrayFTN prevent reproductible compilation
CAST-33799 CCE internal compiler error with miniweather application
CAST-33775 ftn-7206 warning in cce/16.0.0
CAST-33773 INTERNAL COMPILER ERROR with cce 16.0.0
CAST-33711 SEGV from coarray test
CAST-33524 Cray EX - compile error with cce16 and performance issue with cce15.0.1 when using declare target for nested subroutines
CAST-33363 CCE 15.0.1 Very slow to compile lib
CAST-33343 Crayftn Unlimited Polymorphic Assumed Rank Argument
CAST-33342 Crayftn error with pure procedures in submodule
CAST-33324 Fortran with ACC fails on AMD GPU
CAST-33319 incorrect string lengths result after string arrays are initialized with array constructors of different lengths
CAST-33306 `select type` statement fails to identify argument as a `character` type
CAST-32911 needs -O0 to avoid a ftn internal compiler error
CAST-32887 cray-ftn bug with renaming derived type in use statement
CAST-32694 Unified memory does not work with omp declare target directive. Fortran compiler
CAST-32531 ICE using Cray compiler
CAST-32312 CCE Fortran+OpenMP Runtime Error on Crusher (Invalid dope vector)
CAST-31898 Cray Fortran + OpenACC bug
CAST-31821 MFEM: local memory exceeded CCE 15.0.0, ROCm 5.4.0 -O0
CAST-31788 ftn openmp target link directive runtime errors
CAST-31763 Add consistent debug flag behavior across CCE C++ and CCE Fortran compilers
CAST-31636 cray cce compiler OMPT support
CAST-31447 ftn openmp target link directive runtime errors
CAST-31375 Segfault on subroutine exit after omp target exit data clause
CAST-31156 Several problems with Cray CCE compiler version 14.0.3 and GFS weather code
CAST-30542 Cray fortran compiler error on access to module variable for unified memory
CAST-28426 CrayFTN build time issue - demonstrator of pathological behaviour
CAST-22740 Cray Fortran fails to catch bad G format spec
Product and OS Dependencies:
----------------------------
This CCE release is supported on
- HPE Cray supercomputer systems running CSM with COS 2.5.X (SLES15 SP4) or COS 23.11 (SLES15 SP5)
- HPE Cray supercomputer systems running HPCM with COS 2.5.X (SLES15 SP4) or COS 23.11 (SLES15 SP5)
- HPE Cray supercomputer systems running HPCM with RHEL 8.7
- HPE Apollo 2000 Gen10Plus systems running RHEL 8.7
- AMD GPU OpenMP and OpenACC offloading support requires at least ROCm 6.0 (AMD GPU HIP offloading can still use ROCm 5.x)
Notes and Limitations:
----------------------
- The OpenACC type `acc_handle_kind` has been changed to be of size 8 bytes (instead of 4 bytes). This will require recompilation of any codes that use this type.
- Linking Fortran applications may fail with an error message of the form "hidden symbol `<SYMBOL>' in <LIB> is referenced by DSO".
If this error message is seen, it can usually be worked around by adding '-lgcc_s' to your link line. This is known to affect use
of craypat and cray-parallel-netcdf but may be seen without them as well.
- There is a known bug in the CUDA header `cuda_fp16.h` detailed in the upstream issue https://github.com/llvm/llvm-project/issues/60296.
Programs compiling with `-fopenmp` must use CUDA 12.1.1 or later to avoid this bug, or manually change their header as described in the issue.
Documentation:
--------------
- S-5212 Cray Compiling Environment Release Overview (17.0)
- S-3901 Cray Fortran Reference Manual
- S-2179 Cray C and C++ Quick Reference
- Basic man pages: crayftn(1), craycc(1), crayCC(1), intro_openmp()
- Please see https://clang.llvm.org/docs/UsersManual.html or use the -help command line option for
more information on using Clang
Modulefile:
-----------
The following will load the modules necessary to use CCE:
module load PrgEnv-cray
The following will switch to x.y.z version of CCE:
module swap cce cce/x.y.z
Installation instructions:
--------------------------
rpm -ivh cce-17.0.1-202402021627.99c2eadd92a56-4.el8.x86_64.rpm.x86_64.rpm
The following script will set CCE version 17.0.1 default:
/opt/cray/pe/admin-pe/set_default_files/set_default_cce_17.0.1
License:
--------
Except for the third party components and software licensed by HPE through
proprietary agreements, components, files or programs contained within this
package or product are Copyright -2023 Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/cce/17.0.1/ATTRIBUTIONS_17.0.txt
Cray PE DL Plugin with Resiliency Support 22.06.1.2:
Release Date:
-------------
June 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 21.04.1:
o Updated DL framework support
o Introduction of resiliency support
o BFloat16 data type support
Supported Configurations:
-------------------------
Shasta/EX: CPU and AMD GPU support with PrgEnv-gnu
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10.2
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
For info regarding use of the new resiliency feature, see the examples
and README.md in ${CRAYPE_ML_PLUGIN_BASEDIR}/examples/resiliency
Sample Installation instructions:
---------------------------------
Note: Package must be installed or sym linked in the default /opt/cray/pe path.,
ie: use of the --prefix flag for rpm is not supported.
rpm -ivh craype-dl-plugin-ftr-22.06.1.2*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
Shasta/EX:
module load craype-dl-plugin-ftr/22.06.1.2
Certain components, files or programs contained within this package or
product are Copyright 2017-2022 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 21.02.1.3:
Release Date:
-------------
February 2021
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 20.10.1:
o Bug fixes for CUDA 11 support and pip install of include source dist
o Initial CUDA 11.1 support
o Updated DL framework support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC CLE 7: CPU and GPU support
Shasta: CPU-only support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.4
* PyTorch v1.7.1
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-21.02.1.3*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/Shasta:
module load craype-dl-plugin-py3/21.02.1.3
CS:
module load craype-dl-plugin-py3/openmpi/21.02.1.3
module load craype-dl-plugin-py3/mvapich/21.02.1.3
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.09.1:
Release Date:
-------------
September 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.08.1:
o Keras callbacks bug fixes
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.09.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.09.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.12.1:
Release Date:
-------------
December 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.09.1:
o TF v2.9 Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.9
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.12.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.12.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
CrayPE Targets 1.13.0
================================================================================
Release Date:
--------------------------------------------------------------------------------
March, 2023
Purpose:
--------------------------------------------------------------------------------
Adds modulefiles for targets - craype-accel-amd-gfx942
Bugs fixed in this release:
--------------------------------------------------------------------------------
None
Dependencies:
--------------------------------------------------------------------------------
The CrayPE Targets 1.13.0 release is dependent on the following
software products:
CrayPE
set_default_3
lmod_scripts
Documentation:
--------------------------------------------------------------------------------
See manpages for cc, CC, ftn, intro_craype-api and pkg-config
See section 2.6 Using Targeting Modules of the Cray Programming Environment
User's Guide (S-2529-116)
See http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh craype-targets-ex-hpcm-1.13.0-202402062025.1d638354624fb-0.ex.hpcm.noarch.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_craype-targets_1.13.0-tcl
/opt/cray/pe/admin-pe/set_default_files/set_default_craype-targets_1.13.0-lua
Product description:
--------------------------------------------------------------------------------
CrayPE Targets contains the modules for CrayPE targets.
TCL Modules - craype-accel-host craype-accel-amd-gfx90a craype-accel-amd-gfx908 craype-accel-amd-gfx940 craype-accel-amd-gfx942 craype-accel-intel-max craype-accel-nvidia70 craype-accel-nvidia80 craype-accel-nvidia90 craype-hugepages2M craype-hugepages4M craype-hugepages8M craype-hugepages16M craype-hugepages32M craype-hugepages64M craype-hugepages128M craype-hugepages256M craype-hugepages512M craype-hugepages1G craype-hugepages2G craype-arm-grace craype-x86-milan craype-x86-milan-x craype-x86-rome craype-x86-spr craype-x86-spr-hbm craype-x86-trento craype-x86-genoa craype-network-none craype-network-ofi craype-network-ucx
Lmod Modules - craype-accel-host craype-accel-amd-gfx90a craype-accel-amd-gfx908 craype-accel-amd-gfx940 craype-accel-amd-gfx942 craype-accel-intel-max craype-accel-nvidia70 craype-accel-nvidia80 craype-accel-nvidia90 craype-hugepages2M craype-hugepages4M craype-hugepages8M craype-hugepages16M craype-hugepages32M craype-hugepages64M craype-hugepages128M craype-hugepages256M craype-hugepages512M craype-hugepages1G craype-hugepages2G craype-arm-grace craype-x86-milan craype-x86-milan-x craype-x86-rome craype-x86-spr craype-x86-spr-hbm craype-x86-trento craype-x86-genoa craype-network-none craype-network-ofi craype-network-ucx
Copyright 2013-2023 Hewlett Packard Enterprise Development LP
CDEUX driver (craype) 2.7.31.11 - 11
==============
Release Date:
--------------
March 2024
Purpose:
--------
The Cray Development Environment User eXperience driver (CDEUX) (craype)
provides a common set of drivers for all CPE programming environments,
notably the C/C++ (cc/CC) and Fortran (ftn) languages.
New CDEUX driver (craype) features for HPE Cray EX and Apollo systems:
- Support for CCE using distribution GCC toolchain installation.
- Improved support for AMD ROCm installations.
- Improved AARCH64 support.
Key Changes and Bugs Closed:
----------------------------
- CDEUX driver (craype) 2.7.31.11 has fixed issues regarding OpenMP
Product and OS Dependencies:
----------------------------
The CDEUX driver (craype) 2.7.31.11 release is supported on the following
HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Documentation:
---------------
See manpages for cc, CC, ftn, and intro_craype-api.
See the pkg-config manpage or
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Product description:
--------------------
CrayPE contains drivers, cc, CC, and ftn to compile for all CPE
Programming Environments.
Certain components, files or programs contained within this package or product
are Copyright 2013-2023 Hewlett Packard Enterprise Development LP.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright 2023 Hewlett Packard
Enterprise Development LP.
Craypkg-gen 1.3.32
==============
Release date
------------
February 2024
Purpose
-------
Bug fix release.
Bugs fixed in this release
--------------------------
- Preserve build numbers in ROCm versions
- Exclude hip paths in the ROCm pkg-config files for ROCm >= 5.7.0
- Change -D_HIP_PLATFORM_HCC_ to -D__HIP_PLATFORM_AMD__ in pkg-config files for ROCm >= 5.7.0
- Utilize Jinja2 templating for ROCm pkg-config file
- Update logic to generate modulefiles for new file tree in Intel OneAPI 2024.0
- Update the Nvidia HPC SDK compiler compatibility version for aarch64
- Remove non-existent CUPTI path from Nvidia HPC SDK modulefile
Dependencies
------------
The craypkg-gen 1.3.32 release is supported on the following HPE
Cray systems:
- Cray EX systems SLE 15.0 or later and RHEL 8.0 or later.
Driver support for integrating Third Party C, C++, and Fortran libraries
through .pc files using pkg-config is used in CrayPE 2.x and later.
Limitation
----------
- Library dependencies for static libraries are not added to the .pc files
for keywords Requires.private and Libs.private. A warning is issued by
the craypkg-gen tool to advise users to add this information to the
libraries .pc files.
- RPM limits packages to 4GB
Documentation
-------------
Man pages for craypkg-gen are found by executing `module load craypkg-gen`
and then `man craypkg-gen`. See
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Examples for creating modulefiles for Intel, PGI and Python are included in
the craypkg-gen ‘doc’ directory:
/opt/cray/craypkg-gen/1.3.32/doc/intel_example.txt
/opt/cray/craypkg-gen/1.3.32/doc/pgi_example.txt
/opt/cray/craypkg-gen/1.3.32/doc/python_example.txt
Example:
- As an example, the Intel 16.0.3.210 compiler was recently released.
After installing the compiler the administrator creates a modulefile for
this release by executing the following commands:
# module load craypkg-gen
# craypkg-gen -m /opt/intel/compilers_and_libraries_2016.3.210
This version of the Intel compiler is made default by executing the
command:
# /opt/admin-pe/set_default_craypkg/set_default_intel_16.0.3.210
Installation instructions
-------------------------
rpm -ivh craypkg-gen-1.3.32-1.3.32-1.el8.noarch.rpm
To change the product version to default after installation:
/opt/cray/pe/admin-pe/set_default_files/set_default_craypkg-gen_1.3.32
Product description
-------------------
The craypkg-gen 1.3.32 utility provides the system administrator
a tool to integrate third party software with the Cray software stack.
Craypkg-gen assists with integration by creating .pc files for C, C++, and
Fortran libraries, pkg-config enabled modulefiles and RPMs.
The workflow for using craypkg-gen is
1) Build the open source software
2) Create .pc files for libraries
3) Create pkg-config enabled modulefiles
4) Customize pkg-config and modulefiles if needed
Certain components, files or programs contained within this package or product
are Copyright 2013-2024 Hewlett Packard Enterprise Development LP.
Cray DSMML 0.3.0:
==========================
Release Date:
-------------
March, 2024
Product Information:
--------------------
DSMML is a stand-alone memory management library for maintaining distributed
shared symmetric memory heaps for top level PGAS languages and libraries like
Coarray Fortran, UPC, and OpenSHMEM. DSMML allows user libraries to create
multiple symmetric heaps and share information with other libraries. Through
DSMML, we could extract interoperability between PGAS programming models.
Announcements, release informations, supported environments, and backward
compatibility informations about this product can be viewed in the following
location: https://pe-cray.github.io/cray-dsmml/
Purpose:
--------
Cray DSMML 0.3.0 supports the following features:
- Create symmetric heap segments for being used by PGAS languages and
libraries
- Allows options to allocate, deallocate memory buffers from previously
created symmetric heap segments
- Enables sharing information about the symmetric heap segments between
different programming models
- Added support for shared symmetric heap (SSHEAP) that can be used to
provide enhanced SMP data transport in user-libraries like SHMEM, CAF, UPC
The following features are fixed in Cray DSMML 0.3.0 when compared
to Cray DSMML 0.2.2:
- Add support for the library for AARCH64 processor based platforms
- Internal bug fixes involving hugepage cleanups
Refer intro_dsmml(3) manpage for more information on the supported features
and syntax for different routines
Product and OS Dependencies:
----------------------------
The Cray DSMML 0.3.0 release is supported on the following
Cray systems:
* HPE Cray EX systems with CLE
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.3 |
----------------------------+---------------------+
One or more compilers:
* AOCC 2.2 or later
* CCE 9.1 or later
* GNU 9.1 or later
* Intel 19.0 or later
* Nvidia 20.7 or later
Documentation:
--------------
Use the Cray DSMML man pages for more information on the library
functions and use intro_dsmml(3) man page for general information.
Use https://pe-cray.github.io/whitepapers/ for access to different
Cray DSMML specific whitepapers
Announcements, release informations, supported environments, and backward
compatibility informations about this product can be viewed in the following
location: https://pe-cray.github.io/cray-dsmml/
Modulefile:
-----------
module load cray-dsmml/0.3.0
Certain components, files or programs contained within this package or
product have the following Copyright:
Copyright 2018-2021 Hewlett Packard Enterprise Development LP.
FFTW 3.3.10.7
============
Release Date:
-------------
March 2024
Purpose:
--------
This Cray FFTW 3.3.10.7 release is supported on Cray EX (formerly
Shasta) systems. FFTW is supported on the host CPU but not on the
accelerator of Cray systems.
The Cray FFTW 3.3.10.7 release provides the following:
- Compiler and dependency updates
See the Product and OS Dependencies section for details.
Product and OS Dependencies:
----------------------------
The Cray FFTW 3.3.10.7 release is supported on the following Cray systems:
Cray EX Systems with SLES 15 SP4 or later based OS
The FFTW 3.3.10.7 release requires the following software products:
craype 2.7.5 or later
MPT 8.0 or later
One or more of the following major compiler versions:
CCE 17
GCC 12, 13
AOCC 4.1
Intel 2024 or later
Notes and Limitations:
----------------------
Starting with cray-fftw/3.3.6.1 the fftw module has been renamed from fftw
to cray-fftw.
Documentation:
--------------
http://www.fftw.org/index.html#documentation
See the intro_fftw3 manual page for additional information.
Modulefile:
-----------
module load cray-fftw
Installation
------------
To install the rpm, execute:
rpm -ivh cray-fftw-3.3.10.7-202402071520.4d6d346567dfd-1.shasta.el8.x86_64.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_fftw_3.3.10.7
Certain components, files or programs contained within this package or
product are Copyright 2011-2024 Hewlett Packard Enterprise Development LP
# HDF5 1.12.2.11
## Release Date:
* March 2024
## Purpose:
This HDF5 release contains the following:
* Support for compilers on SLES x86_64: CCE 17, Native GCC 12.3.0, Intel-OneAPI 2023.2.0, AOCC 4.2, Nvidia 23.3
* Support for compilers on SLES aarch64: CCE 17, Native GCC 12.3.0, Nvidia 23.11
* Support for compilers on RHEL 8: CCE 17, GCC Toolset 12.2.1, Intel-OneAPI 2023.2.0, AOCC 4.1, Nvidia 23.3
## Product and OS Dependencies:
The HDF5 1.12.2.11 release is supported on the following systems, requiring the following software products:
* CrayPE 2.1.2 or later
* Cray MPT 8.1 or later
SLES x86_64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
SLES aarch64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Nvidia 23.11 or later
RHEL 8
* One or more compilers:
* CCE 17.0 or later
* GCC Toolset 12.2.1 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
## Notes and Limitations:
Users are encouraged to use the Cray compiler scripts (cc, ftn, CC) to include the HDF5 header files and link in the HDF5 libraries, rather than the HDF5 compiler scripts (h5cc, h5fc, h5c++) included in the release.
## Documentation:
http://www.hdfgroup.org/HDF5/doc/index.html
## Modulefile:
`module load cray-hdf5`
OR
`module load cray-hdf5-parallel`
## Product description:
HDF5 is a data model, library, and file format for storing and managing data. It supports an unlimited variety of datatypes, and is designed for flexible and efficient I/O and for high volume and complex data. HDF5 is portable and is extensible, allowing applications to evolve in their use of HDF5. The HDF5 Technology suite includes tools and applications for managing, manipulating, viewing, and analyzing data in the HDF5 format.
## Installation:
`rpm -ivh cray-hdf5-1.12.2.11-*202402200749.acf05c8988d11-3.el8.x86_64.rpm`
The "`*`" in the install command represents compiler version combinations.
To make this the default version, execute:
`/opt/cray/pe/admin-pe/set_default_files/set_default_hdf5_1.12.2.11`
Certain components, files or programs contained within this package or product are Copyright 2019-2024 Hewlett Packard Enterprise Development LP.
Cray LibSci 24.03.0
=====================
Release Date:
-------------
March 2024
Purpose:
--------
Cray LibSci 24.03.0 provides scientific libraries for Cray
HPC systems. Cray LibSci is supported on the host CPU but
not on the accelerator of these systems.
The Cray LibSci 24.03.0 release provides the following:
* support for RHEL gcc-toolset-{10,11,12}
* support for Nvidia HPC SDK 23.11 on aarch64
Cray LibSci 24.03.0 includes the following versions of publicly
available libraries:
* LAPACK 3.10.1 - For further information, see
http://www.netlib.org/lapack
* ScaLAPACK 2.2.0 - (Scalable LAPACK) For further information, see
http://www.netlib.org/scalapack.
