Using Chapel on HPE Cray Systems

The following information is assembled to help Chapel users get up and running on HPE Cray systems including the HPE Cray EX and Cray XC series systems.

Getting Started with Chapel on HPE Cray EX or XC Systems

Chapel is available as a module for HPE Cray EX and XC systems. When it is installed on your system, you do not need to build Chapel from the source release (though you can). Using the module systems on such platforms, you can use Chapel with the default settings and confirm it is correctly installed, as follows:

1. Ensure this required module is loaded. Normally it will be loaded for you, but under some circumstances you may need to load or restore it yourself:

   PrgEnv-gnu or PrgEnv-cray
   

2. Load the Chapel module:

   module load chapel
   

3. Compile an example program using:

   chpl $CHPL_HOME/examples/hello6-taskpar-dist.chpl
   

4. Execute the resulting executable (on four locales):

   ./hello6-taskpar-dist -nl 4
   

This should be all that is necessary to use Chapel on an HPE Cray EX or XC system. If the installation setup by your system administrator deviates from the default settings, or you are interested in other configuration options, see Using Chapel on an HPE Cray System below. If instead you wish to build Chapel from source, continue on to Building Chapel for an HPE Cray System from Source just below.

Note that the Chapel module for HPE Cray EX systems does not currently have the GASNet communication layer built into it. For multilocale execution on EX systems please use the ofi communication layer instead.

For information on obtaining and installing the Chapel module please contact your system administrator.

Getting Started with Chapel on HPE Apollo and Cray CS Systems

On HPE Apollo and Cray CS systems, Chapel is not currently available as a module due to the wide diversity of configurations that these systems support. For this reason, Chapel must be built from source on these systems using the Building Chapel for an HPE Cray System from Source instructions just below.

Building Chapel for an HPE Cray System from Source

1. Set CHPL_HOST_PLATFORM to the string representing your system type.

   For example:

   export CHPL_HOST_PLATFORM=hpe-cray-ex
   

   The following table lists the supported systems and strings. Note that on HPE Cray EX and XC systems, these values should typically be inferred automatically and not need to be set manually. That said, there is also no downside to setting them manually. As with most CHPL_* environment variables, the current set and inferred values can be determined by running $CHPL_HOME/util/printchplenv.

   System	CHPL_HOST_PLATFORM
   EX series	hpe-cray-ex
   Apollo series	hpe-apollo
   XD series	hpe-cray-xd
   XC series	cray-xc
   CS series	cray-cs

2. Optionally, set the CHPL_LAUNCHER environment variable to indicate how Chapel should launch jobs on your system:

   On an HPE Cray EX system, …	set CHPL_LAUNCHER to…
   …queue jobs using SLURM (sbatch)	slurm-srun
   …run jobs interactively using PALS	pals

   On a HPE Apollo or Cray CS system, to…	set CHPL_LAUNCHER to…
   …run jobs interactively on your system	gasnetrun_ibv
   …queue jobs using SLURM (sbatch)	slurm-gasnetrun_ibv
   …queue jobs using LSF (bsub)	lsf-gasnetrun_ibv
   …queue jobs using PBS (qsub)	pbs-gasnetrun_ibv

   On a Cray XC system, to…	set CHPL_LAUNCHER to…
   …run jobs interactively on your system	aprun
   …queue jobs using PBS (qsub)	pbs-aprun
   …queue jobs using SLURM (sbatch)	slurm-srun

   On HPE Cray EX systems, CHPL_LAUNCHER defaults to slurm-srun if srun is in your path and none otherwise.

   On HPE Apollo and Cray CS systems, CHPL_LAUNCHER defaults to gasnetrun_ibv.

   On Cray XC systems, CHPL_LAUNCHER defaults to aprun if aprun is in your path, slurm-srun if srun is in your path and none otherwise.

   You can also set CHPL_LAUNCHER to none if you prefer to manually manage all queuing and job launch commands yourself, though the advantage of using a launcher is to make sure that Chapel maps processes and threads to the appropriate target hardware units.

   For more information on Chapel’s launcher capabilities and options, refer to Chapel Launchers.

3. Select the target compiler that Chapel should use when compiling code for the compute node:

   On an HPE Apollo or Cray CS series system, set the CHPL_TARGET_COMPILER environment variable to indicate which compiler to use (and make sure that the compiler is in your path).

