Return to Section accelerate overview
5.USER-INTEL package¶
The USER-INTEL package was developed by Mike Brown at Intel Corporation. It provides two methods for accelerating simulations, depending on the hardware you have. The first is acceleration on Intel(R) CPUs by running in single, mixed, or double precision with vectorization. The second is acceleration on Intel(R) Xeon Phi(TM) coprocessors via offloading neighbor list and non-bonded force calculations to the Phi. The same C++ code is used in both cases. When offloading to a coprocessor from a CPU, the same routine is run twice, once on the CPU and once with an offload flag.
Note that the USER-INTEL package supports use of the Phi in “offload” mode, not “native” mode like the KOKKOS package.
Also note that the USER-INTEL package can be used in tandem with the USER-OMP package. This is useful when offloading pair style computations to the Phi, so that other styles not supported by the USER-INTEL package, e.g. bond, angle, dihedral, improper, and long-range electrostatics, can run simultaneously in threaded mode on the CPU cores. Since less MPI tasks than CPU cores will typically be invoked when running with coprocessors, this enables the extra CPU cores to be used for useful computation.
As illustrated below, if LAMMPS is built with both the USER-INTEL and USER-OMP packages, this dual mode of operation is made easier to use, via the “-suffix hybrid intel omp” command-line switch or the suffix hybrid intel omp command. Both set a second-choice suffix to “omp” so that styles from the USER-INTEL package will be used if available, with styles from the USER-OMP package as a second choice.
Here is a quick overview of how to use the USER-INTEL package for CPU acceleration, assuming one or more 16-core nodes. More details follow.
use an Intel compiler
use these CCFLAGS settings in Makefile.machine: -fopenmp, -DLAMMPS_MEMALIGN=64, -restrict, -xHost, -fno-alias, -ansi-alias, -override-limits
use these LINKFLAGS settings in Makefile.machine: -fopenmp, -xHost
make yes-user-intel yes-user-omp # including user-omp is optional
make mpi # build with the USER-INTEL package, if settings (including compiler) added to Makefile.mpi
make intel_cpu # or Makefile.intel_cpu already has settings, uses Intel MPI wrapper
Make.py -v -p intel omp -intel cpu -a file mpich_icc # or one-line build via Make.py for MPICH
Make.py -v -p intel omp -intel cpu -a file ompi_icc # or for OpenMPI
Make.py -v -p intel omp -intel cpu -a file intel_cpu # or for Intel MPI wrapper
lmp_machine -sf intel -pk intel 0 omp 16 -in in.script # 1 node, 1 MPI task/node, 16 threads/task, no USER-OMP
mpirun -np 32 lmp_machine -sf intel -in in.script # 2 nodess, 16 MPI tasks/node, no threads, no USER-OMP
lmp_machine -sf hybrid intel omp -pk intel 0 omp 16 -pk omp 16 -in in.script # 1 node, 1 MPI task/node, 16 threads/task, with USER-OMP
mpirun -np 32 -ppn 4 lmp_machine -sf hybrid intel omp -pk omp 4 -pk omp 4 -in in.script # 8 nodes, 4 MPI tasks/node, 4 threads/task, with USER-OMP
Here is a quick overview of how to use the USER-INTEL package for the same CPUs as above (16 cores/node), with an additional Xeon Phi(TM) coprocessor per node. More details follow.
Same as above for building, with these additions/changes:
add the flag -DLMP_INTEL_OFFLOAD to CCFLAGS in Makefile.machine
add the flag -offload to LINKFLAGS in Makefile.machine
for Make.py change "-intel cpu" to "-intel phi", and "file intel_cpu" to "file intel_phi"
mpirun -np 32 lmp_machine -sf intel -pk intel 1 -in in.script # 2 nodes, 16 MPI tasks/node, 240 total threads on coprocessor, no USER-OMP
mpirun -np 16 -ppn 8 lmp_machine -sf intel -pk intel 1 omp 2 -in in.script # 2 nodes, 8 MPI tasks/node, 2 threads/task, 240 total threads on coprocessor, no USER-OMP
mpirun -np 32 -ppn 8 lmp_machine -sf hybrid intel omp -pk intel 1 omp 2 -pk omp 2 -in in.script # 4 nodes, 8 MPI tasks/node, 2 threads/task, 240 total threads on coprocessor, with USER-OMP
Required hardware/software:
Your compiler must support the OpenMP interface. Use of an Intel(R) C++ compiler is recommended, but not required. However, g++ will not recognize some of the settings listed above, so they cannot be used. Optimizations for vectorization have only been tested with the Intel(R) compiler. Use of other compilers may not result in vectorization, or give poor performance.
The recommended version of the Intel(R) compiler is 14.0.1.106. Versions 15.0.1.133 and later are also supported. If using Intel(R) MPI, versions 15.0.2.044 and later are recommended.
To use the offload option, you must have one or more Intel(R) Xeon Phi(TM) coprocessors and use an Intel(R) C++ compiler.