* QDWH 2.0.0 KSVD 1.0.0 - Polar decomposition and SVD packages, see
https://github.com/ecrc/qdwh
https://github.com/ecrc/ksvd
Key Changes and Bugs Closed:
----------------------------
Changes in Cray LibSci 24.03.0
* CAST-34956 - Lua module environment variable issue
Changed in Cray LibSci 23.12.5
* CAST-33777 - LibSci mutex contention issue
* Removed support for Intel KNL
Product and OS Dependencies:
----------------------------
The Cray LibSci 24.03.0 release requires the following:
* SLES 15 SP4/SP5 or RHEL 8.8
* cray-mpich/8.1.0 or later
Supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Cray XD systems with CLE
One or more of the following compiler major versions:
* CCE 17.x (SLES)
* CCE 16.x or later (RHEL)
* GCC 12.3 or later (SLES)
* gcc-toolset-10.3/11.2/12.2 (RHEL)
* AOCC 4.1.0 or later
* AMD ROCm 5.0.0 or later
* Intel 2023.2.0 or later
* Nvidia 23.11 or later (aarch64) and 23.3 or later (x86_64)
Notes and Limitations:
----------------------
OMP threaded versions:
The CrayPE 2.1.2 and later releases add support for link line generation
for the multi-threaded versions of the libsci library based on the OpenMP
flags the user specifies for each compiler:
* CCE by default links to the OpenMP LibSci library. CrayPE will link in
the serial version of LibSci when the CCE flag -hnoomp is used.
* GNU by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the GNU flag -fopenmp is used.
* INTEL by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the INTEL flag -qopenmp is used.
Stack size limit:
The 'ulimit -s unlimited' system stack size setting is required for
cray-libsci on all supported Cray platforms. This is typically set as a
system default by admin but may otherwise need to be set at runtime.
Non-default dynamic linking:
When using a non-default version of cray-libsci with dynamic linking
users should set at runtime and at linktime after loading the desired
cray-libsci module--or set the equivalent for their linux shell:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci man page for additional information.
See the csmlversion man page for information to display version information
on the currently loaded scientific libraries.
Modulefile:
-----------
module load cray-libsci
Installation instructions:
--------------------------
LibSci is now packaged into separate compiler specific RPMs to allow
rpmbuild to correctly include compiler dependencies.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
/opt/cray/libsci/24.03.0/ATTRIBUTIONS
Cray LibSci_ACC 24.03.1
=======================
Release Date:
-------------
March 2024
Purpose:
--------
The Cray LibSci_ACC 24.03.1 release provides accelerated versions of
scientific libraries for HPE Cray systems with accelerators.
Product and OS Dependencies:
----------------------------
Cray LibSci_ACC 24.03.1 requires the following platform:
HPE Cray EX system with AMD MI250X
HPE Cray EX system with AMD MI300A
HPE Cray EX system with Nvidia Hopper
Cray LibSci_ACC 24.03.1 supports the following OS:
RHEL 8.8 x86_64
SLES 15 SP4 x86_64
SLES 15 SP5 x86_64 & aarch64
Notes and Limitations:
----------------------
LAPACK workspace calculations from other implementations or hard-coded
values may not be compatible with libsci_acc. Workspaces reported from
a query to the corresponding libsci_acc function should be used.
The current supported usecase is a one-rank to one-gpu mapping for apps.
Performance improvements can be achieved in programs calling libsci_acc
subroutines by using pinned memory. See the intro_libsci_acc man page for
details.
Use of the aprun option "-cc none" or the srun option "--cpu_bind=none"
to disable core affinity is strongly suggested to maintain performance.
When using non-default cray-libsci_acc versions users should set equivalent
at runtime and linktime after loading the desired cray-libsci_acc module:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci_acc man page for additional information.
Modulefile:
-----------
module load cray-libsci_acc
Sample Installation instructions:
---------------------------------
RPM install command:
rpm -ivh cray-libsci-acc-24.03.1-*.##ARCH##.rpm
The "*" in the install command represents compiler version combinations.
To make this the default version:
##PREFIX##/admin-pe/set_default_files/set_default_libsci_acc_24.03.1
License:
--------
Except for the third party modules and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright -2023 Hewlett
Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in
this package are detailed in the file:
##PREFIX##/libsci_acc/24.03.1/ATTRIBUTIONS_libsci_acc_24.03.1.txt
Cray Lmod 8.7.32
================================================================================
Release Date:
--------------------------------------------------------------------------------
March, 2024
Purpose:
--------------------------------------------------------------------------------
Provide a Lmod 8.7.32 package for CPE customers
Bugs fixed in this release:
--------------------------------------------------------------------------------
N/A
Documentation:
--------------------------------------------------------------------------------
See CPE's Installation and User Guides for documentation.
Also see Lmod's official website: https://lmod.readthedocs.io
Dependencies:
--------------------------------------------------------------------------------
The Lmod Scripts 8.7.32 release is dependent on the following
software:
bc
sed
lua >= 5.3
lua-luaposix
lua-luafilesystem
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh cray-lmod-8.7.32-1.el8.x86_64.rpm
To make this the default version, update the symbolic "lmod" link
to the desired Lmod version directory.
Lmod Scripts 3.2.1
================================================================================
Release Date:
--------------------------------------------------------------------------------
December, 2023
Purpose:
--------------------------------------------------------------------------------
Updates package license.
Bugs fixed in this release:
--------------------------------------------------------------------------------
Documentation:
--------------------------------------------------------------------------------
See system Installation and User Guides for documentation
Dependencies:
--------------------------------------------------------------------------------
The Lmod Scripts 3.2.1 release is dependent on the following
software:
set_default_3
Lua
Lmod
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh lmod_scripts-3.2.1-202311141921.212030d47edcb-0.noarch.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_lmod_scripts_3.2.1
Product description:
--------------------------------------------------------------------------------
Contains Lua scripts which support the Cray implementation of Lmod
Copyright 2020-2023 Hewlett Packard Enterprise Development LP
modules 3.2.11.7
==================
Release Date:
-------------
Dec, 2023
Purpose:
--------
- Package dependency update
Documentation:
---------------
See man-pages for module and modulefile.
Cray MPICH 8.1.29:
=======================================
Release Date:
-------------
February 21, 2024
Purpose:
--------
Cray MPICH 8.1.29 is based upon ANL MPICH 3.4a2 with support for libfabric
and is optimized for the Cray Programming Environment.
Major Differences Cray MPICH 8.1.29 from the XC Cray MPICH include:
- Uses the new ANL MPICH CH4 code path and libfabric for network
support.
- Does not support -default64 mode for Fortran
- Does not support C++ language bindings
New Cray MPICH features for HPE Cray EX and Apollo systems:
- Starting from the 8.1.26 release, Cray MPICH supports the Intel Sapphire Rapids CPU HBM
processor architecture.
- On systems with AMD GPUs, Cray MPICH 8.1.26 supports all ROCm
versions starting from ROCm 5.0, including the latest ROCm 5.5.0
release.
The Cray MPICH 8.1.25 release and prior versions of
Cray MPICH are only compatible with ROCm versions up to (and
including) the ROCm 5.4.0 release.
- Cray MPICH uses the libfabric "verbs;ofi_rxm" provider by default.
This is the supported and optimized OFI libfabric provider for
Slingshot-10 and Apollo systems.
- Cray MPICH offers support for multiple NICs per node. Starting with
version 8.0.8, by default Cray MPICH will use all available NICs on
a node. Several rank-to-NIC assignment policies are supported. For
details on choosing a policy for assigning ranks to NICS, or for
selecting a subset of available NICs, please see the following
environment variables documented in the mpi man page.
MPICH_OFI_NIC_VERBOSE
MPICH_OFI_NIC_POLICY
MPICH_OFI_NIC_MAPPING
MPICH_OFI_NUM_NICS
- Enhancements to the MPICH_OFI_NIC_POLICY NUMA mode have been added.
Starting with version 8.0.14, if the user selects the NUMA policy,
the NIC closest to the rank is selected. A NIC no longer needs to
reside in the same numa node as the rank. If multiple NICs are
assigned to the same numa node, the local ranks will round-robin
between them. Numa distances are analyzed to select the closest NIC.
- Cray MPICH supports creating a full connection grid during MPI_Init.
By default, OFI connections between ranks are set up on demand. This
allows for optimal performance while minimizing memory requirements.
However, for jobs requiring an all-to-all communication pattern, it
may be beneficial to create all OFI connections in a coordinated
manner at startup. See the MPICH_OFI_STARTUP_CONNECT description in
the mpi man page.
- Cray MPICH supports runtime switching to the UCX netmod starting
with version 8.0.14. To do this load the craype-network-ucx module
and module swap between Cray-MPICH and Cray-MPICH-UCX modules. For
more information including relevant environment variables reference
the intro_mpi man page with the Cray-MPICH-UCX module loaded.
- Lmod support for HPE Cray EX starting with Cray MPICH 8.0.16.
Key Changes and Bugs Closed:
----------------------------
Changes in Cray MPICH 8.1.29
New Features:
- Cray MPICH 8.1.29 for aarch64 is in early stage of support and
has some issues. Setting these variables will help circumvent most
known issues.
export MPICH_SMP_SINGLE_COPY_MODE=CMA
export MPICH_MALLOC_FALLBACK=1
Bugs Fixed:
- CAST-34406 - Improve performance in MPI Allreduce benchmark
- CAST-34872 - Fix seg-fault in extremely large allreduce calls
- CAST-34972 - Use dynamic CXI PID to avoid conflicts
- PE-45873 - Added infrastructure for CXI triggered operations non-blocking collectives
- PE-45874 - Add MPI_Ibcast() using CXI triggered operations
- PE-45876 - Add MPI_Igather() using CXI triggered operations
- PE-45877 - Add MPI_Iscatter() using CXI triggered operations
- PE-45878 - Add MPI_Ireduce() using CXI triggered operations
- PE-45879 - Add MPI_Iallreduce() using CXI triggered operations
- PE-49870 - Support MPI_LONG_DOUBLE with CCE for aarch64
- PE-51028 - Improved debugging to diagnose memcpy issues
- PE-51233 - Build libraries for the nvidia compiler on aarch64
- PE-51326 - Fix window creation using MPI_WIN_FLAVOR_SHARED to avoid running out of fi_mr resources
- PE-51365 - Fix bad initialization of hint striping_unit in MPI-IO
- PE-51445 - Support ROCM 6.0 with Cray-MPICH GTL
- PE-51470 - Update Dynamic Process Management documentation
- PE-51524 - Document PBS options for using Spawn on Slingshot
- PE-51738 - Bugfix for GPU NUMA nodes on MI300 GPUs
Product and OS Dependencies:
----------------------------
The Cray MPICH 8.1.29 release is supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.6 |
----------------------------+---------------------+
cray-pals | >= 1.0.6 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
One or more compilers:
* AMD ROCM 5.0 or later
* AOCC 4.1 or later
* CCE 16.0 or later
* GNU 10.3 or later
* Intel 2022.1 or later
* Nvidia 23.3 or later
Notes and Limitations:
----------------------
Limitations in Cray MPICH 8.1.29:
- Cray MPICH 8.1.29 can support only ~2040 simultaneous MPI
communicators. This limit is less the XC Cray MPICH limit of
~4090 simultaneous communicators. Cray intends to raise the
limit in a future release of Cray MPICH for Shasta to at least
the XC limit.
Documentation:
--------------
For more information see the intro_mpi man page.
Modulefile:
-----------
module load cray-mpich/8.1.29
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright -2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/mpich/8.1.29/ATTRIBUTIONS
Copyright -2023 Hewlett Packard Enterprise Development LP
# NetCDF 4.9.0.11
## Release Date:
* March 2024
## Purpose:
This NetCDF release contains the following:
* Support for compilers on SLES x86_64: CCE 17, Native GCC 12.3.0, Intel-OneAPI 2023.2.0, AOCC 4.2, Nvidia 23.3
* Support for compilers on SLES aarch64: CCE 17, Native GCC 12.3.0, Nvidia 23.11
* Support for compilers on RHEL 8: CCE 17, GCC Toolset 12.2.1, Intel-OneAPI 2023.2.0, AOCC 4.1, Nvidia 23.3
## Product and OS Dependencies:
The NetCDF 4.9.0.11 release is supported on the following systems, requiring the following software products:
* Cray HDF5 1.12.2.*
* CrayPE 2.1.2 or later
Cray MPT 8.1 or later
SLES x86_64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
SLES aarch64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Nvidia 23.11 or later
RHEL 8
* One or more compilers:
* CCE 17.0 or later
* GCC Toolset 12.2.1 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
## Notes and Limitations:
For certain F77 source files, when compiling with `gcc/10+`, it may be necessary to configure/compile with:
`export FCFLAGS="-fallow-argument-mismatch"`
`export FFLAGS="-fallow-argument-mismatch"`
This will turn mismatch errors between actual and dummy argument lists to warnings. See:
https://github.com/Unidata/netcdf-fortran/issues/212
https://gcc.gnu.org/gcc-10/changes.html
Unidata now packages Netcdf-4 and legacy Netcdf-3 separately. HPE has decided not to continue supplying the legacy Netcdf-3 package. Due to CCE changes a version of netcdf built with "-sreal64" is neither needed nor provided.
NetCDF is supported on the host CPU but not on the accelerator.
## Documentation:
http://www.unidata.ucar.edu/software/netcdf/docs
## Modulefile:
`module load cray-netcdf`
OR
`module load cray-netcdf-hdf5parallel`
## Product Description:
NetCDF (network Common Data Form) is a set of interfaces for array-oriented data access and a freely-distributed collection of data access libraries for C, Fortran, C++, Java, and other languages. The netCDF libraries support a machine-independent format for representing scientific data. Together, the interfaces, libraries, and format support the creation, access, and sharing of scientific data.
## Installation:
`rpm -ivh cray-netcdf-4.9.0.11-*202402202022.484a8086ac4a4-2.el8.x86_64.rpm`
The "*" in the install command represents compiler version combinations.
To make this the default version, execute:
`/opt/cray/pe/admin-pe/set_default_files/set_default_netcdf_4.9.0.11`
Certain components, files or programs contained within this package or product are Copyright 2019-2024 Hewlett Packard Enterprise Development LP.
Cray PALS 1.3.2:
================
Release Date:
-------------
Nov 2023
Purpose:
--------
Cray Parallel Application Launch Service (PALS) is an application launcher for
Cray PE applications.
Key Changes:
------------
Add host2xname tool
Allow configuring the PMI port range
Add more examples in manpages
Add SLES 15 SP5 build
Add SLES 15 SP4 and SP5 aarch64 build
Add PMI and PMI2 as --pmi options for mpiexec
Periodically ping applications
Add the ability to allocate additional VNIs for special services like DAOS
Add mpiexec --disable-rdzv-get option to disable rendezvous gets in the NICs
Enable --cpu-bind and --gpu-bind=verbose to work with --rankfile
Add PALSD_FANOUT env var for palsd to set fanout tree width
Add ability for simultaneous mpiexecs to divide up PBS job slots
Add RHEL 8.8 build
Add support for AMD & NVIDIA GPU visibility
Change the wire format for PMIx data to make PMIx_Fence more efficient
Increase core file size limit on RHEL systems
Add retries to SPIRE auth token call
Install aprun and mpirun as symlinks
Use position-independent executables
Change F_CXI_COLL_* -> FI_CXI_COLL_* environment variables
Configure Instant On URL on Shasta
Move binaries and libraries to /opt/cray
Change pmi_example to print NIC distances for all ranks
Make PALS command's nodelists consistent
Set ATOM_SOCKET by default
Improve vnid request error reporting
Launch with PMIx support automatically fetch NIC information
Reduce default fanout to 32 nodes
Require PMIx version 4.2.4 or newer
Send the job-step name to vnid when creating a job-step VNI
Include pid in palsd log messages
Raise maximum number of ZMQ sockets to 4096
Munge auth no longer used for some internal RPCs with trusted connections
Increase default send timeout to 8s
Fix errors found with clang scan-build
Fix nohup mpiexec behavior
Improve cleanup when launch RPC fails
Implement the verbose option for GPU binding
Fix support for the cycle keyword with the mpiexec --rankfile option
Fix error checking for GPU binding options
Better handle mpiexec disconnects
Fix segfault on PMIx RPC for missing app
Fix spawn-related memory leaks
Fix misleading warning in line buffer mode
Remove SLES 15 SP3 build
Deprecate PALS_FANOUT and APRUN_FANOUT environment variables
Install cray-pals RPM
Documentation:
--------------
For more information see aprun, mpiexec, palscmd, palsctrl, palsig, and palstat man page.
Modulefile:
-----------
module load cray-pals
Installation instructions:
--------------------------
rpm -ihv cray-pals-<version>-<date>.el8.x86_64.rpm
rpm -ihv cray-palsd-<version>-<date>.el8.x86_64.rpm
Copyright 2022-2023 Hewlett Packard Enterprise Development LP
cray-papi
=================
Release date:
-------------
December 2023
Package:
--------
cray-papi-7.0.1.2-202310131612.70757f986edac-4 x86_64
Purpose:
--------
New version of papi 7.0.1.2 release 4
Documentation:
--------------
Overview: https://github.com/icl-utk-edu/papi
Web Site: https://icl.utk.edu/papi
Product description:
--------------------
PAPI aims to provide the tool designer and application engineer with a
consistent interface and methodology for use of the performance counter
hardware found in most major microprocessors. PAPI enables software
engineers to see, in near real time, the relation between software
performance and processor events.
Dependencies:
-------------
For a list of software used when validating this version of
PAPI on Cray and HPE systems, see the HPE Cray Programming
Environment release announcements.
Copyright 2015-2017,2019-2023 Hewlett Packard Enterprise Development LP
# Parallel NetCDF 1.12.3.11
## Release Date:
* March 2024
## Purpose:
This Parallel NetCDF release contains the following:
* Support for compilers on SLES x86_64: CCE 17, Native GCC 12.3.0, Intel-OneAPI 2023.2.0, AOCC 4.2, Nvidia 23.3
* Support for compilers on SLES aarch64: CCE 17, Native GCC 12.3.0, Nvidia 23.11
* Support for compilers on RHEL 8: CCE 17, GCC Toolset 12.2.1, Intel-OneAPI 2023.2.0, AOCC 4.1, Nvidia 23.3
See the Parallel NetCDF documentation for details and usage information.
## Product and OS Dependencies:
This Parallel NetCDF 1.12.3.11 release is supported on on the following systems, requiring the following software products:
* CrayPE 2.1.2 or later
* Cray MPT 8.1 or later
SLES x86_64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
SLES aarch64
* One or more compilers:
* CCE 17.0 or later
* Native GCC 12.3.0 or later
* Nvidia 23.11 or later
RHEL 8
* One or more compilers:
* CCE 17.0 or later
* GCC Toolset 12.2.1 or later
* Intel-OneAPI 2023.2.0 or later
* Nvidia 22.0 or later
* AOCC 4.1 or later
* AMD 5.0 or later
## Notes and Limitations:
For certain F77 source files, when compiling with `gcc/10+`, it may be necessary to configure/compile with:
`export FCFLAGS="-fallow-argument-mismatch"`
`export FFLAGS="-fallow-argument-mismatch"`
This will turn mismatch errors between actual and dummy argument lists to warnings. See:
https://github.com/Unidata/netcdf-fortran/issues/212
https://gcc.gnu.org/gcc-10/changes.html
## Documentation:
https://parallel-netcdf.github.io/wiki/Documentation.html
## Modulefile:
`module load cray-parallel-netcdf`
## Product Description:
Parallel NetCDF is a library providing high-performance I/O while maintaining file-format compatibility with Unidata's NetCDF.
## Installation:
`rpm -ivh cray-parallel-netcdf-1.12.3.11-*202402200749.41338063d7353-1.el8.x86_64.rpm`
The "`*`" in the install command represents compiler version combinations.