   To request…	set CHPL_TARGET_COMPILER to…
   …the LLVM/clang backend	llvm (default)
   …the GNU compiler (gcc)	gnu
   …the Clang compiler	clang
   …the Intel compiler (icc)	intel

   On an HPE Cray EX or Cray XC system, when using the C back-end, ensure that you have one of the following Programming Environment modules loaded to specify your target compiler:

   PrgEnv-allinea (ARM only)
   PrgEnv-cray
   PrgEnv-gnu
   PrgEnv-intel
   

4. Make sure you’re in the top-level chapel/ directory and make/re-make the compiler and runtime:

   gmake
   

   Note that a single Chapel installation can support multiple configurations simultaneously and that you can switch between them simply by changing any of the above settings. However, each configuration must be built separately. Thus, you can change any of the settings in the steps before this, and then re-run this step in order to create additional installations. Thereafter, you can switch between any of these configurations without rebuilding.

Using Chapel on an HPE Cray System

1. If you are working from a Chapel module:

   1. Load the module using module load chapel

   2. Optionally select a launcher, as in step 2 above

   3. Select a target compiler, as in step 3 above

   If you are working from a source installation:

   1. Set your host platform as in step 1 above

   2. Optionally select a launcher, as in step 2 above

   3. Select a target compiler, as in step 3 above

   4. Set CHPL_HOME and your paths by invoking the appropriate util/setchplenv script for your shell. For example:

   source util/setchplenv.bash
   

2. On HPE Cray EX systems with CHPL_COMM=ofi, optionally, load the Cray PMI modules:

   module load cray-pmi{,-lib}
   

   Often this is not required. Usually the default PMI support has sufficient capabilities to support Chapel program startup. But under certain circumstances it does not, and when you run a Chapel program that was built without these loaded you will see messages like this one:

   [PE_0]:_pmi2_add_kvs:ERROR: The KVS data segment of <num> entries
   is not large enough.  Increase the number of KVS entries by
   setting env variable PMI_MAX_KVS_ENTRIES to a higher value.
   

   Having the Cray PMI modules loaded when the program is compiled will prevent this problem. We expect that eventually these modules will be loaded by default on EX systems, but so far this has not consistently been the case.

3. Compile your Chapel program. For example:

   chpl $CHPL_HOME/examples/hello6-taskpar-dist.chpl
   

   See Compiling Chapel Programs or man chpl for further details.

4. If CHPL_LAUNCHER is set to anything other than none, when you compile a Chapel program for your Cray system, you will see two binaries (e.g., hello6-taskpar-dist and hello6-taskpar-dist_real). The first binary contains code to launch the Chapel program onto the compute nodes, as specified by your CHPL_LAUNCHER setting. The second contains the program code itself; it is not intended to be executed directly from the shell prompt.

   You can use the -v or --dry-run flags to see the commands used by the launcher binary to start your program.

   If CHPL_LAUNCHER is pbs-aprun:

   1. You can optionally specify a queue name using the environment variable CHPL_LAUNCHER_QUEUE. For example:

      export CHPL_LAUNCHER_QUEUE=batch
      

      If this variable is left unset, no queue name will be specified. Alternatively, you can set the queue name on your Chapel program command line using the --queue flag.

   2. You can also optionally set a wall clock time limit for the job using CHPL_LAUNCHER_WALLTIME. For example to specify a 10-minute time limit, use:

      export CHPL_LAUNCHER_WALLTIME=00:10:00
      

      Alternatively, you can set the wall clock time limit on your Chapel program command line using the --walltime flag.

   For further information about launchers, please refer to Chapel Launchers.

5. Execute your Chapel program. Multi-locale executions require the number of locales (compute nodes) to be specified on the command line. For example:

   ./hello6-taskpar-dist -nl 2
   

   Requests the program to be executed using two locales.

6. If your HPE Cray system has compute nodes with varying numbers of cores, you can request nodes with at least a certain number of cores using the variable CHPL_LAUNCHER_CORES_PER_LOCALE. For example, on a system in which some compute nodes have 24 or more cores per compute node, you could request nodes with at least 24 cores using:

   export CHPL_LAUNCHER_CORES_PER_LOCALE=24
   

   This variable may be needed when you are using the aprun launcher and running Chapel programs within batch jobs you are managing yourself. The aprun launcher currently creates aprun commands that request the maximum number of cores per locale found on any locale in the system, irrespective of the fact that the batch job may have a lower limit than that on the number of cores per locale. If the batch job limit is less than the maximum number of cores per locale, you will get the following error message when you try to run a Chapel program:

   apsched: claim exceeds reservation's CPUs
   

   You can work around this by setting CHPL_LAUNCHER_CORES_PER_LOCALE to the same or lesser value as the number of cores per locale specified for the batch job (for example, the mppdepth resource for the PBS qsub command). In the future we hope to achieve better integration between Chapel launchers and workload managers.