Building LAMMPS with the USER-INTEL package:
The lines above illustrate how to include/build with the USER-INTEL package, for either CPU or Phi support, in two steps, using the “make” command. Or how to do it with one command via the src/Make.py script, described in Section 2.4 of the manual. Type “Make.py -h” for help. Because the mechanism for specifing what compiler to use (Intel in this case) is different for different MPI wrappers, 3 versions of the Make.py command are shown.
Note that if you build with support for a Phi coprocessor, the same binary can be used on nodes with or without coprocessors installed. However, if you do not have coprocessors on your system, building without offload support will produce a smaller binary.
If you also build with the USER-OMP package, you can use styles from both packages, as described below.
Note that the CCFLAGS and LINKFLAGS settings in Makefile.machine must include “-fopenmp”. Likewise, if you use an Intel compiler, the CCFLAGS setting must include “-restrict”. For Phi support, the “-DLMP_INTEL_OFFLOAD” (CCFLAGS) and “-offload” (LINKFLAGS) settings are required. The other settings listed above are optional, but will typically improve performance. The Make.py command will add all of these automatically.
If you are compiling on the same architecture that will be used for the runs, adding the flag -xHost to CCFLAGS enables vectorization with the Intel(R) compiler. Otherwise, you must provide the correct compute node architecture to the -x option (e.g. -xAVX).
Example machines makefiles Makefile.intel_cpu and Makefile.intel_phi are included in the src/MAKE/OPTIONS directory with settings that perform well with the Intel(R) compiler. The latter has support for offload to Phi coprocessors; the former does not.
Run with the USER-INTEL package from the command line:
The mpirun or mpiexec command sets the total number of MPI tasks used by LAMMPS (one or multiple per compute node) and the number of MPI tasks used per node. E.g. the mpirun command in MPICH does this via its -np and -ppn switches. Ditto for OpenMPI via -np and -npernode.
If you compute (any portion of) pairwise interactions using USER-INTEL pair styles on the CPU, or use USER-OMP styles on the CPU, you need to choose how many OpenMP threads per MPI task to use. If both packages are used, it must be done for both packages, and the same thread count value should be used for both. Note that the product of MPI tasks * threads/task should not exceed the physical number of cores (on a node), otherwise performance will suffer.
When using the USER-INTEL package for the Phi, you also need to specify the number of coprocessor/node and optionally the number of coprocessor threads per MPI task to use. Note that coprocessor threads (which run on the coprocessor) are totally independent from OpenMP threads (which run on the CPU). The default values for the settings that affect coprocessor threads are typically fine, as discussed below.
As in the lines above, use the “-sf intel” or “-sf hybrid intel omp” command-line switch, which will automatically append “intel” to styles that support it. In the second case, “omp” will be appended if an “intel” style does not exist.
Note that if either switch is used, it also invokes a default command: package intel 1. If the “-sf hybrid intel omp” switch is used, the default USER-OMP command package omp 0 is also invoked (if LAMMPS was built with USER-OMP). Both set the number of OpenMP threads per MPI task via the OMP_NUM_THREADS environment variable. The first command sets the number of Xeon Phi(TM) coprocessors/node to 1 (ignored if USER-INTEL is built for CPU-only), and the precision mode to “mixed” (default value).
You can also use the “-pk intel Nphi” command-line switch to explicitly set Nphi = # of Xeon Phi(TM) coprocessors/node, as well as additional options. Nphi should be >= 1 if LAMMPS was built with coprocessor support, otherswise Nphi = 0 for a CPU-only build. All the available coprocessor threads on each Phi will be divided among MPI tasks, unless the tptask option of the “-pk intel” command-line switch is used to limit the coprocessor threads per MPI task. See the package intel command for details, including the default values used for all its options if not specified, and how to set the number of OpenMP threads via the OMP_NUM_THREADS environment variable if desired.
If LAMMPS was built with the USER-OMP package, you can also use the “-pk omp Nt” command-line switch to explicitly set Nt = # of OpenMP threads per MPI task to use, as well as additional options. Nt should be the same threads per MPI task as set for the USER-INTEL package, e.g. via the “-pk intel Nphi omp Nt” command. Again, see the package omp command for details, including the default values used for all its options if not specified, and how to set the number of OpenMP threads via the OMP_NUM_THREADS environment variable if desired.
Or run with the USER-INTEL package by editing an input script:
The discussion above for the mpirun/mpiexec command, MPI tasks/node, OpenMP threads per MPI task, and coprocessor threads per MPI task is the same.
Use the suffix intel or suffix hybrid intel omp commands, or you can explicitly add an “intel” or “omp” suffix to individual styles in your input script, e.g.
pair_style lj/cut/intel 2.5
You must also use the package intel command, unless the “-sf intel” or “-pk intel” command-line switches were used. It specifies how many coprocessors/node to use, as well as other OpenMP threading and coprocessor options. The package doc page explains how to set the number of OpenMP threads via an environment variable if desired.