To make this the default version, execute:
`/opt/cray/pe/admin-pe/set_default_files/set_default_parallel-netcdf_1.12.3.11`
Certain components, files or programs contained within this package or product are Copyright 2019-2024 Hewlett Packard Enterprise Development LP.
Perftools 23.12.0
===============
Release Date: December, 2023
Purpose:
========
This is a feature and bugfix release for the following systems:
- HPE Cray EX and HPE Cray Supercomputer Systems with HPCM
- HPE Cray EX and HPE Cray Supercomputer Systems with CSM
- HPE Apollo 2000 Gen 10 Plus Systems (x86)
Key enhancements or changes from the previous release:
======================================================
o Upgrade to DWARF 0.8.0
o Upgrade to PAPI 7.0.1.2
- Upstream bugfixes
- Upstream CUDA component version locked from prior release and renamed cuda_7011
- Upstream ROCm component version locked from release 7.0.0.1 and renamed rocm_7001
- Support for Nvidia Grace processor
o Add support for ROCm 5.7.0
o Add support for CUDA 12.2
o Add support for Nvidia H100 GPUs
o Update ROCm trace wrappers to support up to ROCm 5.7
o Add support for PAT_region_push/pop in full trace mode
o Add support for profiling Python 3.11
o Add a DSMML tracegroup
o Improved counter groups and derived metrics for AMD Genoa processors
o Add support for OMPT target callbacks with CCE's OpenMP runtime, enabled
via the CRAYPAT_OMP_TOOL environmnet variable (NOTE: this is primarily intented
for use with pat_run. When instrumenting executables with CCE, proprietary
tracepoints are used in favor of OMPT. To use OMPT instead, '-fno-instrument-openmp'
must be added to compilation.)
o Perftools and its libraries are now compiled using the OS native
gcc and no longer depend on cray-gcc
o Set PAT_RT_ACC_MODEL_NAME in the runtime environment to manually specify
the accelerator model name.
Bugs Fixed:
===========
o Unexpected callers appearing in the calltree when using perftools-lite-loops
o An inappropriate assertion failure when MPI_Get_count returns MPI_UNDEFINED
o A runtime failure when incompatible versions of HIP and pat_run are used
o An issue preventing lambda and template function resolution using Perftools
with Caliper
o Segfaults when managing a hipStream with singletons
o Segfaults when tracing OpenACC codes
o Program hangs at dlopen using pat_run on certain Python codes
o Program hangs using the '-gheap' trace option on RHEL systems
o Incorrect number of PEs/Node reported when using PBS and PALS
o Incorrect launching of Apprentice2 when using reveal launcher
Notes:
======
o This release of Perftools only supports ROCm 5.7.0; SLES 15 support is limited to SP5
o This release of Perftools only supports CUDA 12.0 and later
Known Issues:
=============
o Fortran applications compiled against CCE 16 may fail during instrumentation
with the following error message: "hidden symbol `<SYMBOL>' in <LIB> is
referenced by DSO". To work around this error specify the following during
the pat_build invocation '-Dlink-instr=-lgcc_s'.
o When using OMPT with CCE's openmp runtime, to and from device transfers and
kernel executions originating from a single construct are treated as separate
events. For example, 'target map(tofrom:)' will accumulate 3 events each
time it is encountered.
o Nvidia device drivers do not correctly populate system files used by Perftools
for the Grace-Hopper superchip, impacting driver versions up to: 545.23.06.
Reports generated from applications run on these devices may list the accelerator
device as "Unknown". Additionally, usage of accelerator performance counter groups
is not supported when the device is unresolved. To work around this issue set the
following runtime environment variable prior to running the experiment
'PAT_RT_ACC_MODEL_NAME="GH200 120GB"'.
o Tracing experiments with accelerator performance counters and all sampling experiments
may produce unexpected behavior on Grace-Hopper superchips. At this time, there is
no work around this issue and the capabilities should be treated as experimental.
o Sampling experiments on aarch64 systems lead to fatal errors if
PAT_RT_CALLSTACK_MODE="unwind" is used. The setting is disabled for these situations
and PAT_RT_CALLSTACK_MODE="frames" is used instead.
o The Intel OneAPI Fortran compiler does not correctly produce the ELF .debug_pubnames or
.debug_names section in object files when the "-debug pubnames" option is specified
(as required when the perftools module is loaded). This prevents pat_build from collecting
the necessary information that supports tracing user-defined entry points. Specify the
"-Ddebug-names=n" option to pat_build so pat_build can collect the information from the
ELF .debug_info section.
o The Intel OneAPI Fortran compiler does not accurately represent formal parameters to a
subprogram in some cases. This results in the generation of incorrect intercepts for
user-defined entry points when pat_build -u is specified and results in seg faults when
the instrumented program is executed. Specify the "-Dtrace-force-deref=y" to pat_build
to account for the missing formal parameter information.
Deprecations and Removals:
==========================
o Remove support for PGI programming Environment
o Remove support for following trace groups: craymem, hbw, gni, jemalloc, memkind
Dependencies:
=============
o A PDF reader (such as evince, acroread or okular) is required to use the
app2 online help.
o For a list of software used when validating this version of
Perftools on Cray and HPE systems, see the HPE Cray Programming Environment
release announcements.
Installation instructions:
==========================
For Apollo 80 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 80 (ARM) System (S-8013).
For Apollo 2000 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 2000 Gen10 Plus (x86) System (S-8012).
For HPE Cray EX (Shasta Architecture) systems, refer to the Cray Asynchronous
Installer Guide (S-8003).
Installation of app2 remote client (+ server) on Mac systems:
-------------------------------------------------------------
Apprentice2Installer-23.12.0.dmg
RevealInstaller-23.12.0.dmg
The Cray Apprentice2 and Reveal installers for Mac are included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Mac OS El Capitan through Mojave. Double click on installer
to begin installation. The installer will walk you through the
process for your system.
Installation of app2 remote client (+ server) on Windows 10 systems:
--------------------------------------------------------------------
Apprentice2Installer-23.12.0.exe
The Cray Apprentice2 installer for Windows is included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Windows 10. Double click on installer to begin installation.
The installer will walk you through the process for your system.
Documentation:
==============
See the following man pages:
app2, grid_order, intro_craypat, pat_build, pat_help, pat_info, pat_opts,
pat_report, pat_run, pat_view, reveal
intro_papi
perftools-base, perftools-lite, perftools-preload
accpc, cray_cassini, cray_pm, hwpc, nwpc, papi_counters, uncore
PAPIlicnotices
Search for perftools on https://support.hpe.com to access documentation
on Cray Performance Measurement and Analysis Tools
License:
========
Except for the third party modules discussed below and software licensed
by HPE through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
$CRAYPAT_ROOT/ATTRIBUTIONS_perftools.txt
Cray PMI 6.1.14
==============================================================================
Release Date:
--------------
February 12, 2024
Purpose:
--------
The Cray Process Manager Interface Library provides the interface between the
application launcher and other communication libraries such as MPICH and SHMEM.
Cray PMI 6.1.14 is optimized for the Cray Programming Environment.
New Cray PMI features for HPE Cray EX and Apollo systems:
Key Changes and Bugs Closed:
----------------------------
Starting with Cray-PMI 6.1.0 the cray-pmi and cray-pmi-lib modules have been
combined into a single cray-pmi module. The cray-pmi-lib module will no longer
be installed by new cray-pmi rpms. When the cray-pmi module is loaded, the
pmi libraries will be linked with as-needed flags.
Bug fixes new to Cray PMI 6.1.14 include:
- CAST-34500 - Include unversioned pmi library symlinks in the CPE ld cache
- PE-51739 - Create modern PMI process map code
Product and OS Dependencies:
----------------------------
The Cray PMI 6.1.14 release is supported on the following systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product Dependencies:
----------------------------------------+
cray-pals | >= 1.0.6 |
-----------------------+----------------+
slurm | >= 20.02 |
-----------------------+----------------+
Notes and Limitations:
----------------------
Documentation:
--------------
For more information see the intro_pmi man page.
Modulefile:
-----------
module load cray-pmi/6.1.14
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright 2020-2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/pmi/6.1.14/ATTRIBUTIONS
cray-python 3.11.7
====================
Release Date
------------
March 2024
Purpose
-------
Cray Python is an implementation of the Python programming language for the Cray
Programming Environment. The numpy and scipy modules are configured to call Cray
Libsci routines. The mpi4py module is configured to call Cray MPICH routines.
Cray Python is designed to run Python codes on the compute nodes of an HPE
supercomputer. HPE does not make changes to the Python source or any of its
libraries nor does it plan to make changes in future releases.
The cray-python 3.11.7 release contains
- python-3.11.7
- numpy-1.24.4
- scipy-1.10.1
- mpi4py-3.1.4
- dask-2023.6.1
Product and OS Dependencies
---------------------------
The cray-python 3.11.7 release is supported on
- Cray EX systems running SLES 15 or RHEL 8
Documentation
-------------
https://www.python.org/doc
Changelog
---------
https://docs.python.org/release/3.11.7/whatsnew/changelog.html
Modulefile
----------
module load cray-python/3.11.7
Installation
------------
rpm -ihv cray-python-3.11.7-20240208204348.1cdbf56-1.el8.x86_64.rpm
To make this the default version, execute
[PREFIX]/admin-pe/set_default_files/set_default_python_3.11.7
Certain components, files, or programs contained within this package are
© Copyright 2021-2024 Hewlett Packard Enterprise Development LP
cray-R 4.3.2
============
Release Date
------------
March 2024
Purpose
-------
The cray-R 4.3.2 release.
cray-R is configured to call cray-libsci routines.
Product and OS Dependencies
---------------------------
The cray-R 4.3.2 release is supported on
- Cray EX systems running SLES 15 or RHEL 8
Documentation
-------------
https://cran.r-project.org/manuals.html
Modulefile
----------
module load cray-R/4.3.2
Installation
------------
rpm -ihv cray-R-4.3.2-20240208204434.1115d98-1.el8.x86_64.rpm
To make this the default version, execute:
[PREFIX]/admin-pe/set_default_files/set_default_R_4.3.2
Certain components, files or programs contained within this package or product are
Copyright 2018-2024 Hewlett Packard Enterprise Development LP
Cray OpenSHMEMX 11.7.1:
===============================
Release Date:
-------------
Mar, 2024
Purpose:
--------
OpenSHMEM is a Partitioned Global Address Space (PGAS) library interface
specification, which is the culmination of a standardization effort among
many implementers and users of SHMEM programming model. SHMEM has a long
history as a parallel programming model on HPE Cray supercomputer systems.
For the past two decades SHMEM library implementation in HPE Cray systems
evolved through different generations. Cray OpenSHMEMX is a proprietary,
OpenSHMEM specification compliant SHMEM implementation for HPE Cray
product line.
Key Changes:
------------
Major differences in Cray OpenSHMEMX 11.7.1 from
Cray OpenSHMEMX version 11.7.0 includes the following:
- Add support for logging HPE Slingshot NIC counter values
- Add support for using the library on platforms with AARCH64
processors
- Improve performance of shmem_quiet memory ordering operation
- Improve performance for allreduce collective operations when all
PEs are involved in the communication operations
Bugs Closed:
------------
The following bugs are fixed as part of the Cray OpenSHMEMX 11.7.1
release:
- Fix C11 generic APIs for shmem_sync operations
- Fix library usage with GNU and Intel compilers
Product and OS Dependencies:
----------------------------
The Cray OpenSHMEMX 11.7.1 release is supported on the following
Cray systems:
* HPE Cray EX systems with CLE
Product and OS Dependencies by network type:
--------------------------------------------------+
| HPE Slingshot |
----------------------------+---------------------+
craype | >= 2.7.3 |
----------------------------+---------------------+
cray-pals | >= 0.3.5 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
cray-dsmml | >= 0.2.2 |
----------------------------+---------------------+
One or more compilers:
* AOCC 2.2 or later
* CCE 9.1 or later
* GNU 9.1 or later
* Intel 19.0 or later
* Nvidia 20.7 or later
Notes and Limitations:
----------------------
Announcements, release information, supported environments, limitations,
usage models, and backward compatibility informations about this product
can be viewed in the following location:
https://pe-cray.github.io/cray-openshmemx/
Documentation:
--------------
Use the Cray OpenSHMEMX man pages for more information on the library
functions and use intro_shmem(3) man page for general information.
Initial support for Cray OpenSHMEMX man pages are derived from OpenSHMEM
standards specification document
Use https://pe-cray.github.io/whitepapers/ for access to different
Cray OpenSHMEMX specific whitepapers
Modulefile:
-----------
module load cray-openshmemx/11.7.1
License:
--------
Components, files, and programs contained within this package or
product are Copyright Hewlett Packard Enterprise Development LP
CPE Release Notes For SLES 15 SP4
HPE Cray Programming Environment 24.03 Full Release Notes for SLES15 SP4
=========================================================
HPE Cray Compiling Environment (CCE) 17.0.1
===========================================
Release Date:
-------------
March 2024
Purpose:
--------
CCE 17.0.1 provides Fortran, C, and C++ compilers for
HPE Cray supercomputer and HPE Apollo 2000 Gen10Plus systems.
Key Changes and Support with CCE 17.0.1:
----------------------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
- ROCm 6.0 Suppport
- Added OMPT device tracing support to enable profiling
- Tuned support for MI300A
- SLES gcc/13 support
- Address sanitizer support on GPUs in Fortran
-----------
CCE 17.0.0
- LLVM 17 base (Merges up to November 6th 2023)
- Fortran
- Support offload of DO CONCURRENT iterations to accelerators with unified shared memory
- Starting with 17.0, support passing any Fortran contained procedures to C/C++
- Code with both 17.0 and pre-17.0 objects files must be linked with 17.0 ftn
- Added the following Fortran 2023 Features:
- US 01 & 02 Allow much longer statement lines and overall statement length
- US 03 SPLIT and TOKENIZE intrinsics
- US 04 Trig functions that work in degrees
- US 06 selected_logical_kind intrinsic function
- US 07 Additional named constants to specify kinds in ISO_FORTRAN_ENV
- US 10 The at edit descriptor
- US 11 Control over leading zeros in output of real values
- US 15 Simple procedures
- US 20 Reduction specifier for do concurrent
- UK 01 Extend the intrinsic procedure c_f_pointer to allow its pointer result to have specified lower bounds
- A public namelist group can have a private namelist group object.
- Added the following OpenMP 5.1 Features:
- 'error' directive
- 'omp_get_mapped_ptr' routine
- 'omp_display_env' routine
- 'thread_limit' clause on 'target' construct (C/C++ only)
- New initial device number value
- loop construct
- Augmented OpenMP 5.X 'declare mapper' to support all Fortran derived types
- Updated GPU parallelism mapping policy to address performance concerns in some user applications
- Updated GPU unified shared memory software architecture to take advantage of supported hardware
- Full tuned support for Nvidia Grace Hopper
- Initial support for AMD MI300
- Experimental support for Sanitizers on AMD GPUs (C & C++ Only)
User and Application Impact:
----------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
-----------
CCE-17.0.0
- For SLES/COS based systems, the linking GCC library has been changed to use the OS Distributed version available through SLES channels
- For SLES/COS based systems, shared libraries built by prior CCE versions must be rebuilt with CCE 17
- For RHEL based systems, we continue to use gcc-toolset provided by Red Hat, updating to gcc-toolset-12 as our baseline
- For RHEL based systems, shared libraries built by prior CCE versions do not need to be rebuilt
Issues or Bugs Resolved:
------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
CAST-34955 CCE17 Regression with OpenMP Offload and Fortran Abstract Type
CAST-34862 fortran openmp default map(first private:scalar) disables reduction function
CAST-34547 roc-obj doesn't work with executable built with cce
CAST-34528 fortran arrays dimensioned 0 with omp target offload -runtime error
CAST-34517 Cray compiler segfaults on ARCHER2 when default IPA enabled
CAST-33381 ONETEP reaises a CCE internal compiler error (segfault)
CAST-31166 Cray Fortran compiler bug - Finalization
-----------
CCE-17.0.0
CAST-34375 ftn openmp target offload gets Internal Compiler Error
CAST-34321 GPU-RDC flag breaks ROCm/5.7.0 + CCE/16.0.1 interoperability
CAST-34267 Cray Clang 16 compiler crash when on Kokkos sources
CAST-34235 internal functions of elemental function are not compiling
CAST-34208 internal subprogram to ELEMENTAL subprogram
CAST-34118 ftn bounds checking -R b causes internal compiler error
CAST-34114 ftn openmp target offload gets Internal Compiler Error
CAST-33882 CrayFTN prevent reproductible compilation
CAST-33799 CCE internal compiler error with miniweather application
CAST-33775 ftn-7206 warning in cce/16.0.0
CAST-33773 INTERNAL COMPILER ERROR with cce 16.0.0
CAST-33711 SEGV from coarray test
CAST-33524 Cray EX - compile error with cce16 and performance issue with cce15.0.1 when using declare target for nested subroutines
CAST-33363 CCE 15.0.1 Very slow to compile lib
CAST-33343 Crayftn Unlimited Polymorphic Assumed Rank Argument
CAST-33342 Crayftn error with pure procedures in submodule
CAST-33324 Fortran with ACC fails on AMD GPU
CAST-33319 incorrect string lengths result after string arrays are initialized with array constructors of different lengths
CAST-33306 `select type` statement fails to identify argument as a `character` type
CAST-32911 needs -O0 to avoid a ftn internal compiler error
CAST-32887 cray-ftn bug with renaming derived type in use statement
CAST-32694 Unified memory does not work with omp declare target directive. Fortran compiler
CAST-32531 ICE using Cray compiler
CAST-32312 CCE Fortran+OpenMP Runtime Error on Crusher (Invalid dope vector)
CAST-31898 Cray Fortran + OpenACC bug
CAST-31821 MFEM: local memory exceeded CCE 15.0.0, ROCm 5.4.0 -O0
CAST-31788 ftn openmp target link directive runtime errors
CAST-31763 Add consistent debug flag behavior across CCE C++ and CCE Fortran compilers
CAST-31636 cray cce compiler OMPT support
CAST-31447 ftn openmp target link directive runtime errors
CAST-31375 Segfault on subroutine exit after omp target exit data clause
CAST-31156 Several problems with Cray CCE compiler version 14.0.3 and GFS weather code
CAST-30542 Cray fortran compiler error on access to module variable for unified memory
CAST-28426 CrayFTN build time issue - demonstrator of pathological behaviour
CAST-22740 Cray Fortran fails to catch bad G format spec
Product and OS Dependencies:
----------------------------
This CCE release is supported on
- HPE Cray supercomputer systems running CSM with COS 2.5.X (SLES15 SP4) or COS 23.11 (SLES15 SP5)
- HPE Cray supercomputer systems running HPCM with COS 2.5.X (SLES15 SP4) or COS 23.11 (SLES15 SP5)
- HPE Cray supercomputer systems running HPCM with RHEL 8.7
- HPE Apollo 2000 Gen10Plus systems running RHEL 8.7
- AMD GPU OpenMP and OpenACC offloading support requires at least ROCm 6.0 (AMD GPU HIP offloading can still use ROCm 5.x)
Notes and Limitations:
----------------------
- The OpenACC type `acc_handle_kind` has been changed to be of size 8 bytes (instead of 4 bytes). This will require recompilation of any codes that use this type.
- Linking Fortran applications may fail with an error message of the form "hidden symbol `<SYMBOL>' in <LIB> is referenced by DSO".
If this error message is seen, it can usually be worked around by adding '-lgcc_s' to your link line. This is known to affect use
of craypat and cray-parallel-netcdf but may be seen without them as well.
- There is a known bug in the CUDA header `cuda_fp16.h` detailed in the upstream issue https://github.com/llvm/llvm-project/issues/60296.
Programs compiling with `-fopenmp` must use CUDA 12.1.1 or later to avoid this bug, or manually change their header as described in the issue.