7. If your HPE Cray system has compute nodes with varying numbers of CPUs per compute unit, you can request nodes with a certain number of CPUs per compute unit using the variable CHPL_LAUNCHER_CPUS_PER_CU. For example, on a Cray XC system with some nodes having at least 2 CPUs per compute unit, to request running on those nodes you would use:

   export CHPL_LAUNCHER_CPUS_PER_CU=2
   

   Currently, the only legal values for CHPL_LAUNCHER_CPUS_PER_CU are 0 (the default), 1, and 2.

For more information on…	see…
…CHPL_* environment settings	Setting up Your Environment for Chapel
…Compiling Chapel programs	Compiling Chapel Programs
…Launcher options	Chapel Launchers
…Executing Chapel programs	Executing Chapel Programs
…Running multi-locale Chapel programs	Multilocale Chapel Execution

Cray File Systems and Chapel execution

For best results, it is recommended that you execute your Chapel program by placing the binaries on a file system shared between the login node and compute nodes (typically Lustre), as this will provide the greatest degree of transparency when executing your program. In some cases, running a Chapel program from a non-shared file system will make it impossible to launch onto the compute nodes. In other cases, the launch will succeed, but any files read or written by the Chapel program will be opened relative to the compute node’s file system rather than the login node’s.

Special Notes for Cray XC Series Systems

Native ugni Communication Layer

The Multilocale Chapel Execution page describes the runtime communication layer implementations that can be used by Chapel programs. In addition to the standard ones, Chapel supports a Cray-specific ugni communication layer. The ugni communication layer interacts with the system’s network interface very closely through a lightweight interface called uGNI (user Generic Network Interface). On Cray XC systems the ugni communication layer is the default.

Using the ugni Communications Layer

To use ugni communications:

1. Leave your CHPL_COMM environment variable unset or set it to ugni:

   export CHPL_COMM=ugni
   

   This specifies that you wish to use the Cray-specific communication layer.

2. (Optional) Load an appropriate craype-hugepages module. For example:

   module load craype-hugepages16M
   

   The ugni communication layer can be used with or without so-called hugepages. Performance for remote variable references is much better when hugepages are used. The only downside of using hugepages is that the tasking layer may not be able to detect task stack overflows by means of guard pages (see below).

   To use hugepages, you must have a craype-hugepages module loaded both when building your program and when running it. There are several hugepage modules, with suffixes indicating the page size they support. For example, craype-hugepages16M supports 16 MiB hugepages. It does not matter which craype-hugepages module you have loaded when you build your program. Any of them will do. Which one you have loaded when you run a program does matter, however. For general use, the Chapel group recommends the craype-hugepages16M module. You can read on for more information about hugepage modules if you would like, but the recommended craype-hugepages16M module will probably give you satisfactory results.

   The Cray network interface chips (NICs) can only address memory that has been registered with them. In practical terms, the Aries(TM) NIC on Cray XC systems is not limited as to how much memory it can register. However, it does have an on-board cache of 512 registered page table entries, and registering more than this can cause reduced performance if the program’s memory reference pattern causes refills in this cache. We have seen up to a 15% reduction from typical nightly XC-16 performance in an ra-rmo run using hugepages small enough that every reference should have missed in this cache. Covering an entire 128 GiB XC compute node with only 512 hugepages will require at least the craype-hugepages256M module’s 256 MiB hugepages.

   Offsetting this, using larger hugepages may reduce performance because it can result in poorer NUMA affinity. With the ugni communication layer, arrays larger than 2 hugepages are allocated separately from the heap, which improves NUMA affinity. An obvious side effect of using larger hugepages is that an array has to be larger to qualify. Thus, achieving the best performance for any given program may require striking a balance between using larger hugepages to reduce NIC page table cache refills and using smaller ones to improve NUMA locality.

