SMPI Integration Testing

Proxy apps are scale models of real, massive HPC applications: each of them exhibits the same communication and computation patterns than the massive application that it stands for. But they last only a few thousands lines instead of some millions of lines. These proxy apps are usually provided for educational purpose, and also to ensure that the represented large HPC applications will correctly work with the next generation of runtimes and hardware. See also this explanation of the proxy apps' purpose.

SMPI is an implementation of MPI executing the application on top of the SimGrid simulator. This allows to study the application behavior in many details that would be hard to observe if not in a simulator, or on platforms that do not exist yet. The predictive power of SMPI was shown to be very good, provided that you correctly model the target platform [🗎, 🗎] while very scalable [🗎].

This repository gathers several collections of proxy apps, and reports their support by the SMPI implementation. This is tested nightly on a dedicated jenkins server.

Full Scale Applications

In addition to the proxy apps testing presented here, some real-scale HPC projects directly use SMPI as a basis for their regression and integration tests:

Application	Build status
BigDFT is a DFT massively parallel electronic structure code using a wavelet basis set.
StarPU is an unified runtime system for heterogeneous multicore architectures.
OpenMPI-test-suite is a test suite used by OpenMPI and several other MPI libraries
Intel-benchmarks-suite is a benchmarks suite by Intel
NPB is a test suite designed to help evaluate the performance of parallel supercomputers

ECP Proxy Application

Benchmark	Lines	Lang	Script	Status	Build Patch	Code Patch
AMG	4,658	C	🎫		▪️	▪️
CLAMR	109,477	C++	🎫		▪️	🎫
Chatterbug	1,050	C++	🎫		▪️	▪️
Comb	1,965	C++	🎫		▪️	🎫
CoMD	4,658	C	🎫		▪️	▪️
CoSP2	2,199	C	🎫		▪️	🎫
CloverLeaf	37,477	C, F90	🎫		▪️	▪️
CloverLeaf3D	11,200	C, F90	🎫		▪️	▪️
EBMS	841	C++, F90	🎫		▪️	▪️
Ember	1,300	C	🎫		▪️	▪️
ExaMiniMD	6,184	C++	🎫		🎫	only if sampling is needed : 🎫
CabanaMD	33,000	C++	🎫
HPCCG	1,548	C++	🎫		▪️	▪️
HPGMG	12,073	C	🎫		▪️	▪️
Kripke	5,350	C++	🎫		▪️	▪️
MINITRI	1,534	C++	🎫		▪️	▪️
MiniAero	4,645	C++	🎫		🎫	▪️
MiniAMR	8,329	C	🎫		▪️	▪️
MiniFE	19,907	C	🎫		▪️	▪️
MiniGhost	8,962	F90	🎫		🎫	▪️
MiniGMG	4,257	C	🎫		▪️	▪️
MiniMD	9,344	C++	🎫		🎫	▪️
MiniSMAC2D	8,329	F90	🎫		▪️	▪️
MiniVite	2,375	C++	🎫		▪️	▪️
Nekbone	38,674	C, F77	🎫		▪️	▪️
MiniXyce	2,020	C++	🎫		▪️	▪️
PENNANT	3,464	C++	🎫		🎫	▪️
PETSC	750,000	C, C++, F90	🎫		▪️	▪️
Picsar	57,800	F90	🎫		▪️	🎫
ProfugusMC	107,400	C++	🎫		▪️	▪️
Quicksilver	9,821	C++	🎫		▪️	🎫
SimpleMOC	2,864	C	🎫		🎫	▪️
SNAP	4,640	F90	🎫		▪️	▪️
SNBONE	5,897	F90	🎫		🎫	🎫
SWFFT	3,827	C++, C, F90	🎫		▪️	only if sampling is needed : 🎫, only for shared allocs : 🎫
Sw4lite	48,436	C	🎫		▪️	only if sampling is needed : 🎫, only for shared allocs : 🎫
TeaLeaf	5,729	F90	🎫		▪️	▪️
Tycho2	6,343	C++	🎫		▪️	🎫
VPFFT	3,384	C++	🎫		🎫	▪️
XSBench	1,086	C (OMP)	🎫		▪️	▪️