Documentation:
--------------
- S-5212 Cray Compiling Environment Release Overview (17.0)
- S-3901 Cray Fortran Reference Manual
- S-2179 Cray C and C++ Quick Reference
- Basic man pages: crayftn(1), craycc(1), crayCC(1), intro_openmp()
- Please see https://clang.llvm.org/docs/UsersManual.html or use the -help command line option for
more information on using Clang
Modulefile:
-----------
The following will load the modules necessary to use CCE:
module load PrgEnv-cray
The following will switch to x.y.z version of CCE:
module swap cce cce/x.y.z
Installation instructions:
--------------------------
rpm -ivh cce-17.0.1-202402021627.99c2eadd92a56-3.sles15sp4.x86_64.rpm.x86_64.rpm
The following script will set CCE version 17.0.1 default:
/opt/cray/pe/admin-pe/set_default_files/set_default_cce_17.0.1
License:
--------
Except for the third party components and software licensed by HPE through
proprietary agreements, components, files or programs contained within this
package or product are Copyright -2023 Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/cce/17.0.1/ATTRIBUTIONS_17.0.txt
Cray PE DL Plugin with Resiliency Support 22.06.1.2:
Release Date:
-------------
June 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 21.04.1:
o Updated DL framework support
o Introduction of resiliency support
o BFloat16 data type support
Supported Configurations:
-------------------------
Shasta/EX: CPU and AMD GPU support with PrgEnv-gnu
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10.2
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
For info regarding use of the new resiliency feature, see the examples
and README.md in ${CRAYPE_ML_PLUGIN_BASEDIR}/examples/resiliency
Sample Installation instructions:
---------------------------------
Note: Package must be installed or sym linked in the default /opt/cray/pe path.,
ie: use of the --prefix flag for rpm is not supported.
rpm -ivh craype-dl-plugin-ftr-22.06.1.2*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
Shasta/EX:
module load craype-dl-plugin-ftr/22.06.1.2
Certain components, files or programs contained within this package or
product are Copyright 2017-2022 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 21.02.1.3:
Release Date:
-------------
February 2021
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 20.10.1:
o Bug fixes for CUDA 11 support and pip install of include source dist
o Initial CUDA 11.1 support
o Updated DL framework support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC CLE 7: CPU and GPU support
Shasta: CPU-only support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.4
* PyTorch v1.7.1
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-21.02.1.3*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/Shasta:
module load craype-dl-plugin-py3/21.02.1.3
CS:
module load craype-dl-plugin-py3/openmpi/21.02.1.3
module load craype-dl-plugin-py3/mvapich/21.02.1.3
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.06.1.2:
Release Date:
-------------
June 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.04.1.1:
o Updated DL framework support
o BFloat16 data type support
o ROCm 5.x support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC CLE 7: CPU and GPU support
Shasta: CPU-only support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.06.1.2*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/Shasta:
module load craype-dl-plugin-py3/22.06.1.2
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.08.1:
Release Date:
-------------
August 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.06.1.2:
o PyTorch Apex Amp Optimization Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.08.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.08.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.09.1:
Release Date:
-------------
September 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.08.1:
o Keras callbacks bug fixes
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.09.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.09.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.12.1:
Release Date:
-------------
December 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.09.1:
o TF v2.9 Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.9
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.12.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.12.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 23.09.1:
Release Date:
-------------
September 2023
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.12.1:
o ROCm 5.3.x support
o TF v2.11 Support
o PT v1.12 Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support for both AMD and NVIDIA
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.11
* PyTorch v1.12
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using TensorFlow, if you see an exception error related to apply_gradients() then
downgrade your keras installation version in your python environment.
pip install keras==2.10.0
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-23.09.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/23.09.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
HPE Cray DL Plugin 24.03.1:
Release Date:
-------------
March 2024
Purpose:
--------
The following changes have been made since HPE Cray DL Plugin 23.09.1:
o AMD MI300A APU support
o TensorFlow v2.15 support
o PyTorch v2.0.1 support
o ROCm 6.0.x support
o Cray-mpich 8.1.27 support
o GCC 12.2.0 support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support for both AMD and NVIDIA
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.15
* PyTorch v2.0.1
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using TensorFlow, if you see an exception error related to apply_gradients() then
downgrade your keras installation version in your python environment.
pip install keras==2.10.0
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-24.03.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/24.03.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
CrayPE Targets 1.13.0
================================================================================
Release Date:
--------------------------------------------------------------------------------
March, 2023
Purpose:
--------------------------------------------------------------------------------
Adds modulefiles for targets - craype-accel-amd-gfx942
Bugs fixed in this release:
--------------------------------------------------------------------------------
None
Dependencies:
--------------------------------------------------------------------------------
The CrayPE Targets 1.13.0 release is dependent on the following
software products:
CrayPE
set_default_3
lmod_scripts
Documentation:
--------------------------------------------------------------------------------
See manpages for cc, CC, ftn, intro_craype-api, intro_hugepages and pkg-config
See section 2.6 Using Targeting Modules of the Cray Programming Environment
User's Guide (S-2529-116)
See http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh craype-targets-ex-1.13.0-202402062025.1d638354624fb-0.ex.noarch.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_craype-targets_1.13.0-tcl
/opt/cray/pe/admin-pe/set_default_files/set_default_craype-targets_1.13.0-lua
Product description:
--------------------------------------------------------------------------------
CrayPE Targets contains the modules for CrayPE targets.
TCL Modules - craype-accel-host craype-accel-amd-gfx90a craype-accel-amd-gfx908 craype-accel-amd-gfx940 craype-accel-amd-gfx942 craype-accel-nvidia70 craype-accel-nvidia80 craype-accel-nvidia90 craype-hugepages2M craype-hugepages4M craype-hugepages8M craype-hugepages16M craype-hugepages32M craype-hugepages64M craype-hugepages128M craype-hugepages256M craype-hugepages512M craype-hugepages1G craype-hugepages2G craype-arm-grace craype-x86-milan craype-x86-milan-x craype-x86-rome craype-x86-spr craype-x86-spr-hbm craype-x86-trento craype-x86-genoa craype-network-none craype-network-ofi craype-network-ucx
Lmod Modules - craype-accel-host craype-accel-amd-gfx90a craype-accel-amd-gfx908 craype-accel-amd-gfx940 craype-accel-amd-gfx942 craype-accel-nvidia70 craype-accel-nvidia80 craype-accel-nvidia90 craype-hugepages2M craype-hugepages4M craype-hugepages8M craype-hugepages16M craype-hugepages32M craype-hugepages64M craype-hugepages128M craype-hugepages256M craype-hugepages512M craype-hugepages1G craype-hugepages2G craype-arm-grace craype-x86-milan craype-x86-milan-x craype-x86-rome craype-x86-spr craype-x86-spr-hbm craype-x86-trento craype-x86-genoa craype-network-none craype-network-ofi craype-network-ucx
Copyright 2013-2023 Hewlett Packard Enterprise Development LP
CDEUX driver (craype) 2.7.31 - 11
==============
Release Date:
--------------
March 2024
Purpose:
--------
The Cray Development Environment User eXperience driver (CDEUX) (craype)
provides a common set of drivers for all CPE programming environments,
notably the C/C++ (cc/CC) and Fortran (ftn) languages.
New CDEUX driver (craype) features for HPE Cray EX and Apollo systems:
- Support for CCE using distribution GCC toolchain installation.
- Improved support for AMD ROCm installations.
- Improved AARCH64 support.
Key Changes and Bugs Closed:
----------------------------
- CDEUX driver (craype) 2.7.31 has fixed issues regarding OpenMP
Product and OS Dependencies:
----------------------------
The CDEUX driver (craype) 2.7.31 release is supported on the following
HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Documentation:
---------------
See manpages for cc, CC, ftn, and intro_craype-api.
See the pkg-config manpage or
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Product description:
--------------------
CrayPE contains drivers, cc, CC, and ftn to compile for all CPE
Programming Environments.
Certain components, files or programs contained within this package or product
are Copyright 2013-2023 Hewlett Packard Enterprise Development LP.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright 2023 Hewlett Packard
Enterprise Development LP.
Craypkg-gen 1.3.32
==============
Release date
------------
February 2024
Purpose
-------
Bug fix release.
Bugs fixed in this release
--------------------------
- Preserve build numbers in ROCm versions
- Exclude hip paths in the ROCm pkg-config files for ROCm >= 5.7.0
- Change -D_HIP_PLATFORM_HCC_ to -D__HIP_PLATFORM_AMD__ in pkg-config files for ROCm >= 5.7.0
- Utilize Jinja2 templating for ROCm pkg-config file
- Update logic to generate modulefiles for new file tree in Intel OneAPI 2024.0
- Update the Nvidia HPC SDK compiler compatibility version for aarch64
- Remove non-existent CUPTI path from Nvidia HPC SDK modulefile
Dependencies
------------
The craypkg-gen 1.3.32 release is supported on the following HPE
Cray systems:
- Cray EX systems SLE 15.0 or later and RHEL 8.0 or later.
Driver support for integrating Third Party C, C++, and Fortran libraries
through .pc files using pkg-config is used in CrayPE 2.x and later.
Limitation
----------
- Library dependencies for static libraries are not added to the .pc files
for keywords Requires.private and Libs.private. A warning is issued by
the craypkg-gen tool to advise users to add this information to the
libraries .pc files.
- RPM limits packages to 4GB
Documentation
-------------
Man pages for craypkg-gen are found by executing `module load craypkg-gen`
and then `man craypkg-gen`. See
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Examples for creating modulefiles for Intel, PGI and Python are included in
the craypkg-gen ‘doc’ directory:
/opt/cray/craypkg-gen/1.3.32/doc/intel_example.txt
/opt/cray/craypkg-gen/1.3.32/doc/pgi_example.txt
/opt/cray/craypkg-gen/1.3.32/doc/python_example.txt
Example:
- As an example, the Intel 16.0.3.210 compiler was recently released.
After installing the compiler the administrator creates a modulefile for
this release by executing the following commands:
# module load craypkg-gen
# craypkg-gen -m /opt/intel/compilers_and_libraries_2016.3.210
This version of the Intel compiler is made default by executing the
command:
# /opt/admin-pe/set_default_craypkg/set_default_intel_16.0.3.210
Installation instructions
-------------------------
rpm -ivh craypkg-gen-1.3.32-1.3.32-1.sles15sp4.noarch.rpm
To change the product version to default after installation:
/opt/cray/pe/admin-pe/set_default_files/set_default_craypkg-gen_1.3.32
Product description
-------------------
The craypkg-gen 1.3.32 utility provides the system administrator
a tool to integrate third party software with the Cray software stack.
Craypkg-gen assists with integration by creating .pc files for C, C++, and
Fortran libraries, pkg-config enabled modulefiles and RPMs.
The workflow for using craypkg-gen is
1) Build the open source software
2) Create .pc files for libraries
3) Create pkg-config enabled modulefiles
4) Customize pkg-config and modulefiles if needed
Certain components, files or programs contained within this package or product
are Copyright 2013-2024 Hewlett Packard Enterprise Development LP.
Cray DSMML 0.3.0:
==========================
Release Date:
-------------
March, 2024
Product Information:
--------------------
DSMML is a stand-alone memory management library for maintaining distributed
shared symmetric memory heaps for top level PGAS languages and libraries like
Coarray Fortran, UPC, and OpenSHMEM. DSMML allows user libraries to create
multiple symmetric heaps and share information with other libraries. Through
DSMML, we could extract interoperability between PGAS programming models.
Announcements, release informations, supported environments, and backward
compatibility informations about this product can be viewed in the following
location: https://pe-cray.github.io/cray-dsmml/
Purpose:
--------
Cray DSMML 0.3.0 supports the following features:
- Create symmetric heap segments for being used by PGAS languages and
libraries
- Allows options to allocate, deallocate memory buffers from previously
created symmetric heap segments
- Enables sharing information about the symmetric heap segments between
different programming models
- Added support for shared symmetric heap (SSHEAP) that can be used to
provide enhanced SMP data transport in user-libraries like SHMEM, CAF, UPC
The following features are fixed in Cray DSMML 0.3.0 when compared
to Cray DSMML 0.2.2:
- Add support for the library for AARCH64 processor based platforms
- Internal bug fixes involving hugepage cleanups
Refer intro_dsmml(3) manpage for more information on the supported features
and syntax for different routines
Product and OS Dependencies:
----------------------------
The Cray DSMML 0.3.0 release is supported on the following
Cray systems:
* HPE Cray EX systems with CLE
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.3 |
----------------------------+---------------------+
One or more compilers:
* AOCC 2.2 or later
* CCE 9.1 or later
* GNU 9.1 or later
* Intel 19.0 or later
* Nvidia 20.7 or later
Documentation:
--------------
Use the Cray DSMML man pages for more information on the library
functions and use intro_dsmml(3) man page for general information.
Use https://pe-cray.github.io/whitepapers/ for access to different
Cray DSMML specific whitepapers
Announcements, release informations, supported environments, and backward
compatibility informations about this product can be viewed in the following
location: https://pe-cray.github.io/cray-dsmml/
Modulefile:
-----------
module load cray-dsmml/0.3.0
Certain components, files or programs contained within this package or
product have the following Copyright:
Copyright 2018-2021 Hewlett Packard Enterprise Development LP.
FFTW 3.3.10.7
============
Release Date:
-------------
March 2024
Purpose:
--------
This Cray FFTW 3.3.10.7 release is supported on Cray EX (formerly
Shasta) systems. FFTW is supported on the host CPU but not on the
accelerator of Cray systems.
The Cray FFTW 3.3.10.7 release provides the following:
- Compiler and dependency updates
See the Product and OS Dependencies section for details.
Product and OS Dependencies:
----------------------------
The Cray FFTW 3.3.10.7 release is supported on the following Cray systems:
Cray EX Systems with SLES 15 SP4 or later based OS
The FFTW 3.3.10.7 release requires the following software products:
craype 2.7.5 or later
MPT 8.0 or later
One or more of the following major compiler versions:
CCE 17
GCC 12, 13
AOCC 4.1
Intel 2024 or later
Notes and Limitations:
----------------------
Starting with cray-fftw/3.3.6.1 the fftw module has been renamed from fftw
to cray-fftw.
Documentation:
--------------
http://www.fftw.org/index.html#documentation
See the intro_fftw3 manual page for additional information.
Modulefile:
-----------
module load cray-fftw
Installation
------------
To install the rpm, execute:
rpm -ivh cray-fftw-3.3.10.7-202402071520.4d6d346567dfd-1.shasta.sles15sp4.x86_64.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_fftw_3.3.10.7
Certain components, files or programs contained within this package or
product are Copyright 2011-2024 Hewlett Packard Enterprise Development LP
Cray LibSci 24.03.0
=====================
Release Date:
-------------
March 2024
Purpose:
--------
Cray LibSci 24.03.0 provides scientific libraries for Cray
HPC systems. Cray LibSci is supported on the host CPU but
not on the accelerator of these systems.
The Cray LibSci 24.03.0 release provides the following:
* support for RHEL gcc-toolset-{10,11,12}
* support for Nvidia HPC SDK 23.11 on aarch64
Cray LibSci 24.03.0 includes the following versions of publicly
available libraries:
* LAPACK 3.10.1 - For further information, see
http://www.netlib.org/lapack
* ScaLAPACK 2.2.0 - (Scalable LAPACK) For further information, see
http://www.netlib.org/scalapack.
* QDWH 2.0.0 KSVD 1.0.0 - Polar decomposition and SVD packages, see
https://github.com/ecrc/qdwh
https://github.com/ecrc/ksvd
Key Changes and Bugs Closed:
----------------------------
Changes in Cray LibSci 24.03.0
* CAST-34956 - Lua module environment variable issue
Changed in Cray LibSci 23.12.5
* CAST-33777 - LibSci mutex contention issue
* Removed support for Intel KNL
Product and OS Dependencies:
----------------------------
The Cray LibSci 24.03.0 release requires the following:
* SLES 15 SP4/SP5 or RHEL 8.8
* cray-mpich/8.1.0 or later
Supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Cray XD systems with CLE
One or more of the following compiler major versions:
* CCE 17.x (SLES)
* CCE 16.x or later (RHEL)
* GCC 12.3 or later (SLES)
* gcc-toolset-10.3/11.2/12.2 (RHEL)
* AOCC 4.1.0 or later
* AMD ROCm 5.0.0 or later
* Intel 2023.2.0 or later
* Nvidia 23.11 or later (aarch64) and 23.3 or later (x86_64)
Notes and Limitations:
----------------------
OMP threaded versions:
The CrayPE 2.1.2 and later releases add support for link line generation
for the multi-threaded versions of the libsci library based on the OpenMP
flags the user specifies for each compiler:
* CCE by default links to the OpenMP LibSci library. CrayPE will link in
the serial version of LibSci when the CCE flag -hnoomp is used.
* GNU by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the GNU flag -fopenmp is used.
* INTEL by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the INTEL flag -qopenmp is used.
Stack size limit:
The 'ulimit -s unlimited' system stack size setting is required for
cray-libsci on all supported Cray platforms. This is typically set as a
system default by admin but may otherwise need to be set at runtime.
Non-default dynamic linking:
When using a non-default version of cray-libsci with dynamic linking
users should set at runtime and at linktime after loading the desired
cray-libsci module--or set the equivalent for their linux shell:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci man page for additional information.
See the csmlversion man page for information to display version information
on the currently loaded scientific libraries.
Modulefile:
-----------
module load cray-libsci
Installation instructions:
--------------------------
LibSci is now packaged into separate compiler specific RPMs to allow
rpmbuild to correctly include compiler dependencies.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
/opt/cray/libsci/24.03.0/ATTRIBUTIONS
Cray LibSci_ACC 24.03.1
=======================
Release Date:
-------------
March 2024
Purpose:
--------
The Cray LibSci_ACC 24.03.1 release provides accelerated versions of
scientific libraries for HPE Cray systems with accelerators.
Product and OS Dependencies:
----------------------------
Cray LibSci_ACC 24.03.1 requires the following platform:
HPE Cray EX system with AMD MI250X
HPE Cray EX system with AMD MI300A
HPE Cray EX system with Nvidia Hopper
Cray LibSci_ACC 24.03.1 supports the following OS:
RHEL 8.8 x86_64
SLES 15 SP4 x86_64
SLES 15 SP5 x86_64 & aarch64
Notes and Limitations:
----------------------
LAPACK workspace calculations from other implementations or hard-coded
values may not be compatible with libsci_acc. Workspaces reported from
a query to the corresponding libsci_acc function should be used.
The current supported usecase is a one-rank to one-gpu mapping for apps.
Performance improvements can be achieved in programs calling libsci_acc
subroutines by using pinned memory. See the intro_libsci_acc man page for
details.
Use of the aprun option "-cc none" or the srun option "--cpu_bind=none"
to disable core affinity is strongly suggested to maintain performance.
When using non-default cray-libsci_acc versions users should set equivalent
at runtime and linktime after loading the desired cray-libsci_acc module:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci_acc man page for additional information.
Modulefile:
-----------
module load cray-libsci_acc
Sample Installation instructions:
---------------------------------
RPM install command:
rpm -ivh cray-libsci-acc-24.03.1-*.##ARCH##.rpm
The "*" in the install command represents compiler version combinations.
To make this the default version:
##PREFIX##/admin-pe/set_default_files/set_default_libsci_acc_24.03.1
License:
--------
Except for the third party modules and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright -2023 Hewlett
Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in
this package are detailed in the file:
##PREFIX##/libsci_acc/24.03.1/ATTRIBUTIONS_libsci_acc_24.03.1.txt
Cray Lmod 8.7.32
================================================================================
Release Date:
--------------------------------------------------------------------------------
March, 2024
Purpose:
--------------------------------------------------------------------------------
Provide a Lmod 8.7.32 package for CPE customers
Bugs fixed in this release:
--------------------------------------------------------------------------------
N/A
Documentation:
--------------------------------------------------------------------------------
See CPE's Installation and User Guides for documentation.