   Note that when hugepages are used with the ugni comm layer, tasking layers cannot use guard pages for stack overflow detection. Qthreads tasking cannot detect stack overflow except by means of guard pages, so if ugni communications is combined with qthreads tasking and a hugepage module is loaded, stack overflow detection is unavailable.

Network Atomics

The Aries networks on Cray XC series systems support remote atomic memory operations (AMOs). When the CHPL_NETWORK_ATOMICS environment variable is set to ugni, the following operations on remote atomics are done using the network:

32- and 64-bit signed and unsigned integer types:
32- and 64-bit real types:
  read()
  write()
  exchange()
  compareAndSwap()
  add(), fetchAdd()
  sub(), fetchSub()

32- and 64-bit signed and unsigned integer types:
  or(),  fetchOr()
  and(), fetchAnd()
  xor(), fetchXor()

All of the operations shown above are done natively by the network hardware except 64-bit real add, which is disabled in hardware and thus done using on statements.

ugni Communication Layer and the Heap

The “heap” is an area of memory used for dynamic allocation of everything from user data to internal management data structures. When running on Cray XC systems using the default configuration with the ugni comm layer and a craype-hugepages module loaded, the heap is used for all dynamic allocations except data space for arrays larger than 2 hugepages. (See Using the ugni Communications Layer, just above, for more about hugepages.) It is normally extended dynamically, as needed. But if desired, the heap can instead be created at a specified fixed size at the beginning of execution. In some cases this will reduce certain internal comm layer overheads and marginally improve performance.

The disadvantage of a fixed heap is that it usually produces worse NUMA affinity, it limits available heap memory to the specified fixed size, and it limits memory for arrays to whatever remains after the fixed-size heap is created. If either of the latter are less than what a program needs, it will terminate prematurely with an “Out of memory” message.

To specify a fixed heap, set the CHPL_RT_MAX_HEAP_SIZE environment variable to indicate its size. For the value of this variable you can use any of the following formats, where num is a positive integer number:

Format	Resulting Heap Size
num	num bytes
num[kK]	num * 2**10 bytes
num[mM]	num * 2**20 bytes
num[gG]	num * 2**30 bytes
num%	percentage of compute node physical memory

Any of the following would specify an approximately 1 GiB heap on a 128-GiB compute node, for example:

export CHPL_RT_MAX_HEAP_SIZE=1073741824
export CHPL_RT_MAX_HEAP_SIZE=1048576k
export CHPL_RT_MAX_HEAP_SIZE=1024m
export CHPL_RT_MAX_HEAP_SIZE=1g
export CHPL_RT_MAX_HEAP_SIZE=1% # 1.28 GiB, really

Note that the resulting heap size may get rounded up to match the page alignment. How much this will add, if any, depends on the hugepage size in any craype-hugepage module you have loaded at the time you execute the program. It may also be reduced, if some resource limitation prevents making the heap as large as requested.

ugni Communication Layer Registered Memory Regions

The ugni communication layer maintains information about every memory region it registers with Aries NIC. Roughly speaking there are a few memory regions for each tasking layer thread, plus one for each array larger than 2 hugepages allocated and registered separately from the heap. By default the comm layer can handle up to 16k (2**14) total memory regions, which is plenty under normal circumstances. In the event a program needs more than this, a message like the following will be printed:

warning: no more registered memory region table entries (max is 16384).
         Change using CHPL_RT_COMM_UGNI_MAX_MEM_REGIONS.

To provide for more registered regions, set the CHPL_RT_COMM_UGNI_MAX_MEM_REGIONS environment variable to a number indicating how many you want to allow. For example:

export CHPL_RT_COMM_UGNI_MAX_MEM_REGIONS=30000

Note that there are certain comm layer overheads that are proportional to the number of registered memory regions, so allowing a very high number of them may lead to reduced performance.

ugni Hugepage-related Warnings

Communication performance with ugni is so much better when hugepages are used that if you do not use them, the runtime will print the following warning when a multilocale program starts:

warning: without hugepages, communication performance will suffer

If you definitely do not want to use hugepages you can quiet this warning by giving the --quiet or -q option when you run the executable. Otherwise, load a hugepage module as described above in Using the ugni Communications Layer before running.