The following apps of this collection are not included:

• Requiring OpenMP only : PathFinder, CoHMM, NuT, RSBench, CoGL
• Laghos (depends on HYPRE, METIS, MFEM)
• FleCSALE (depends on Exodus, FleCSI, ParMETIS)
• PlasmApp (depends on Trilinos)
• LCALS, AMR_Exp_Parabolic (code source not available)
• ASPA, RajaPerformanceSuite : (no MPI)

CORAL benchmarks

Benchmark	Lines	Lang	Script	Status	Build Patch	Code Patch
amg2013	75,000	C	🎫		▪️	▪️
HACCIO	2,000	C++	🎫		▪️	▪️
KMI_HASH	4,239	C	🎫		▪️	▪️
LULESH	5,000	C	🎫		▪️	▪️
IOR	14,000	C	🎫		▪️	▪️
UMT2013	34,000	C, F90, C++	🎫		🎫	▪️
Qbox	45,000	C++	🎫		🎫	▪️
LSMS	200,000	C++, F90, C	🎫		🎫	▪️
MCB	13,000	C++	🎫		🎫	▪️
CLOMP	1,374	C	🎫		▪️	▪️

The following apps of this collection are not included:

• too heavy : QMCPACK (supported, but compilation takes 30+ minutes and several GB of disk space) (📖)
• BigSort : works, but needs to generate/read input files through MPI/IO calls. SMPI's MPI/IO implementation does not actually read/write files on the disk, so data can't be read (📖 📦)
• OpenMP only : Graph500 Memory benchmarks, FTQ, XSBench, NEKbonemk, HACCmk, UMTmk, AMGmk, MILCmk, GFMCmk.
• Using Python: Pynmaic.
• CAM-SE (📖📦) Depends on NETCDF, Lapack, cBlas.
• NAMD (📖📦) Requires mpixlc.

Trinity-Nersc benchmarks

Benchmark	Lines	Lang	Script	Status	Build Patch	Code Patch
SMB - mpiheader	418	C	🎫		🎫	▪️
SMB - msgrate	362	C	🎫		▪️	▪️
ziatest	255	C	🎫		▪️	▪️
psnap	384	C	🎫		▪️	▪️
mdtest	2,187	C	🎫		▪️	▪️
mpimemu	4,980	C	🎫		▪️	▪️
OMB_MPI	1,406	C	🎫		▪️	▪️
OMB_MPI (one-sided)	1,679	C	🎫		▪️	▪️
OMB_MPI (collective)	1,269	C	🎫		▪️	▪️
MiniFE	4,968	C++	🎫		▪️	▪️
GTC	5,591	F90	🎫		▪️	▪️
MILC	82,645	C	🎫		▪️	▪️
MiniDFT	30,874	C, F90	🎫		(with MKL)	▪️
Stream	747	C, F77	🎫		▪️	▪️

The following apps of this collection are not included:

• UPC-FC (📖📦): Depending on UPC
• Using Python: UMT.

CodeVault benchmarks

Benchmark	Lines	Lang	Script	Status	Build Patch	Code Patch
MCM	453	C++	🎫		▪️	▪️
n-Body_methods	2,401	C++	🎫		▪️	▪️
basic MPI IO	380	C++	🎫		▪️	▪️
read2shmem	202	C++	🎫		▪️	▪️
bhtree_mpi	1,200	C++	🎫		▪️	▪️

The following apps of this collection are not included:

• Requesting OpenMP: Dense_linear_algebra, N-body_methods (bhtree, hermite4, naive), Structured_grids.
• Using unsupported MPI primitives : Unstructed_grid (halo exchange)
• Sparse_linear_algebra (depends on PETSc)
• Spectral_methods (depends on OpenMP, CUDA, OpenCL, FFTW)
• Unstructured_grids ( depends on libmesh)

MeteoFrance Proxy Applications

HPL

Benchmark	Lines	Lang	SMPI	Patch
Vanilla	15,807	C		▪️
Optimized	15,807	C		▪️

Legend