Also see Lmod's official website: https://lmod.readthedocs.io
Dependencies:
--------------------------------------------------------------------------------
The Lmod Scripts 8.7.32 release is dependent on the following
software:
bc
sed
lua >= 5.3
lua-luaposix
lua-luafilesystem
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh cray-lmod-8.7.32-1.sles15sp4.x86_64.rpm
To make this the default version, update the symbolic "lmod" link
to the desired Lmod version directory.
Lmod Scripts 3.2.1
================================================================================
Release Date:
--------------------------------------------------------------------------------
December, 2023
Purpose:
--------------------------------------------------------------------------------
Updates package license.
Bugs fixed in this release:
--------------------------------------------------------------------------------
Documentation:
--------------------------------------------------------------------------------
See system Installation and User Guides for documentation
Dependencies:
--------------------------------------------------------------------------------
The Lmod Scripts 3.2.1 release is dependent on the following
software:
set_default_3
Lua
Lmod
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh lmod_scripts-3.2.1-202311141921.212030d47edcb-0.noarch.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_lmod_scripts_3.2.1
Product description:
--------------------------------------------------------------------------------
Contains Lua scripts which support the Cray implementation of Lmod
Copyright 2020-2023 Hewlett Packard Enterprise Development LP
modules 3.2.11.7
==================
Release Date:
-------------
Dec, 2023
Purpose:
--------
- Package dependency update
Documentation:
---------------
See man-pages for module and modulefile.
Cray MPICH 8.1.29:
=======================================
Release Date:
-------------
February 21, 2024
Purpose:
--------
Cray MPICH 8.1.29 is based upon ANL MPICH 3.4a2 with support for libfabric
and is optimized for the Cray Programming Environment.
Major Differences Cray MPICH 8.1.29 from the XC Cray MPICH include:
- Uses the new ANL MPICH CH4 code path and libfabric for network
support.
- Does not support -default64 mode for Fortran
- Does not support C++ language bindings
New Cray MPICH features for HPE Cray EX and Apollo systems:
- Starting from the 8.1.26 release, Cray MPICH supports the Intel Sapphire Rapids CPU HBM
processor architecture.
- On systems with AMD GPUs, Cray MPICH 8.1.26 supports all ROCm
versions starting from ROCm 5.0, including the latest ROCm 5.5.0
release.
The Cray MPICH 8.1.25 release and prior versions of
Cray MPICH are only compatible with ROCm versions up to (and
including) the ROCm 5.4.0 release.
- Cray MPICH uses the libfabric "verbs;ofi_rxm" provider by default.
This is the supported and optimized OFI libfabric provider for
Slingshot-10 and Apollo systems.
- Cray MPICH offers support for multiple NICs per node. Starting with
version 8.0.8, by default Cray MPICH will use all available NICs on
a node. Several rank-to-NIC assignment policies are supported. For
details on choosing a policy for assigning ranks to NICS, or for
selecting a subset of available NICs, please see the following
environment variables documented in the mpi man page.
MPICH_OFI_NIC_VERBOSE
MPICH_OFI_NIC_POLICY
MPICH_OFI_NIC_MAPPING
MPICH_OFI_NUM_NICS
- Enhancements to the MPICH_OFI_NIC_POLICY NUMA mode have been added.
Starting with version 8.0.14, if the user selects the NUMA policy,
the NIC closest to the rank is selected. A NIC no longer needs to
reside in the same numa node as the rank. If multiple NICs are
assigned to the same numa node, the local ranks will round-robin
between them. Numa distances are analyzed to select the closest NIC.
- Cray MPICH supports creating a full connection grid during MPI_Init.
By default, OFI connections between ranks are set up on demand. This
allows for optimal performance while minimizing memory requirements.
However, for jobs requiring an all-to-all communication pattern, it
may be beneficial to create all OFI connections in a coordinated
manner at startup. See the MPICH_OFI_STARTUP_CONNECT description in
the mpi man page.
- Cray MPICH supports runtime switching to the UCX netmod starting
with version 8.0.14. To do this load the craype-network-ucx module
and module swap between Cray-MPICH and Cray-MPICH-UCX modules. For
more information including relevant environment variables reference
the intro_mpi man page with the Cray-MPICH-UCX module loaded.
- Lmod support for HPE Cray EX starting with Cray MPICH 8.0.16.
Key Changes and Bugs Closed:
----------------------------
Changes in Cray MPICH 8.1.29
New Features:
- Cray MPICH 8.1.29 for aarch64 is in early stage of support and
has some issues. Setting these variables will help circumvent most
known issues.
export MPICH_SMP_SINGLE_COPY_MODE=CMA
export MPICH_MALLOC_FALLBACK=1
Bugs Fixed:
- CAST-34406 - Improve performance in MPI Allreduce benchmark
- CAST-34872 - Fix seg-fault in extremely large allreduce calls
- CAST-34972 - Use dynamic CXI PID to avoid conflicts
- PE-45873 - Added infrastructure for CXI triggered operations non-blocking collectives
- PE-45874 - Add MPI_Ibcast() using CXI triggered operations
- PE-45876 - Add MPI_Igather() using CXI triggered operations
- PE-45877 - Add MPI_Iscatter() using CXI triggered operations
- PE-45878 - Add MPI_Ireduce() using CXI triggered operations
- PE-45879 - Add MPI_Iallreduce() using CXI triggered operations
- PE-49870 - Support MPI_LONG_DOUBLE with CCE for aarch64
- PE-51028 - Improved debugging to diagnose memcpy issues
- PE-51233 - Build libraries for the nvidia compiler on aarch64
- PE-51326 - Fix window creation using MPI_WIN_FLAVOR_SHARED to avoid running out of fi_mr resources
- PE-51365 - Fix bad initialization of hint striping_unit in MPI-IO
- PE-51445 - Support ROCM 6.0 with Cray-MPICH GTL
- PE-51470 - Update Dynamic Process Management documentation
- PE-51524 - Document PBS options for using Spawn on Slingshot
- PE-51738 - Bugfix for GPU NUMA nodes on MI300 GPUs
Product and OS Dependencies:
----------------------------
The Cray MPICH 8.1.29 release is supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.6 |
----------------------------+---------------------+
cray-pals | >= 1.0.6 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
One or more compilers:
* AMD ROCM 5.0 or later
* AOCC 4.1 or later
* CCE 17.0 or later
* GNU 12.3 or later
* Intel 2022.1 or later
* Nvidia 23.3 or later
Notes and Limitations:
----------------------
Limitations in Cray MPICH 8.1.29:
- Cray MPICH 8.1.29 can support only ~2040 simultaneous MPI
communicators. This limit is less the XC Cray MPICH limit of
~4090 simultaneous communicators. Cray intends to raise the
limit in a future release of Cray MPICH for Shasta to at least
the XC limit.
Documentation:
--------------
For more information see the intro_mpi man page.
Modulefile:
-----------
module load cray-mpich/8.1.29
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright -2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/mpich/8.1.29/ATTRIBUTIONS
Copyright -2023 Hewlett Packard Enterprise Development LP
Copyright 2022-2024 Hewlett Packard Enterprise Development LP
Product Release Information
================================================================================
Product identification:
-----------------------
HPE Cray MPIxlate: version 1.0.4 (built on 2024-02-16T14:55:44Z from 9bd5b45)
Purpose:
--------
HPE Cray MPIxlate enables applications compiled using an MPI library that
is not binary compatible with HPE Cray MPI, to be run without recompilation
on supported HPE platforms.
Warnings & Limitations:
-----------------------
- See NOTES in mpixlate(1)
Changes from previous release:
------------------------------
- New features:
* None
- Defects Fixed:
* None
Platform, Operating System and Product dependencies:
----------------------------------------------------
Supported HPE platforms:
- aarch64 and x86_64 HPE Cray EX systems with HPCM or CSM
Supported Operating Systems:
- SLES 15 SP4, SP5
- COS 2.5, 23.11
Product dependencies:
-----------------+----------
cray-mpich[-ucx] | >= 8.1.29
-----------------+----------
Documentation:
--------------
mpixlate(1)
Environment Modules
-------------------
module load cray-mpixlate/1.0.4
================================================================================
Cray PALS 1.3.2:
================
Release Date:
-------------
Nov 2023
Purpose:
--------
Cray Parallel Application Launch Service (PALS) is an application launcher for
Cray PE applications.
Key Changes:
------------
Add host2xname tool
Allow configuring the PMI port range
Add more examples in manpages
Add SLES 15 SP5 build
Add SLES 15 SP4 and SP5 aarch64 build
Add PMI and PMI2 as --pmi options for mpiexec
Periodically ping applications
Add the ability to allocate additional VNIs for special services like DAOS
Add mpiexec --disable-rdzv-get option to disable rendezvous gets in the NICs
Enable --cpu-bind and --gpu-bind=verbose to work with --rankfile
Add PALSD_FANOUT env var for palsd to set fanout tree width
Add ability for simultaneous mpiexecs to divide up PBS job slots
Add RHEL 8.8 build
Add support for AMD & NVIDIA GPU visibility
Change the wire format for PMIx data to make PMIx_Fence more efficient
Increase core file size limit on RHEL systems
Add retries to SPIRE auth token call
Install aprun and mpirun as symlinks
Use position-independent executables
Change F_CXI_COLL_* -> FI_CXI_COLL_* environment variables
Configure Instant On URL on Shasta
Move binaries and libraries to /opt/cray
Change pmi_example to print NIC distances for all ranks
Make PALS command's nodelists consistent
Set ATOM_SOCKET by default
Improve vnid request error reporting
Launch with PMIx support automatically fetch NIC information
Reduce default fanout to 32 nodes
Require PMIx version 4.2.4 or newer
Send the job-step name to vnid when creating a job-step VNI
Include pid in palsd log messages
Raise maximum number of ZMQ sockets to 4096
Munge auth no longer used for some internal RPCs with trusted connections
Increase default send timeout to 8s
Fix errors found with clang scan-build
Fix nohup mpiexec behavior
Improve cleanup when launch RPC fails
Implement the verbose option for GPU binding
Fix support for the cycle keyword with the mpiexec --rankfile option
Fix error checking for GPU binding options
Better handle mpiexec disconnects
Fix segfault on PMIx RPC for missing app
Fix spawn-related memory leaks
Fix misleading warning in line buffer mode
Remove SLES 15 SP3 build
Deprecate PALS_FANOUT and APRUN_FANOUT environment variables
Install cray-pals RPM
Documentation:
--------------
For more information see aprun, mpiexec, palscmd, palsctrl, palsig, and palstat man page.
Modulefile:
-----------
module load cray-pals
Installation instructions:
--------------------------
rpm -ihv cray-pals-<version>-<date>.el8.x86_64.rpm
rpm -ihv cray-palsd-<version>-<date>.el8.x86_64.rpm
Copyright 2022-2023 Hewlett Packard Enterprise Development LP
cray-papi
=================
Release date:
-------------
December 2023
Package:
--------
cray-papi-7.0.1.2-202310131612.70757f986edac-4 x86_64
Purpose:
--------
New version of papi 7.0.1.2 release 4
Documentation:
--------------
Overview: https://github.com/icl-utk-edu/papi
Web Site: https://icl.utk.edu/papi
Product description:
--------------------
PAPI aims to provide the tool designer and application engineer with a
consistent interface and methodology for use of the performance counter
hardware found in most major microprocessors. PAPI enables software
engineers to see, in near real time, the relation between software
performance and processor events.
Dependencies:
-------------
For a list of software used when validating this version of
PAPI on Cray and HPE systems, see the HPE Cray Programming
Environment release announcements.
Copyright 2015-2017,2019-2023 Hewlett Packard Enterprise Development LP
Perftools 23.12.0
===============
Release Date: December, 2023
Purpose:
========
This is a feature and bugfix release for the following systems:
- HPE Cray EX and HPE Cray Supercomputer Systems with HPCM
- HPE Cray EX and HPE Cray Supercomputer Systems with CSM
- HPE Apollo 2000 Gen 10 Plus Systems (x86)
Key enhancements or changes from the previous release:
======================================================
o Upgrade to DWARF 0.8.0
o Upgrade to PAPI 7.0.1.2
- Upstream bugfixes
- Upstream CUDA component version locked from prior release and renamed cuda_7011
- Upstream ROCm component version locked from release 7.0.0.1 and renamed rocm_7001
- Support for Nvidia Grace processor
o Add support for ROCm 5.7.0
o Add support for CUDA 12.2
o Add support for Nvidia H100 GPUs
o Update ROCm trace wrappers to support up to ROCm 5.7
o Add support for PAT_region_push/pop in full trace mode
o Add support for profiling Python 3.11
o Add a DSMML tracegroup
o Improved counter groups and derived metrics for AMD Genoa processors
o Add support for OMPT target callbacks with CCE's OpenMP runtime, enabled
via the CRAYPAT_OMP_TOOL environmnet variable (NOTE: this is primarily intented
for use with pat_run. When instrumenting executables with CCE, proprietary
tracepoints are used in favor of OMPT. To use OMPT instead, '-fno-instrument-openmp'
must be added to compilation.)
o Perftools and its libraries are now compiled using the OS native
gcc and no longer depend on cray-gcc
o Set PAT_RT_ACC_MODEL_NAME in the runtime environment to manually specify
the accelerator model name.
Bugs Fixed:
===========
o Unexpected callers appearing in the calltree when using perftools-lite-loops
o An inappropriate assertion failure when MPI_Get_count returns MPI_UNDEFINED
o A runtime failure when incompatible versions of HIP and pat_run are used
o An issue preventing lambda and template function resolution using Perftools
with Caliper
o Segfaults when managing a hipStream with singletons
o Segfaults when tracing OpenACC codes
o Program hangs at dlopen using pat_run on certain Python codes
o Program hangs using the '-gheap' trace option on RHEL systems
o Incorrect number of PEs/Node reported when using PBS and PALS
o Incorrect launching of Apprentice2 when using reveal launcher
Notes:
======
o This release of Perftools only supports ROCm 5.7.0; SLES 15 support is limited to SP5
o This release of Perftools only supports CUDA 12.0 and later
Known Issues:
=============
o Fortran applications compiled against CCE 16 may fail during instrumentation
with the following error message: "hidden symbol `<SYMBOL>' in <LIB> is
referenced by DSO". To work around this error specify the following during
the pat_build invocation '-Dlink-instr=-lgcc_s'.
o When using OMPT with CCE's openmp runtime, to and from device transfers and
kernel executions originating from a single construct are treated as separate
events. For example, 'target map(tofrom:)' will accumulate 3 events each
time it is encountered.
o Nvidia device drivers do not correctly populate system files used by Perftools
for the Grace-Hopper superchip, impacting driver versions up to: 545.23.06.
Reports generated from applications run on these devices may list the accelerator
device as "Unknown". Additionally, usage of accelerator performance counter groups
is not supported when the device is unresolved. To work around this issue set the
following runtime environment variable prior to running the experiment
'PAT_RT_ACC_MODEL_NAME="GH200 120GB"'.
o Tracing experiments with accelerator performance counters and all sampling experiments
may produce unexpected behavior on Grace-Hopper superchips. At this time, there is
no work around this issue and the capabilities should be treated as experimental.
o Sampling experiments on aarch64 systems lead to fatal errors if
PAT_RT_CALLSTACK_MODE="unwind" is used. The setting is disabled for these situations
and PAT_RT_CALLSTACK_MODE="frames" is used instead.
o The Intel OneAPI Fortran compiler does not correctly produce the ELF .debug_pubnames or
.debug_names section in object files when the "-debug pubnames" option is specified
(as required when the perftools module is loaded). This prevents pat_build from collecting
the necessary information that supports tracing user-defined entry points. Specify the
"-Ddebug-names=n" option to pat_build so pat_build can collect the information from the
ELF .debug_info section.
o The Intel OneAPI Fortran compiler does not accurately represent formal parameters to a
subprogram in some cases. This results in the generation of incorrect intercepts for
user-defined entry points when pat_build -u is specified and results in seg faults when
the instrumented program is executed. Specify the "-Dtrace-force-deref=y" to pat_build
to account for the missing formal parameter information.
Deprecations and Removals:
==========================
o Remove support for PGI programming Environment
o Remove support for following trace groups: craymem, hbw, gni, jemalloc, memkind
Dependencies:
=============
o A PDF reader (such as evince, acroread or okular) is required to use the
app2 online help.
o For a list of software used when validating this version of
Perftools on Cray and HPE systems, see the HPE Cray Programming Environment
release announcements.
Installation instructions:
==========================
For Apollo 80 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 80 (ARM) System (S-8013).
For Apollo 2000 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 2000 Gen10 Plus (x86) System (S-8012).
For HPE Cray EX (Shasta Architecture) systems, refer to the Cray Asynchronous
Installer Guide (S-8003).
Installation of app2 remote client (+ server) on Mac systems:
-------------------------------------------------------------
Apprentice2Installer-23.12.0.dmg
RevealInstaller-23.12.0.dmg
The Cray Apprentice2 and Reveal installers for Mac are included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Mac OS El Capitan through Mojave. Double click on installer
to begin installation. The installer will walk you through the
process for your system.
Installation of app2 remote client (+ server) on Windows 10 systems:
--------------------------------------------------------------------
Apprentice2Installer-23.12.0.exe
The Cray Apprentice2 installer for Windows is included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Windows 10. Double click on installer to begin installation.
The installer will walk you through the process for your system.
Documentation:
==============
See the following man pages:
app2, grid_order, intro_craypat, pat_build, pat_help, pat_info, pat_opts,
pat_report, pat_run, pat_view, reveal
intro_papi
perftools-base, perftools-lite, perftools-preload
accpc, cray_cassini, cray_pm, hwpc, nwpc, papi_counters, uncore
PAPIlicnotices
Search for perftools on https://support.hpe.com to access documentation
on Cray Performance Measurement and Analysis Tools
License:
========
Except for the third party modules discussed below and software licensed
by HPE through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
$CRAYPAT_ROOT/ATTRIBUTIONS_perftools.txt
Cray PMI 6.1.14
==============================================================================
Release Date:
--------------
February 12, 2024
Purpose:
--------
The Cray Process Manager Interface Library provides the interface between the
application launcher and other communication libraries such as MPICH and SHMEM.
Cray PMI 6.1.14 is optimized for the Cray Programming Environment.
New Cray PMI features for HPE Cray EX and Apollo systems:
Key Changes and Bugs Closed:
----------------------------
Starting with Cray-PMI 6.1.0 the cray-pmi and cray-pmi-lib modules have been
combined into a single cray-pmi module. The cray-pmi-lib module will no longer
be installed by new cray-pmi rpms. When the cray-pmi module is loaded, the
pmi libraries will be linked with as-needed flags.
Bug fixes new to Cray PMI 6.1.14 include:
- CAST-34500 - Include unversioned pmi library symlinks in the CPE ld cache
- PE-51739 - Create modern PMI process map code
Product and OS Dependencies:
----------------------------
The Cray PMI 6.1.14 release is supported on the following systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product Dependencies:
----------------------------------------+
cray-pals | >= 1.0.6 |
-----------------------+----------------+
slurm | >= 20.02 |
-----------------------+----------------+
Notes and Limitations:
----------------------
Documentation:
--------------
For more information see the intro_pmi man page.