If LAMMPS was also built with the USER-OMP package, you must also use the package omp command to enable that package, unless the “-sf hybrid intel omp” or “-pk omp” command-line switches were used. It specifies how many OpenMP threads per MPI task to use (should be same as the setting for the USER-INTEL package), as well as other options. Its doc page explains how to set the number of OpenMP threads via an environment variable if desired.
Speed-ups to expect:
If LAMMPS was not built with coprocessor support (CPU only) when including the USER-INTEL package, then acclerated styles will run on the CPU using vectorization optimizations and the specified precision. This may give a substantial speed-up for a pair style, particularly if mixed or single precision is used.
If LAMMPS was built with coproccesor support, the pair styles will run on one or more Intel(R) Xeon Phi(TM) coprocessors (per node). The performance of a Xeon Phi versus a multi-core CPU is a function of your hardware, which pair style is used, the number of atoms/coprocessor, and the precision used on the coprocessor (double, single, mixed).
See the Benchmark page of the LAMMPS web site for performance of the USER-INTEL package on different hardware.
Note
Setting core affinity is often used to pin MPI tasks and OpenMP threads to a core or group of cores so that memory access can be uniform. Unless disabled at build time, affinity for MPI tasks and OpenMP threads on the host (CPU) will be set by default on the host when using offload to a coprocessor. In this case, it is unnecessary to use other methods to control affinity (e.g. taskset, numactl, I_MPI_PIN_DOMAIN, etc.). This can be disabled in an input script with the no_affinity option to the package intel command or by disabling the option at build time (by adding -DINTEL_OFFLOAD_NOAFFINITY to the CCFLAGS line of your Makefile). Disabling this option is not recommended, especially when running on a machine with hyperthreading disabled.
Guidelines for best performance on an Intel(R) Xeon Phi(TM) coprocessor:
- The default for the package intel command is to have all the MPI tasks on a given compute node use a single Xeon Phi(TM) coprocessor. In general, running with a large number of MPI tasks on each node will perform best with offload. Each MPI task will automatically get affinity to a subset of the hardware threads available on the coprocessor. For example, if your card has 61 cores, with 60 cores available for offload and 4 hardware threads per core (240 total threads), running with 24 MPI tasks per node will cause each MPI task to use a subset of 10 threads on the coprocessor. Fine tuning of the number of threads to use per MPI task or the number of threads to use per core can be accomplished with keyword settings of the package intel command.
- If desired, only a fraction of the pair style computation can be offloaded to the coprocessors. This is accomplished by using the balance keyword in the package intel command. A balance of 0 runs all calculations on the CPU. A balance of 1 runs all calculations on the coprocessor. A balance of 0.5 runs half of the calculations on the coprocessor. Setting the balance to -1 (the default) will enable dynamic load balancing that continously adjusts the fraction of offloaded work throughout the simulation. This option typically produces results within 5 to 10 percent of the optimal fixed balance.
- When using offload with CPU hyperthreading disabled, it may help performance to use fewer MPI tasks and OpenMP threads than available cores. This is due to the fact that additional threads are generated internally to handle the asynchronous offload tasks.
- If running short benchmark runs with dynamic load balancing, adding a short warm-up run (10-20 steps) will allow the load-balancer to find a near-optimal setting that will carry over to additional runs.
- If pair computations are being offloaded to an Intel(R) Xeon Phi(TM) coprocessor, a diagnostic line is printed to the screen (not to the log file), during the setup phase of a run, indicating that offload mode is being used and indicating the number of coprocessor threads per MPI task. Additionally, an offload timing summary is printed at the end of each run. When offloading, the frequency for atom sorting is changed to 1 so that the per-atom data is effectively sorted at every rebuild of the neighbor lists.
- For simulations with long-range electrostatics or bond, angle, dihedral, improper calculations, computation and data transfer to the coprocessor will run concurrently with computations and MPI communications for these calculations on the host CPU. The USER-INTEL package has two modes for deciding which atoms will be handled by the coprocessor. This choice is controlled with the ghost keyword of the package intel command. When set to 0, ghost atoms (atoms at the borders between MPI tasks) are not offloaded to the card. This allows for overlap of MPI communication of forces with computation on the coprocessor when the newton setting is “on”. The default is dependent on the style being used, however, better performance may be achieved by setting this option explictly.
Restrictions¶
When offloading to a coprocessor, hybrid styles that require skip lists for neighbor builds cannot be offloaded. Using hybrid/overlay is allowed. Only one intel accelerated style may be used with hybrid styles. Special_bonds exclusion lists are not currently supported with offload, however, the same effect can often be accomplished by setting cutoffs for excluded atom types to 0. None of the pair styles in the USER-INTEL package currently support the “inner”, “middle”, “outer” options for rRESPA integration via the run_style respa command; only the “pair” option is supported.