When you are using hugepages and do not have a fixed heap (that is, the CHPL_RT_MAX_HEAP_SIZE environment variable is not set), the Chapel runtime expects certain hugepage-related environment variables to have been set by the Chapel launcher. If you do not use a Chapel launcher you have to provide these settings yourself. Not doing so will result in one or both of the following messages:

warning: dynamic heap on hugepages needs HUGETLB_NO_RESERVE set to something
warning: dynamic heap on hugepages needs CHPL_JE_MALLOC_CONF set properly

To quiet these warnings, use the following settings:

export HUGETLB_NO_RESERVE=yes
export CHPL_JE_MALLOC_CONF=purge:decay,lg_chunk:log2HPS

where log2HPS is the base-2 log of the hugepage size. For example, with 16 MiB hugepages you would use:

export CHPL_JE_MALLOC_CONF=purge:decay,lg_chunk:24

gasnet Communication Layer

The GASNet-based communication layer discussed in the Multilocale Chapel Execution page can be used on all Cray systems. For best performance it should be used with native substrates and fixed segments, though even then its performance will rarely match that of the ugni communication layer. The relevant configurations are:

CHPL_COMM=gasnet
  CHPL_GASNET_SEGMENT=fast or large

In these configurations the heap is created with a fixed size at the beginning of execution. The default size works well in most cases but if it doesn’t a different size can be specified, as discussed in the following section.

Note that as of Chapel 2.4, the ‘aries’ substrate is no longer supported by GASNet. It is still possible to use the ‘ofi’ or ‘mpi’ substrates to target a Cray XC with GASNet, but this is not well supported or tested.

gasnet Communication Layer and the Heap

In contrast to the dynamic heap extension available in the ugni comm layer, when the gasnet comm layer is used with a native substrate for higher network performance, the runtime must know up front the maximum size the heap will grow to during execution.

In these cases the heap is used for all dynamic allocations, including arrays. By default it will occupy as much of the free memory on each compute node as the runtime can acquire, less some small amount to allow for demands from other (system) programs running there. Advanced users may want to make the heap smaller than the default. Programs start more quickly with a smaller heap, and in the unfortunate event that you need to produce core files, those will be written more quickly if the heap is smaller. Specify the heap size using the CHPL_RT_MAX_HEAP_SIZE environment variable, as discussed above in ugni Communication Layer and the Heap. But be aware that just as in the CHPL_COMM=ugni case, if you reduce the heap size to less than the amount your program actually needs and then run it, it will terminate prematurely due to not having enough memory.

Note that for CHPL_COMM=gasnet, CHPL_RT_MAX_HEAP_SIZE is synonymous with GASNET_MAX_SEGSIZE, and the former overrides the latter if both are set.

Known Constraints and Bugs

• Our PBS launcher explicitly supports PBS Pro, Moab/Torque, and the NCCS site versions of PBS. It may also work with other versions. If our PBS launcher does not work for you, you can fall back on a more manual launch of your program. For example, supposing the program is compiled to myprogram:

  • Launch the myprogram_real binary manually using aprun and your own qsub script or command.

  • Use ./myprogram --generate-qsub-script to generate a qsub script. Then edit the generated script and launch the myprogram_real binary manually as above.

• Redirecting stdin when executing a Chapel program under PBS/qsub may not work due to limitations of qsub.

• For EX and XC systems, there is a known issue with the Cray MPI release that causes some programs to assert and then hang during exit. A workaround is to set the environment variable, MPICH_GNI_DYNAMIC_CONN to disabled. Setting this environment variable affects all MPI programs, so remember to unset it after running your Chapel program.

• The amount of memory available to a Chapel program running over GASNet with the aries conduit is allocated at program start up. The default memory segment size may be too high on some platforms, resulting in an internal Chapel error or a GASNet initialization error such as:

  node 1 log gasnetc_init_segment() at $CHPL_HOME/third-party/gasnet/gasnet-src/aries-conduit/gasnet_aries.c:<line#>: MemRegister segment fault 8 at  0x2aab6ae00000 60000000, code GNI_RC_ERROR_RESOURCE
  

  If your Chapel program exits with such an error, try setting the environment variable CHPL_RT_MAX_HEAP_SIZE or GASNET_MAX_SEGSIZE to a lower value than the default (say 1G) and re-running your program. For more information, refer to the discussion of CHPL_RT_MAX_HEAP_SIZE above and/or the discussion of GASNET_MAX_SEGSIZE here:

  $CHPL_HOME/third-party/gasnet/gasnet-src/README