Modulefile:
-----------
module load cray-pmi/6.1.14
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright 2020-2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/pmi/6.1.14/ATTRIBUTIONS
cray-python 3.11.7
====================
Release Date
------------
March 2024
Purpose
-------
Cray Python is an implementation of the Python programming language for the Cray
Programming Environment. The numpy and scipy modules are configured to call Cray
Libsci routines. The mpi4py module is configured to call Cray MPICH routines.
Cray Python is designed to run Python codes on the compute nodes of an HPE
supercomputer. HPE does not make changes to the Python source or any of its
libraries nor does it plan to make changes in future releases.
The cray-python 3.11.7 release contains
- python-3.11.7
- numpy-1.24.4
- scipy-1.10.1
- mpi4py-3.1.4
- dask-2023.6.1
Product and OS Dependencies
---------------------------
The cray-python 3.11.7 release is supported on
- Cray EX systems running SLES 15 or RHEL 8
Documentation
-------------
https://www.python.org/doc
Changelog
---------
https://docs.python.org/release/3.11.7/whatsnew/changelog.html
Modulefile
----------
module load cray-python/3.11.7
Installation
------------
rpm -ihv cray-python-3.11.7-20240208204342.1cdbf56-1.sles15sp4.x86_64.rpm
To make this the default version, execute
[PREFIX]/admin-pe/set_default_files/set_default_python_3.11.7
Certain components, files, or programs contained within this package are
© Copyright 2021-2024 Hewlett Packard Enterprise Development LP
cray-R 4.3.2
============
Release Date
------------
March 2024
Purpose
-------
The cray-R 4.3.2 release.
cray-R is configured to call cray-libsci routines.
Product and OS Dependencies
---------------------------
The cray-R 4.3.2 release is supported on
- Cray EX systems running SLES 15 or RHEL 8
Documentation
-------------
https://cran.r-project.org/manuals.html
Modulefile
----------
module load cray-R/4.3.2
Installation
------------
rpm -ihv cray-R-4.3.2-20240208204433.1115d98-1.sles15sp4.x86_64.rpm
To make this the default version, execute:
[PREFIX]/admin-pe/set_default_files/set_default_R_4.3.2
Certain components, files or programs contained within this package or product are
Copyright 2018-2024 Hewlett Packard Enterprise Development LP
Cray OpenSHMEMX 11.7.1:
===============================
Release Date:
-------------
Mar, 2024
Purpose:
--------
OpenSHMEM is a Partitioned Global Address Space (PGAS) library interface
specification, which is the culmination of a standardization effort among
many implementers and users of SHMEM programming model. SHMEM has a long
history as a parallel programming model on HPE Cray supercomputer systems.
For the past two decades SHMEM library implementation in HPE Cray systems
evolved through different generations. Cray OpenSHMEMX is a proprietary,
OpenSHMEM specification compliant SHMEM implementation for HPE Cray
product line.
Key Changes:
------------
Major differences in Cray OpenSHMEMX 11.7.1 from
Cray OpenSHMEMX version 11.7.0 includes the following:
- Add support for logging HPE Slingshot NIC counter values
- Add support for using the library on platforms with AARCH64
processors
- Improve performance of shmem_quiet memory ordering operation
- Improve performance for allreduce collective operations when all
PEs are involved in the communication operations
Bugs Closed:
------------
The following bugs are fixed as part of the Cray OpenSHMEMX 11.7.1
release:
- Fix C11 generic APIs for shmem_sync operations
- Fix library usage with GNU and Intel compilers
Product and OS Dependencies:
----------------------------
The Cray OpenSHMEMX 11.7.1 release is supported on the following
Cray systems:
* HPE Cray EX systems with CLE
Product and OS Dependencies by network type:
--------------------------------------------------+
| HPE Slingshot |
----------------------------+---------------------+
craype | >= 2.7.3 |
----------------------------+---------------------+
cray-pals | >= 0.3.5 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
cray-dsmml | >= 0.2.2 |
----------------------------+---------------------+
One or more compilers:
* AOCC 2.2 or later
* CCE 9.1 or later
* GNU 9.1 or later
* Intel 19.0 or later
* Nvidia 20.7 or later
Notes and Limitations:
----------------------
Announcements, release information, supported environments, limitations,
usage models, and backward compatibility informations about this product
can be viewed in the following location:
https://pe-cray.github.io/cray-openshmemx/
Documentation:
--------------
Use the Cray OpenSHMEMX man pages for more information on the library
functions and use intro_shmem(3) man page for general information.
Initial support for Cray OpenSHMEMX man pages are derived from OpenSHMEM
standards specification document
Use https://pe-cray.github.io/whitepapers/ for access to different
Cray OpenSHMEMX specific whitepapers
Modulefile:
-----------
module load cray-openshmemx/11.7.1
License:
--------
Components, files, and programs contained within this package or
product are Copyright Hewlett Packard Enterprise Development LP
CPE Release Notes For SLES 15 SP5
HPE Cray Programming Environment 24.03 Full Release Notes for SLES15 SP5
=========================================================
HPE Cray Compiling Environment (CCE) 17.0.1
===========================================
Release Date:
-------------
March 2024
Purpose:
--------
CCE 17.0.1 provides Fortran, C, and C++ compilers for
HPE Cray supercomputer and HPE Apollo 2000 Gen10Plus systems.
Key Changes and Support with CCE 17.0.1:
----------------------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
- ROCm 6.0 Suppport
- Added OMPT device tracing support to enable profiling
- Tuned support for MI300A
- SLES gcc/13 support
- Address sanitizer support on GPUs in Fortran
-----------
CCE 17.0.0
- LLVM 17 base (Merges up to November 6th 2023)
- Fortran
- Support offload of DO CONCURRENT iterations to accelerators with unified shared memory
- Starting with 17.0, support passing any Fortran contained procedures to C/C++
- Code with both 17.0 and pre-17.0 objects files must be linked with 17.0 ftn
- Added the following Fortran 2023 Features:
- US 01 & 02 Allow much longer statement lines and overall statement length
- US 03 SPLIT and TOKENIZE intrinsics
- US 04 Trig functions that work in degrees
- US 06 selected_logical_kind intrinsic function
- US 07 Additional named constants to specify kinds in ISO_FORTRAN_ENV
- US 10 The at edit descriptor
- US 11 Control over leading zeros in output of real values
- US 15 Simple procedures
- US 20 Reduction specifier for do concurrent
- UK 01 Extend the intrinsic procedure c_f_pointer to allow its pointer result to have specified lower bounds
- A public namelist group can have a private namelist group object.
- Added the following OpenMP 5.1 Features:
- 'error' directive
- 'omp_get_mapped_ptr' routine
- 'omp_display_env' routine
- 'thread_limit' clause on 'target' construct (C/C++ only)
- New initial device number value
- loop construct
- Augmented OpenMP 5.X 'declare mapper' to support all Fortran derived types
- Updated GPU parallelism mapping policy to address performance concerns in some user applications
- Updated GPU unified shared memory software architecture to take advantage of supported hardware
- Full tuned support for Nvidia Grace Hopper
- Initial support for AMD MI300
- Experimental support for Sanitizers on AMD GPUs (C & C++ Only)
User and Application Impact:
----------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
-----------
CCE-17.0.0
- For SLES/COS based systems, the linking GCC library has been changed to use the OS Distributed version available through SLES channels
- For SLES/COS based systems, shared libraries built by prior CCE versions must be rebuilt with CCE 17
- For RHEL based systems, we continue to use gcc-toolset provided by Red Hat, updating to gcc-toolset-12 as our baseline
- For RHEL based systems, shared libraries built by prior CCE versions do not need to be rebuilt
Issues or Bugs Resolved:
------------------------
CCE 17.0.1:
-----------
CCE-17.0.1
CAST-34955 CCE17 Regression with OpenMP Offload and Fortran Abstract Type
CAST-34862 fortran openmp default map(first private:scalar) disables reduction function
CAST-34547 roc-obj doesn't work with executable built with cce
CAST-34528 fortran arrays dimensioned 0 with omp target offload -runtime error
CAST-34517 Cray compiler segfaults on ARCHER2 when default IPA enabled
CAST-33381 ONETEP reaises a CCE internal compiler error (segfault)
CAST-31166 Cray Fortran compiler bug - Finalization
-----------
CCE-17.0.0
CAST-34375 ftn openmp target offload gets Internal Compiler Error
CAST-34321 GPU-RDC flag breaks ROCm/5.7.0 + CCE/16.0.1 interoperability
CAST-34267 Cray Clang 16 compiler crash when on Kokkos sources
CAST-34235 internal functions of elemental function are not compiling
CAST-34208 internal subprogram to ELEMENTAL subprogram
CAST-34118 ftn bounds checking -R b causes internal compiler error
CAST-34114 ftn openmp target offload gets Internal Compiler Error
CAST-33882 CrayFTN prevent reproductible compilation
CAST-33799 CCE internal compiler error with miniweather application
CAST-33775 ftn-7206 warning in cce/16.0.0
CAST-33773 INTERNAL COMPILER ERROR with cce 16.0.0
CAST-33711 SEGV from coarray test
CAST-33524 Cray EX - compile error with cce16 and performance issue with cce15.0.1 when using declare target for nested subroutines
CAST-33363 CCE 15.0.1 Very slow to compile lib
CAST-33343 Crayftn Unlimited Polymorphic Assumed Rank Argument
CAST-33342 Crayftn error with pure procedures in submodule
CAST-33324 Fortran with ACC fails on AMD GPU
CAST-33319 incorrect string lengths result after string arrays are initialized with array constructors of different lengths
CAST-33306 `select type` statement fails to identify argument as a `character` type
CAST-32911 needs -O0 to avoid a ftn internal compiler error
CAST-32887 cray-ftn bug with renaming derived type in use statement
CAST-32694 Unified memory does not work with omp declare target directive. Fortran compiler
CAST-32531 ICE using Cray compiler
CAST-32312 CCE Fortran+OpenMP Runtime Error on Crusher (Invalid dope vector)
CAST-31898 Cray Fortran + OpenACC bug
CAST-31821 MFEM: local memory exceeded CCE 15.0.0, ROCm 5.4.0 -O0
CAST-31788 ftn openmp target link directive runtime errors
CAST-31763 Add consistent debug flag behavior across CCE C++ and CCE Fortran compilers
CAST-31636 cray cce compiler OMPT support
CAST-31447 ftn openmp target link directive runtime errors
CAST-31375 Segfault on subroutine exit after omp target exit data clause
CAST-31156 Several problems with Cray CCE compiler version 14.0.3 and GFS weather code
CAST-30542 Cray fortran compiler error on access to module variable for unified memory
CAST-28426 CrayFTN build time issue - demonstrator of pathological behaviour
CAST-22740 Cray Fortran fails to catch bad G format spec
Product and OS Dependencies:
----------------------------
This CCE release is supported on
- HPE Cray supercomputer systems running CSM with COS 2.5.X (SLES15 SP4) or COS 23.11 (SLES15 SP5)
- HPE Cray supercomputer systems running HPCM with COS 2.5.X (SLES15 SP4) or COS 23.11 (SLES15 SP5)
- HPE Cray supercomputer systems running HPCM with RHEL 8.7
- HPE Apollo 2000 Gen10Plus systems running RHEL 8.7
- AMD GPU OpenMP and OpenACC offloading support requires at least ROCm 6.0 (AMD GPU HIP offloading can still use ROCm 5.x)
Notes and Limitations:
----------------------
- The OpenACC type `acc_handle_kind` has been changed to be of size 8 bytes (instead of 4 bytes). This will require recompilation of any codes that use this type.
- Linking Fortran applications may fail with an error message of the form "hidden symbol `<SYMBOL>' in <LIB> is referenced by DSO".
If this error message is seen, it can usually be worked around by adding '-lgcc_s' to your link line. This is known to affect use
of craypat and cray-parallel-netcdf but may be seen without them as well.
- There is a known bug in the CUDA header `cuda_fp16.h` detailed in the upstream issue https://github.com/llvm/llvm-project/issues/60296.
Programs compiling with `-fopenmp` must use CUDA 12.1.1 or later to avoid this bug, or manually change their header as described in the issue.
Documentation:
--------------
- S-5212 Cray Compiling Environment Release Overview (17.0)
- S-3901 Cray Fortran Reference Manual
- S-2179 Cray C and C++ Quick Reference
- Basic man pages: crayftn(1), craycc(1), crayCC(1), intro_openmp()
- Please see https://clang.llvm.org/docs/UsersManual.html or use the -help command line option for
more information on using Clang
Modulefile:
-----------
The following will load the modules necessary to use CCE:
module load PrgEnv-cray
The following will switch to x.y.z version of CCE:
module swap cce cce/x.y.z
Installation instructions:
--------------------------
rpm -ivh cce-17.0.1-202402021627.99c2eadd92a56-3.sles15sp4.x86_64.rpm.x86_64.rpm
The following script will set CCE version 17.0.1 default:
/opt/cray/pe/admin-pe/set_default_files/set_default_cce_17.0.1
License:
--------
Except for the third party components and software licensed by HPE through
proprietary agreements, components, files or programs contained within this
package or product are Copyright -2023 Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/cce/17.0.1/ATTRIBUTIONS_17.0.txt
Cray PE DL Plugin with Resiliency Support 22.06.1.2:
Release Date:
-------------
June 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 21.04.1:
o Updated DL framework support
o Introduction of resiliency support
o BFloat16 data type support
Supported Configurations:
-------------------------
Shasta/EX: CPU and AMD GPU support with PrgEnv-gnu
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10.2
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
For info regarding use of the new resiliency feature, see the examples
and README.md in ${CRAYPE_ML_PLUGIN_BASEDIR}/examples/resiliency
Sample Installation instructions:
---------------------------------
Note: Package must be installed or sym linked in the default /opt/cray/pe path.,
ie: use of the --prefix flag for rpm is not supported.
rpm -ivh craype-dl-plugin-ftr-22.06.1.2*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
Shasta/EX:
module load craype-dl-plugin-ftr/22.06.1.2
Certain components, files or programs contained within this package or
product are Copyright 2017-2022 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 21.02.1.3:
Release Date:
-------------
February 2021
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 20.10.1:
o Bug fixes for CUDA 11 support and pip install of include source dist
o Initial CUDA 11.1 support
o Updated DL framework support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC CLE 7: CPU and GPU support
Shasta: CPU-only support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.4
* PyTorch v1.7.1
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-21.02.1.3*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/Shasta:
module load craype-dl-plugin-py3/21.02.1.3
CS:
module load craype-dl-plugin-py3/openmpi/21.02.1.3
module load craype-dl-plugin-py3/mvapich/21.02.1.3
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.06.1.2:
Release Date:
-------------
June 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.04.1.1:
o Updated DL framework support
o BFloat16 data type support
o ROCm 5.x support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC CLE 7: CPU and GPU support
Shasta: CPU-only support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.06.1.2*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/Shasta:
module load craype-dl-plugin-py3/22.06.1.2
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.08.1:
Release Date:
-------------
August 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.06.1.2:
o PyTorch Apex Amp Optimization Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.08.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.08.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.09.1:
Release Date:
-------------
September 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.08.1:
o Keras callbacks bug fixes
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: CPU-only support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.6
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.09.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.09.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 22.12.1:
Release Date:
-------------
December 2022
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.09.1:
o TF v2.9 Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.9
* PyTorch v1.10
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using Tensorflow 1.14 binaries distributed by Google and Intel require portions of the DL
Plugin to be compiled with gcc 4.8. If using the included source distribution of the
DL Plugin to install the Python packagesinto a given Python installation, gcc 4.8
will need to be used to install said packages, assuming Google and Intel distributed
Tensorflow 1.14 binaries are used. For TensorFlow 1.15 and greater, use of gcc 7.x
is required instead.
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-22.12.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/22.12.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
Cray PE DL Plugin 23.09.1:
Release Date:
-------------
September 2023
Purpose:
--------
The following changes have been made since Cray PE DL Plugin 22.12.1:
o ROCm 5.3.x support
o TF v2.11 Support
o PT v1.12 Support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support for both AMD and NVIDIA
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.11
* PyTorch v1.12
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using TensorFlow, if you see an exception error related to apply_gradients() then
downgrade your keras installation version in your python environment.
pip install keras==2.10.0
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-23.09.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/23.09.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
HPE Cray DL Plugin 24.03.1:
Release Date:
-------------
March 2024
Purpose:
--------
The following changes have been made since HPE Cray DL Plugin 23.09.1:
o AMD MI300A APU support
o TensorFlow v2.15 support
o PyTorch v2.0.1 support
o ROCm 6.0.x support
o Cray-mpich 8.1.27 support
o GCC 12.2.0 support
Supported Configurations:
-------------------------
CS: OpenMPI 4.0.2 and cray-mvapich 2.3.2 or newer with CPU and GPU support
XC: CLE 7 CPU and GPU support
EX: CPU and GPU support for both AMD and NVIDIA
Apollo: RHEL8.6 CPU and GPU support
The following Deep Learning frameworks and versions are supported:
* TensorFlow v2.15
* PyTorch v2.0.1
Documentation:
--------------
For more information see the intro_dl_plugin man pages.
Known Issues:
-------------
* Using TensorFlow, if you see an exception error related to apply_gradients() then
downgrade your keras installation version in your python environment.
pip install keras==2.10.0
Sample Installation instructions:
rpm -ivh craype-dl-plugin-py3-24.03.1*.x86_64.rpm
The "*" in the install command represents date and hash specific information.
Modulefile:
-----------
XC/EX:
module load craype-dl-plugin-py3/24.03.1
CS:
module load craype-dl-plugin-py3/openmpi/21.04.1
module load craype-dl-plugin-py3/mvapich/21.04.1
Certain components, files or programs contained within this package or
product are Copyright 2017-2021 Hewlett Packard Enterprise Development LP.
CrayPE Targets 1.13.0
================================================================================
Release Date:
--------------------------------------------------------------------------------
March, 2023
Purpose:
--------------------------------------------------------------------------------
Adds modulefiles for targets - craype-accel-amd-gfx942
Bugs fixed in this release:
--------------------------------------------------------------------------------
None
Dependencies:
--------------------------------------------------------------------------------
The CrayPE Targets 1.13.0 release is dependent on the following
software products:
CrayPE
set_default_3
lmod_scripts
Documentation:
--------------------------------------------------------------------------------
See manpages for cc, CC, ftn, intro_craype-api, intro_hugepages and pkg-config
See section 2.6 Using Targeting Modules of the Cray Programming Environment
User's Guide (S-2529-116)
See http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh craype-targets-ex-1.13.0-202402062025.1d638354624fb-0.ex.noarch.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_craype-targets_1.13.0-tcl
/opt/cray/pe/admin-pe/set_default_files/set_default_craype-targets_1.13.0-lua
Product description:
--------------------------------------------------------------------------------
CrayPE Targets contains the modules for CrayPE targets.
TCL Modules - craype-accel-host craype-accel-amd-gfx90a craype-accel-amd-gfx908 craype-accel-amd-gfx940 craype-accel-amd-gfx942 craype-accel-nvidia70 craype-accel-nvidia80 craype-accel-nvidia90 craype-hugepages2M craype-hugepages4M craype-hugepages8M craype-hugepages16M craype-hugepages32M craype-hugepages64M craype-hugepages128M craype-hugepages256M craype-hugepages512M craype-hugepages1G craype-hugepages2G craype-arm-grace craype-x86-milan craype-x86-milan-x craype-x86-rome craype-x86-spr craype-x86-spr-hbm craype-x86-trento craype-x86-genoa craype-network-none craype-network-ofi craype-network-ucx
Lmod Modules - craype-accel-host craype-accel-amd-gfx90a craype-accel-amd-gfx908 craype-accel-amd-gfx940 craype-accel-amd-gfx942 craype-accel-nvidia70 craype-accel-nvidia80 craype-accel-nvidia90 craype-hugepages2M craype-hugepages4M craype-hugepages8M craype-hugepages16M craype-hugepages32M craype-hugepages64M craype-hugepages128M craype-hugepages256M craype-hugepages512M craype-hugepages1G craype-hugepages2G craype-arm-grace craype-x86-milan craype-x86-milan-x craype-x86-rome craype-x86-spr craype-x86-spr-hbm craype-x86-trento craype-x86-genoa craype-network-none craype-network-ofi craype-network-ucx
Copyright 2013-2023 Hewlett Packard Enterprise Development LP
CDEUX driver (craype) 2.7.31.11 - 11
==============
Release Date:
--------------
March 2024
Purpose:
--------
The Cray Development Environment User eXperience driver (CDEUX) (craype)
provides a common set of drivers for all CPE programming environments,
notably the C/C++ (cc/CC) and Fortran (ftn) languages.
New CDEUX driver (craype) features for HPE Cray EX and Apollo systems:
- Support for CCE using distribution GCC toolchain installation.
- Improved support for AMD ROCm installations.
- Improved AARCH64 support.
Key Changes and Bugs Closed:
----------------------------
- CDEUX driver (craype) 2.7.31.11 has fixed issues regarding OpenMP
Product and OS Dependencies:
----------------------------
The CDEUX driver (craype) 2.7.31.11 release is supported on the following
HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Documentation:
---------------
See manpages for cc, CC, ftn, and intro_craype-api.
See the pkg-config manpage or
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Product description:
--------------------
CrayPE contains drivers, cc, CC, and ftn to compile for all CPE
Programming Environments.
Certain components, files or programs contained within this package or product
are Copyright 2013-2023 Hewlett Packard Enterprise Development LP.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright 2023 Hewlett Packard
Enterprise Development LP.
Craypkg-gen 1.3.32
==============
Release date
------------
February 2024
Purpose
-------
Bug fix release.
Bugs fixed in this release
--------------------------
- Preserve build numbers in ROCm versions
- Exclude hip paths in the ROCm pkg-config files for ROCm >= 5.7.0
- Change -D_HIP_PLATFORM_HCC_ to -D__HIP_PLATFORM_AMD__ in pkg-config files for ROCm >= 5.7.0
- Utilize Jinja2 templating for ROCm pkg-config file
- Update logic to generate modulefiles for new file tree in Intel OneAPI 2024.0
- Update the Nvidia HPC SDK compiler compatibility version for aarch64
- Remove non-existent CUPTI path from Nvidia HPC SDK modulefile
Dependencies
------------
The craypkg-gen 1.3.32 release is supported on the following HPE
Cray systems:
- Cray EX systems SLE 15.0 or later and RHEL 8.0 or later.
Driver support for integrating Third Party C, C++, and Fortran libraries
through .pc files using pkg-config is used in CrayPE 2.x and later.
Limitation
----------
- Library dependencies for static libraries are not added to the .pc files
for keywords Requires.private and Libs.private. A warning is issued by
the craypkg-gen tool to advise users to add this information to the
libraries .pc files.
- RPM limits packages to 4GB
Documentation
-------------
Man pages for craypkg-gen are found by executing `module load craypkg-gen`
and then `man craypkg-gen`. See
http://www.freedesktop.org/wiki/Software/pkg-config for a pkg-config
introduction.
Examples for creating modulefiles for Intel, PGI and Python are included in
the craypkg-gen ‘doc’ directory:
/opt/cray/craypkg-gen/1.3.32/doc/intel_example.txt
/opt/cray/craypkg-gen/1.3.32/doc/pgi_example.txt
/opt/cray/craypkg-gen/1.3.32/doc/python_example.txt
Example:
- As an example, the Intel 16.0.3.210 compiler was recently released.
After installing the compiler the administrator creates a modulefile for
this release by executing the following commands:
# module load craypkg-gen
# craypkg-gen -m /opt/intel/compilers_and_libraries_2016.3.210
This version of the Intel compiler is made default by executing the
command:
# /opt/admin-pe/set_default_craypkg/set_default_intel_16.0.3.210
Installation instructions
-------------------------
rpm -ivh craypkg-gen-1.3.32-1.3.32-1.sles15sp5.noarch.rpm
To change the product version to default after installation:
/opt/cray/pe/admin-pe/set_default_files/set_default_craypkg-gen_1.3.32
Product description
-------------------
The craypkg-gen 1.3.32 utility provides the system administrator
a tool to integrate third party software with the Cray software stack.
Craypkg-gen assists with integration by creating .pc files for C, C++, and
Fortran libraries, pkg-config enabled modulefiles and RPMs.
The workflow for using craypkg-gen is
1) Build the open source software
2) Create .pc files for libraries
3) Create pkg-config enabled modulefiles
4) Customize pkg-config and modulefiles if needed
Certain components, files or programs contained within this package or product
are Copyright 2013-2024 Hewlett Packard Enterprise Development LP.
Cray DSMML 0.3.0:
==========================
Release Date:
-------------
March, 2024
Product Information:
--------------------
DSMML is a stand-alone memory management library for maintaining distributed
shared symmetric memory heaps for top level PGAS languages and libraries like
Coarray Fortran, UPC, and OpenSHMEM. DSMML allows user libraries to create
multiple symmetric heaps and share information with other libraries. Through
DSMML, we could extract interoperability between PGAS programming models.
Announcements, release informations, supported environments, and backward
compatibility informations about this product can be viewed in the following
location: https://pe-cray.github.io/cray-dsmml/
Purpose:
--------
Cray DSMML 0.3.0 supports the following features:
- Create symmetric heap segments for being used by PGAS languages and
libraries
- Allows options to allocate, deallocate memory buffers from previously
created symmetric heap segments
- Enables sharing information about the symmetric heap segments between
different programming models
- Added support for shared symmetric heap (SSHEAP) that can be used to
provide enhanced SMP data transport in user-libraries like SHMEM, CAF, UPC
The following features are fixed in Cray DSMML 0.3.0 when compared
to Cray DSMML 0.2.2:
- Add support for the library for AARCH64 processor based platforms
- Internal bug fixes involving hugepage cleanups
Refer intro_dsmml(3) manpage for more information on the supported features
and syntax for different routines
Product and OS Dependencies:
----------------------------
The Cray DSMML 0.3.0 release is supported on the following
Cray systems:
* HPE Cray EX systems with CLE
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.3 |
----------------------------+---------------------+
One or more compilers:
* AOCC 2.2 or later
* CCE 9.1 or later
* GNU 9.1 or later
* Intel 19.0 or later
* Nvidia 20.7 or later
Documentation:
--------------
Use the Cray DSMML man pages for more information on the library
functions and use intro_dsmml(3) man page for general information.
Use https://pe-cray.github.io/whitepapers/ for access to different
Cray DSMML specific whitepapers
Announcements, release informations, supported environments, and backward
compatibility informations about this product can be viewed in the following
location: https://pe-cray.github.io/cray-dsmml/
Modulefile:
-----------
module load cray-dsmml/0.3.0
Certain components, files or programs contained within this package or
product have the following Copyright:
Copyright 2018-2021 Hewlett Packard Enterprise Development LP.
FFTW 3.3.10.7
============
Release Date:
-------------
March 2024
Purpose:
--------
This Cray FFTW 3.3.10.7 release is supported on Cray EX (formerly
Shasta) systems. FFTW is supported on the host CPU but not on the
accelerator of Cray systems.
The Cray FFTW 3.3.10.7 release provides the following:
- Compiler and dependency updates
See the Product and OS Dependencies section for details.
Product and OS Dependencies:
----------------------------
The Cray FFTW 3.3.10.7 release is supported on the following Cray systems:
Cray EX Systems with SLES 15 SP4 or later based OS
The FFTW 3.3.10.7 release requires the following software products:
craype 2.7.5 or later
MPT 8.0 or later
One or more of the following major compiler versions:
CCE 17
GCC 12, 13
AOCC 4.1
Intel 2024 or later
Notes and Limitations:
----------------------
Starting with cray-fftw/3.3.6.1 the fftw module has been renamed from fftw
to cray-fftw.
Documentation:
--------------
http://www.fftw.org/index.html#documentation
See the intro_fftw3 manual page for additional information.
Modulefile:
-----------
module load cray-fftw
Installation
------------
To install the rpm, execute:
rpm -ivh cray-fftw-3.3.10.7-202402071520.4d6d346567dfd-1.shasta.sles15sp4.x86_64.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_fftw_3.3.10.7
Certain components, files or programs contained within this package or
product are Copyright 2011-2024 Hewlett Packard Enterprise Development LP
Cray LibSci 24.03.0
=====================
Release Date:
-------------
March 2024
Purpose:
--------
Cray LibSci 24.03.0 provides scientific libraries for Cray
HPC systems. Cray LibSci is supported on the host CPU but
not on the accelerator of these systems.
The Cray LibSci 24.03.0 release provides the following:
* support for RHEL gcc-toolset-{10,11,12}
* support for Nvidia HPC SDK 23.11 on aarch64
Cray LibSci 24.03.0 includes the following versions of publicly
available libraries:
* LAPACK 3.10.1 - For further information, see
http://www.netlib.org/lapack
* ScaLAPACK 2.2.0 - (Scalable LAPACK) For further information, see
http://www.netlib.org/scalapack.
* QDWH 2.0.0 KSVD 1.0.0 - Polar decomposition and SVD packages, see
https://github.com/ecrc/qdwh
https://github.com/ecrc/ksvd
Key Changes and Bugs Closed:
----------------------------
Changes in Cray LibSci 24.03.0
* CAST-34956 - Lua module environment variable issue
Changed in Cray LibSci 23.12.5
* CAST-33777 - LibSci mutex contention issue
* Removed support for Intel KNL
Product and OS Dependencies:
----------------------------
The Cray LibSci 24.03.0 release requires the following:
* SLES 15 SP4/SP5 or RHEL 8.8
* cray-mpich/8.1.0 or later
Supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Cray XD systems with CLE
One or more of the following compiler major versions:
* CCE 17.x (SLES)
* CCE 16.x or later (RHEL)
* GCC 12.3 or later (SLES)
* gcc-toolset-10.3/11.2/12.2 (RHEL)
* AOCC 4.1.0 or later
* AMD ROCm 5.0.0 or later
* Intel 2023.2.0 or later
* Nvidia 23.11 or later (aarch64) and 23.3 or later (x86_64)
Notes and Limitations:
----------------------
OMP threaded versions:
The CrayPE 2.1.2 and later releases add support for link line generation
for the multi-threaded versions of the libsci library based on the OpenMP
flags the user specifies for each compiler:
* CCE by default links to the OpenMP LibSci library. CrayPE will link in
the serial version of LibSci when the CCE flag -hnoomp is used.
* GNU by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the GNU flag -fopenmp is used.
* INTEL by default links serial LibSci library. CrayPE will link in the
OpenMP version of LibSci when the INTEL flag -qopenmp is used.
Stack size limit:
The 'ulimit -s unlimited' system stack size setting is required for
cray-libsci on all supported Cray platforms. This is typically set as a
system default by admin but may otherwise need to be set at runtime.
Non-default dynamic linking:
When using a non-default version of cray-libsci with dynamic linking
users should set at runtime and at linktime after loading the desired
cray-libsci module--or set the equivalent for their linux shell:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci man page for additional information.
See the csmlversion man page for information to display version information
on the currently loaded scientific libraries.
Modulefile:
-----------
module load cray-libsci
Installation instructions:
--------------------------
LibSci is now packaged into separate compiler specific RPMs to allow
rpmbuild to correctly include compiler dependencies.
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
/opt/cray/libsci/24.03.0/ATTRIBUTIONS
Cray LibSci_ACC 24.03.1
=======================
Release Date:
-------------
March 2024
Purpose:
--------
The Cray LibSci_ACC 24.03.1 release provides accelerated versions of
scientific libraries for HPE Cray systems with accelerators.
Product and OS Dependencies:
----------------------------
Cray LibSci_ACC 24.03.1 requires the following platform:
HPE Cray EX system with AMD MI250X
HPE Cray EX system with AMD MI300A
HPE Cray EX system with Nvidia Hopper
Cray LibSci_ACC 24.03.1 supports the following OS:
RHEL 8.8 x86_64
SLES 15 SP4 x86_64
SLES 15 SP5 x86_64 & aarch64
Notes and Limitations:
----------------------
LAPACK workspace calculations from other implementations or hard-coded
values may not be compatible with libsci_acc. Workspaces reported from
a query to the corresponding libsci_acc function should be used.
The current supported usecase is a one-rank to one-gpu mapping for apps.
Performance improvements can be achieved in programs calling libsci_acc
subroutines by using pinned memory. See the intro_libsci_acc man page for
details.
Use of the aprun option "-cc none" or the srun option "--cpu_bind=none"
to disable core affinity is strongly suggested to maintain performance.
When using non-default cray-libsci_acc versions users should set equivalent
at runtime and linktime after loading the desired cray-libsci_acc module:
export LD_LIBRARY_PATH=$CRAY_LD_LIBRARY_PATH:$LD_LIBRARY_PATH
Documentation:
--------------
See the intro_libsci_acc man page for additional information.
Modulefile:
-----------
module load cray-libsci_acc
Sample Installation instructions:
---------------------------------
RPM install command:
rpm -ivh cray-libsci-acc-24.03.1-*.##ARCH##.rpm
The "*" in the install command represents compiler version combinations.
To make this the default version:
##PREFIX##/admin-pe/set_default_files/set_default_libsci_acc_24.03.1
License:
--------
Except for the third party modules and software licensed by HPE
through proprietary agreements, components, files or programs
contained within this package or product are Copyright -2023 Hewlett
Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in
this package are detailed in the file:
##PREFIX##/libsci_acc/24.03.1/ATTRIBUTIONS_libsci_acc_24.03.1.txt
Cray Lmod 8.7.32
================================================================================
Release Date:
--------------------------------------------------------------------------------
March, 2024
Purpose:
--------------------------------------------------------------------------------
Provide a Lmod 8.7.32 package for CPE customers
Bugs fixed in this release:
--------------------------------------------------------------------------------
N/A
Documentation:
--------------------------------------------------------------------------------
See CPE's Installation and User Guides for documentation.
Also see Lmod's official website: https://lmod.readthedocs.io
Dependencies:
--------------------------------------------------------------------------------
The Lmod Scripts 8.7.32 release is dependent on the following
software:
bc
sed
lua >= 5.3
lua-luaposix
lua-luafilesystem
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh cray-lmod-8.7.32-1.sles15sp5.x86_64.rpm
To make this the default version, update the symbolic "lmod" link
to the desired Lmod version directory.
Lmod Scripts 3.2.1
================================================================================
Release Date:
--------------------------------------------------------------------------------
December, 2023
Purpose:
--------------------------------------------------------------------------------
Updates package license.
Bugs fixed in this release:
--------------------------------------------------------------------------------
Documentation:
--------------------------------------------------------------------------------
See system Installation and User Guides for documentation
Dependencies:
--------------------------------------------------------------------------------
The Lmod Scripts 3.2.1 release is dependent on the following
software:
set_default_3
Lua
Lmod
Installation instructions:
--------------------------------------------------------------------------------
rpm -Uvh lmod_scripts-3.2.1-202311141921.212030d47edcb-0.noarch.rpm
To make this the default version, execute:
/opt/cray/pe/admin-pe/set_default_files/set_default_lmod_scripts_3.2.1
Product description:
--------------------------------------------------------------------------------
Contains Lua scripts which support the Cray implementation of Lmod
Copyright 2020-2023 Hewlett Packard Enterprise Development LP
modules 3.2.11.7
==================
Release Date:
-------------
Dec, 2023
Purpose:
--------
- Package dependency update
Documentation:
---------------
See man-pages for module and modulefile.
Cray MPICH 8.1.29:
=======================================
Release Date:
-------------
February 21, 2024
Purpose:
--------
Cray MPICH 8.1.29 is based upon ANL MPICH 3.4a2 with support for libfabric
and is optimized for the Cray Programming Environment.
Major Differences Cray MPICH 8.1.29 from the XC Cray MPICH include:
- Uses the new ANL MPICH CH4 code path and libfabric for network
support.
- Does not support -default64 mode for Fortran
- Does not support C++ language bindings
New Cray MPICH features for HPE Cray EX and Apollo systems:
- Starting from the 8.1.26 release, Cray MPICH supports the Intel Sapphire Rapids CPU HBM
processor architecture.
- On systems with AMD GPUs, Cray MPICH 8.1.26 supports all ROCm
versions starting from ROCm 5.0, including the latest ROCm 5.5.0
release.
The Cray MPICH 8.1.25 release and prior versions of
Cray MPICH are only compatible with ROCm versions up to (and
including) the ROCm 5.4.0 release.
- Cray MPICH uses the libfabric "verbs;ofi_rxm" provider by default.
This is the supported and optimized OFI libfabric provider for
Slingshot-10 and Apollo systems.
- Cray MPICH offers support for multiple NICs per node. Starting with
version 8.0.8, by default Cray MPICH will use all available NICs on
a node. Several rank-to-NIC assignment policies are supported. For
details on choosing a policy for assigning ranks to NICS, or for
selecting a subset of available NICs, please see the following
environment variables documented in the mpi man page.
MPICH_OFI_NIC_VERBOSE
MPICH_OFI_NIC_POLICY
MPICH_OFI_NIC_MAPPING
MPICH_OFI_NUM_NICS
- Enhancements to the MPICH_OFI_NIC_POLICY NUMA mode have been added.
Starting with version 8.0.14, if the user selects the NUMA policy,
the NIC closest to the rank is selected. A NIC no longer needs to
reside in the same numa node as the rank. If multiple NICs are
assigned to the same numa node, the local ranks will round-robin
between them. Numa distances are analyzed to select the closest NIC.
- Cray MPICH supports creating a full connection grid during MPI_Init.
By default, OFI connections between ranks are set up on demand. This
allows for optimal performance while minimizing memory requirements.
However, for jobs requiring an all-to-all communication pattern, it
may be beneficial to create all OFI connections in a coordinated
manner at startup. See the MPICH_OFI_STARTUP_CONNECT description in
the mpi man page.
- Cray MPICH supports runtime switching to the UCX netmod starting
with version 8.0.14. To do this load the craype-network-ucx module
and module swap between Cray-MPICH and Cray-MPICH-UCX modules. For
more information including relevant environment variables reference
the intro_mpi man page with the Cray-MPICH-UCX module loaded.
- Lmod support for HPE Cray EX starting with Cray MPICH 8.0.16.
Key Changes and Bugs Closed:
----------------------------
Changes in Cray MPICH 8.1.29
New Features:
- Cray MPICH 8.1.29 for aarch64 is in early stage of support and
has some issues. Setting these variables will help circumvent most
known issues.
export MPICH_SMP_SINGLE_COPY_MODE=CMA
export MPICH_MALLOC_FALLBACK=1
Bugs Fixed:
- CAST-34406 - Improve performance in MPI Allreduce benchmark
- CAST-34872 - Fix seg-fault in extremely large allreduce calls
- CAST-34972 - Use dynamic CXI PID to avoid conflicts
- PE-45873 - Added infrastructure for CXI triggered operations non-blocking collectives
- PE-45874 - Add MPI_Ibcast() using CXI triggered operations
- PE-45876 - Add MPI_Igather() using CXI triggered operations
- PE-45877 - Add MPI_Iscatter() using CXI triggered operations
- PE-45878 - Add MPI_Ireduce() using CXI triggered operations
- PE-45879 - Add MPI_Iallreduce() using CXI triggered operations
- PE-49870 - Support MPI_LONG_DOUBLE with CCE for aarch64
- PE-51028 - Improved debugging to diagnose memcpy issues
- PE-51233 - Build libraries for the nvidia compiler on aarch64
- PE-51326 - Fix window creation using MPI_WIN_FLAVOR_SHARED to avoid running out of fi_mr resources
- PE-51365 - Fix bad initialization of hint striping_unit in MPI-IO
- PE-51445 - Support ROCM 6.0 with Cray-MPICH GTL
- PE-51470 - Update Dynamic Process Management documentation
- PE-51524 - Document PBS options for using Spawn on Slingshot
- PE-51738 - Bugfix for GPU NUMA nodes on MI300 GPUs
Product and OS Dependencies:
----------------------------
The Cray MPICH 8.1.29 release is supported on the following HPE systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product and OS Dependencies by network type:
--------------------------------------------------+
| Shasta |
----------------------------+---------------------+
craype | >= 2.7.6 |
----------------------------+---------------------+
cray-pals | >= 1.0.6 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
One or more compilers:
* AMD ROCM 5.0 or later
* AOCC 4.1 or later
* CCE 17.0 or later
* GNU 12.3 or later
* Intel 2022.1 or later
* Nvidia 23.3 or later
Notes and Limitations:
----------------------
Limitations in Cray MPICH 8.1.29:
- Cray MPICH 8.1.29 can support only ~2040 simultaneous MPI
communicators. This limit is less the XC Cray MPICH limit of
~4090 simultaneous communicators. Cray intends to raise the
limit in a future release of Cray MPICH for Shasta to at least
the XC limit.
Documentation:
--------------
For more information see the intro_mpi man page.
Modulefile:
-----------
module load cray-mpich/8.1.29
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright -2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/mpich/8.1.29/ATTRIBUTIONS
Copyright -2023 Hewlett Packard Enterprise Development LP
Copyright 2022-2024 Hewlett Packard Enterprise Development LP
Product Release Information
================================================================================
Product identification:
-----------------------
HPE Cray MPIxlate: version 1.0.4 (built on 2024-02-16T14:55:41Z from 9bd5b45)
Purpose:
--------
HPE Cray MPIxlate enables applications compiled using an MPI library that
is not binary compatible with HPE Cray MPI, to be run without recompilation
on supported HPE platforms.
Warnings & Limitations:
-----------------------
- See NOTES in mpixlate(1)
Changes from previous release:
------------------------------
- New features:
* None
- Defects Fixed:
* None
Platform, Operating System and Product dependencies:
----------------------------------------------------
Supported HPE platforms:
- aarch64 and x86_64 HPE Cray EX systems with HPCM or CSM
Supported Operating Systems:
- SLES 15 SP4, SP5
- COS 2.5, 23.11
Product dependencies:
-----------------+----------
cray-mpich[-ucx] | >= 8.1.29
-----------------+----------
Documentation:
--------------
mpixlate(1)
Environment Modules
-------------------
module load cray-mpixlate/1.0.4
================================================================================
Cray PALS 1.3.2:
================
Release Date:
-------------
Nov 2023
Purpose:
--------
Cray Parallel Application Launch Service (PALS) is an application launcher for
Cray PE applications.
Key Changes:
------------
Add host2xname tool
Allow configuring the PMI port range
Add more examples in manpages
Add SLES 15 SP5 build
Add SLES 15 SP4 and SP5 aarch64 build
Add PMI and PMI2 as --pmi options for mpiexec
Periodically ping applications
Add the ability to allocate additional VNIs for special services like DAOS
Add mpiexec --disable-rdzv-get option to disable rendezvous gets in the NICs
Enable --cpu-bind and --gpu-bind=verbose to work with --rankfile
Add PALSD_FANOUT env var for palsd to set fanout tree width
Add ability for simultaneous mpiexecs to divide up PBS job slots
Add RHEL 8.8 build
Add support for AMD & NVIDIA GPU visibility
Change the wire format for PMIx data to make PMIx_Fence more efficient
Increase core file size limit on RHEL systems
Add retries to SPIRE auth token call
Install aprun and mpirun as symlinks
Use position-independent executables
Change F_CXI_COLL_* -> FI_CXI_COLL_* environment variables
Configure Instant On URL on Shasta
Move binaries and libraries to /opt/cray
Change pmi_example to print NIC distances for all ranks
Make PALS command's nodelists consistent
Set ATOM_SOCKET by default
Improve vnid request error reporting
Launch with PMIx support automatically fetch NIC information
Reduce default fanout to 32 nodes
Require PMIx version 4.2.4 or newer
Send the job-step name to vnid when creating a job-step VNI
Include pid in palsd log messages
Raise maximum number of ZMQ sockets to 4096
Munge auth no longer used for some internal RPCs with trusted connections
Increase default send timeout to 8s
Fix errors found with clang scan-build
Fix nohup mpiexec behavior
Improve cleanup when launch RPC fails
Implement the verbose option for GPU binding
Fix support for the cycle keyword with the mpiexec --rankfile option
Fix error checking for GPU binding options
Better handle mpiexec disconnects
Fix segfault on PMIx RPC for missing app
Fix spawn-related memory leaks
Fix misleading warning in line buffer mode
Remove SLES 15 SP3 build
Deprecate PALS_FANOUT and APRUN_FANOUT environment variables
Install cray-pals RPM
Documentation:
--------------
For more information see aprun, mpiexec, palscmd, palsctrl, palsig, and palstat man page.
Modulefile:
-----------
module load cray-pals
Installation instructions:
--------------------------
rpm -ihv cray-pals-<version>-<date>.el8.x86_64.rpm
rpm -ihv cray-palsd-<version>-<date>.el8.x86_64.rpm
Copyright 2022-2023 Hewlett Packard Enterprise Development LP
cray-papi
=================
Release date:
-------------
December 2023
Package:
--------
cray-papi-7.0.1.2-202310131612.70757f986edac-4 x86_64
Purpose:
--------
New version of papi 7.0.1.2 release 4
Documentation:
--------------
Overview: https://github.com/icl-utk-edu/papi
Web Site: https://icl.utk.edu/papi
Product description:
--------------------
PAPI aims to provide the tool designer and application engineer with a
consistent interface and methodology for use of the performance counter
hardware found in most major microprocessors. PAPI enables software
engineers to see, in near real time, the relation between software
performance and processor events.
Dependencies:
-------------
For a list of software used when validating this version of
PAPI on Cray and HPE systems, see the HPE Cray Programming
Environment release announcements.
Copyright 2015-2017,2019-2023 Hewlett Packard Enterprise Development LP
Perftools 23.12.0
===============
Release Date: December, 2023
Purpose:
========
This is a feature and bugfix release for the following systems:
- HPE Cray EX and HPE Cray Supercomputer Systems with HPCM
- HPE Cray EX and HPE Cray Supercomputer Systems with CSM
- HPE Apollo 2000 Gen 10 Plus Systems (x86)
Key enhancements or changes from the previous release:
======================================================
o Upgrade to DWARF 0.8.0
o Upgrade to PAPI 7.0.1.2
- Upstream bugfixes
- Upstream CUDA component version locked from prior release and renamed cuda_7011
- Upstream ROCm component version locked from release 7.0.0.1 and renamed rocm_7001
- Support for Nvidia Grace processor
o Add support for ROCm 5.7.0
o Add support for CUDA 12.2
o Add support for Nvidia H100 GPUs
o Update ROCm trace wrappers to support up to ROCm 5.7
o Add support for PAT_region_push/pop in full trace mode
o Add support for profiling Python 3.11
o Add a DSMML tracegroup
o Improved counter groups and derived metrics for AMD Genoa processors
o Add support for OMPT target callbacks with CCE's OpenMP runtime, enabled
via the CRAYPAT_OMP_TOOL environmnet variable (NOTE: this is primarily intented
for use with pat_run. When instrumenting executables with CCE, proprietary
tracepoints are used in favor of OMPT. To use OMPT instead, '-fno-instrument-openmp'
must be added to compilation.)
o Perftools and its libraries are now compiled using the OS native
gcc and no longer depend on cray-gcc
o Set PAT_RT_ACC_MODEL_NAME in the runtime environment to manually specify
the accelerator model name.
Bugs Fixed:
===========
o Unexpected callers appearing in the calltree when using perftools-lite-loops
o An inappropriate assertion failure when MPI_Get_count returns MPI_UNDEFINED
o A runtime failure when incompatible versions of HIP and pat_run are used
o An issue preventing lambda and template function resolution using Perftools
with Caliper
o Segfaults when managing a hipStream with singletons
o Segfaults when tracing OpenACC codes
o Program hangs at dlopen using pat_run on certain Python codes
o Program hangs using the '-gheap' trace option on RHEL systems
o Incorrect number of PEs/Node reported when using PBS and PALS
o Incorrect launching of Apprentice2 when using reveal launcher
Notes:
======
o This release of Perftools only supports ROCm 5.7.0; SLES 15 support is limited to SP5
o This release of Perftools only supports CUDA 12.0 and later
Known Issues:
=============
o Fortran applications compiled against CCE 16 may fail during instrumentation
with the following error message: "hidden symbol `<SYMBOL>' in <LIB> is
referenced by DSO". To work around this error specify the following during
the pat_build invocation '-Dlink-instr=-lgcc_s'.
o When using OMPT with CCE's openmp runtime, to and from device transfers and
kernel executions originating from a single construct are treated as separate
events. For example, 'target map(tofrom:)' will accumulate 3 events each
time it is encountered.
o Nvidia device drivers do not correctly populate system files used by Perftools
for the Grace-Hopper superchip, impacting driver versions up to: 545.23.06.
Reports generated from applications run on these devices may list the accelerator
device as "Unknown". Additionally, usage of accelerator performance counter groups
is not supported when the device is unresolved. To work around this issue set the
following runtime environment variable prior to running the experiment
'PAT_RT_ACC_MODEL_NAME="GH200 120GB"'.
o Tracing experiments with accelerator performance counters and all sampling experiments
may produce unexpected behavior on Grace-Hopper superchips. At this time, there is
no work around this issue and the capabilities should be treated as experimental.
o Sampling experiments on aarch64 systems lead to fatal errors if
PAT_RT_CALLSTACK_MODE="unwind" is used. The setting is disabled for these situations
and PAT_RT_CALLSTACK_MODE="frames" is used instead.
o The Intel OneAPI Fortran compiler does not correctly produce the ELF .debug_pubnames or
.debug_names section in object files when the "-debug pubnames" option is specified
(as required when the perftools module is loaded). This prevents pat_build from collecting
the necessary information that supports tracing user-defined entry points. Specify the
"-Ddebug-names=n" option to pat_build so pat_build can collect the information from the
ELF .debug_info section.
o The Intel OneAPI Fortran compiler does not accurately represent formal parameters to a
subprogram in some cases. This results in the generation of incorrect intercepts for
user-defined entry points when pat_build -u is specified and results in seg faults when
the instrumented program is executed. Specify the "-Dtrace-force-deref=y" to pat_build
to account for the missing formal parameter information.
Deprecations and Removals:
==========================
o Remove support for PGI programming Environment
o Remove support for following trace groups: craymem, hbw, gni, jemalloc, memkind
Dependencies:
=============
o A PDF reader (such as evince, acroread or okular) is required to use the
app2 online help.
o For a list of software used when validating this version of
Perftools on Cray and HPE systems, see the HPE Cray Programming Environment
release announcements.
Installation instructions:
==========================
For Apollo 80 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 80 (ARM) System (S-8013).
For Apollo 2000 systems, please see the HPE Cray Programming Environments
Installation Guide: Apollo 2000 Gen10 Plus (x86) System (S-8012).
For HPE Cray EX (Shasta Architecture) systems, refer to the Cray Asynchronous
Installer Guide (S-8003).
Installation of app2 remote client (+ server) on Mac systems:
-------------------------------------------------------------
Apprentice2Installer-23.12.0.dmg
RevealInstaller-23.12.0.dmg
The Cray Apprentice2 and Reveal installers for Mac are included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Mac OS El Capitan through Mojave. Double click on installer
to begin installation. The installer will walk you through the
process for your system.
Installation of app2 remote client (+ server) on Windows 10 systems:
--------------------------------------------------------------------
Apprentice2Installer-23.12.0.exe
The Cray Apprentice2 installer for Windows is included in the
perftools-clients rpm, and placed in
$CRAYPAT_ROOT/share/desktop_installers/
on a Cray Shasta User Access Node (UAN) or on a Cray login node when the
perftools software is installed.
Download the Cray Apprentice2 installer onto a desktop or laptop
running Windows 10. Double click on installer to begin installation.
The installer will walk you through the process for your system.
Documentation:
==============
See the following man pages:
app2, grid_order, intro_craypat, pat_build, pat_help, pat_info, pat_opts,
pat_report, pat_run, pat_view, reveal
intro_papi
perftools-base, perftools-lite, perftools-preload
accpc, cray_cassini, cray_pm, hwpc, nwpc, papi_counters, uncore
PAPIlicnotices
Search for perftools on https://support.hpe.com to access documentation
on Cray Performance Measurement and Analysis Tools
License:
========
Except for the third party modules discussed below and software licensed
by HPE through proprietary agreements, components, files or programs
contained within this package or product are Copyright 2001-2023
Hewlett Packard Enterprise Development LP.
Attribution notices for open source licensed software contained in this
package are detailed in the file:
$CRAYPAT_ROOT/ATTRIBUTIONS_perftools.txt
Cray PMI 6.1.14
==============================================================================
Release Date:
--------------
February 12, 2024
Purpose:
--------
The Cray Process Manager Interface Library provides the interface between the
application launcher and other communication libraries such as MPICH and SHMEM.
Cray PMI 6.1.14 is optimized for the Cray Programming Environment.
New Cray PMI features for HPE Cray EX and Apollo systems:
Key Changes and Bugs Closed:
----------------------------
Starting with Cray-PMI 6.1.0 the cray-pmi and cray-pmi-lib modules have been
combined into a single cray-pmi module. The cray-pmi-lib module will no longer
be installed by new cray-pmi rpms. When the cray-pmi module is loaded, the
pmi libraries will be linked with as-needed flags.
Bug fixes new to Cray PMI 6.1.14 include:
- CAST-34500 - Include unversioned pmi library symlinks in the CPE ld cache
- PE-51739 - Create modern PMI process map code
Product and OS Dependencies:
----------------------------
The Cray PMI 6.1.14 release is supported on the following systems:
* HPE Cray EX systems with CLE
* HPE Apollo systems as part of the Cray Programming Environment
Product Dependencies:
----------------------------------------+
cray-pals | >= 1.0.6 |
-----------------------+----------------+
slurm | >= 20.02 |
-----------------------+----------------+
Notes and Limitations:
----------------------
Documentation:
--------------
For more information see the intro_pmi man page.
Modulefile:
-----------
module load cray-pmi/6.1.14
License:
--------
Except for the third party components and software licensed by HPE
through proprietary agreements, components, files or programs contained
within this package or product are Copyright 2020-2021 Hewlett Packard
Enterprise Development LP.
Attribution notices for open source licensed software for this
package are detailed in the file:
/opt/cray/pe/pmi/6.1.14/ATTRIBUTIONS
cray-python 3.11.7
====================
Release Date
------------
March 2024
Purpose
-------
Cray Python is an implementation of the Python programming language for the Cray
Programming Environment. The numpy and scipy modules are configured to call Cray
Libsci routines. The mpi4py module is configured to call Cray MPICH routines.
Cray Python is designed to run Python codes on the compute nodes of an HPE
supercomputer. HPE does not make changes to the Python source or any of its
libraries nor does it plan to make changes in future releases.
The cray-python 3.11.7 release contains
- python-3.11.7
- numpy-1.24.4
- scipy-1.10.1
- mpi4py-3.1.4
- dask-2023.6.1
Product and OS Dependencies
---------------------------
The cray-python 3.11.7 release is supported on
- Cray EX systems running SLES 15 or RHEL 8
Documentation
-------------
https://www.python.org/doc
Changelog
---------
https://docs.python.org/release/3.11.7/whatsnew/changelog.html
Modulefile
----------
module load cray-python/3.11.7
Installation
------------
rpm -ihv cray-python-3.11.7-20240208204342.1cdbf56-1.sles15sp4.x86_64.rpm
To make this the default version, execute
[PREFIX]/admin-pe/set_default_files/set_default_python_3.11.7
Certain components, files, or programs contained within this package are
© Copyright 2021-2024 Hewlett Packard Enterprise Development LP
cray-R 4.3.2
============
Release Date
------------
March 2024
Purpose
-------
The cray-R 4.3.2 release.
cray-R is configured to call cray-libsci routines.
Product and OS Dependencies
---------------------------
The cray-R 4.3.2 release is supported on
- Cray EX systems running SLES 15 or RHEL 8
Documentation
-------------
https://cran.r-project.org/manuals.html
Modulefile
----------
module load cray-R/4.3.2
Installation
------------
rpm -ihv cray-R-4.3.2-20240208204433.1115d98-1.sles15sp4.x86_64.rpm
To make this the default version, execute:
[PREFIX]/admin-pe/set_default_files/set_default_R_4.3.2
Certain components, files or programs contained within this package or product are
Copyright 2018-2024 Hewlett Packard Enterprise Development LP
Cray OpenSHMEMX 11.7.1:
===============================
Release Date:
-------------
Mar, 2024
Purpose:
--------
OpenSHMEM is a Partitioned Global Address Space (PGAS) library interface
specification, which is the culmination of a standardization effort among
many implementers and users of SHMEM programming model. SHMEM has a long
history as a parallel programming model on HPE Cray supercomputer systems.
For the past two decades SHMEM library implementation in HPE Cray systems
evolved through different generations. Cray OpenSHMEMX is a proprietary,
OpenSHMEM specification compliant SHMEM implementation for HPE Cray
product line.
Key Changes:
------------
Major differences in Cray OpenSHMEMX 11.7.1 from
Cray OpenSHMEMX version 11.7.0 includes the following:
- Add support for logging HPE Slingshot NIC counter values
- Add support for using the library on platforms with AARCH64
processors
- Improve performance of shmem_quiet memory ordering operation
- Improve performance for allreduce collective operations when all
PEs are involved in the communication operations
Bugs Closed:
------------
The following bugs are fixed as part of the Cray OpenSHMEMX 11.7.1
release:
- Fix C11 generic APIs for shmem_sync operations
- Fix library usage with GNU and Intel compilers
Product and OS Dependencies:
----------------------------
The Cray OpenSHMEMX 11.7.1 release is supported on the following
Cray systems:
* HPE Cray EX systems with CLE
Product and OS Dependencies by network type:
--------------------------------------------------+
| HPE Slingshot |
----------------------------+---------------------+
craype | >= 2.7.3 |
----------------------------+---------------------+
cray-pals | >= 0.3.5 |
----------------------------+---------------------+
cray-pmi | >= 6.0.1 |
----------------------------+---------------------+
libfabric | >= 1.9.0 |
----------------------------+---------------------+
cray-dsmml | >= 0.2.2 |
----------------------------+---------------------+
One or more compilers:
* AOCC 2.2 or later
* CCE 9.1 or later
* GNU 9.1 or later
* Intel 19.0 or later
* Nvidia 20.7 or later
Notes and Limitations:
----------------------
Announcements, release information, supported environments, limitations,
usage models, and backward compatibility informations about this product
can be viewed in the following location:
https://pe-cray.github.io/cray-openshmemx/
Documentation:
--------------
Use the Cray OpenSHMEMX man pages for more information on the library
functions and use intro_shmem(3) man page for general information.
Initial support for Cray OpenSHMEMX man pages are derived from OpenSHMEM
standards specification document
Use https://pe-cray.github.io/whitepapers/ for access to different
Cray OpenSHMEMX specific whitepapers
Modulefile:
-----------
module load cray-openshmemx/11.7.1
License:
--------
Components, files, and programs contained within this package or
product are Copyright Hewlett Packard Enterprise Development LP