The Combined Parallel Subscripted Subscript (COMPASS) Benchmark Suite

This COMPASS suite is a collection of benchmarks exhibiting special patterns referred to as subscripted subscript patterns such as a[b[i]]. This repository serves as the artifact for our LCPC paper titled - COMPASS: A Combined Parallel Subscripted Subscript Benchmark Suite

The benchmarks have been taken from several well-known benchmark suites listed below.

Obtaining the Codes

This repository contains the benchmark codes listed below:

Code	Source	Original Source link
Amgmk-v1.0	CORAL Benchmark Codes	(https://asc.llnl.gov/coral-benchmarks)
UA-NPB-1.0.3	NAS Parallel Benchmarks	(https://github.com/akshay9594/SNU_NPB-1.0.3)
CHOLMOD	SuiteSparse	(https://github.com/DrTimothyAldenDavis/SuiteSparse)
IncompleteCholesky (C version)	Sparselib++	(https://github.com/king-mahdi/IEGenLib)
IS-NPB-1.0.3	NAS Parallel Benchmarks	(https://github.com/akshay9594/SNU_NPB-1.0.3)
SDDMM (C version)	Published Paper	(https://github.com/isratnisa/SDDMM_GPU)

The benchmarks can be obtained by cloning the git repository using the commands-

cd $HOME or cd ~
git clone (https://github.com/akshay9594/COMPASS.git)
cd COMPASS

Input Files

• The benchmarks : Amgmk, UA-NPB and IS-NPB use internally generated inputs
• The benchmarks : CHOLMOD, SDDMM and IncompleteCholesky require input sparse matrices (matrix market format) which can be obtained from the SuiteSparse matrix collection - (http://sparse.tamu.edu/)

Compiling and Running the benchmarks

There are 2 main directories - "serial" and "parallel" which contain the serial and parallel versions of the codes. The process of compiling and running the codes is the same for both versions of a benchmark.

Compilation

1. Compiling amgmk, IncompleteCholesky and SDDMM

   • Navigate to the source of these benchmarks and find the Makefile
   • Use the "make" command to run the Makefile and generate the executable.
   • "sddmm", "AMGMk" and "icm7" are the generated executables.

2. Compiling IS and UA

   • Navigate to the directory with the source files and compile using the Makefile.
   • The command to compile is: "make CLASS=A" (for input Class A)
   • The supported classes are A through C
   • The executable is placed in the "bin" sub-directory.

3. CHOLMOD Supernodal (SuiteSparse)

   • Navigate to the directories SuiteSparse_config, AMD, CAMD and COLAMD first.
   • Hit the command "make" to compile the source files in each of the above directories in the order mentioned.
   • Then navigate to the directory CHOLMOD and hit "make" to compile.
   • If there are issues during compilation, check the make file - SuiteSparse_config.mk within SuiteSparse_config directory.
   • The executable for CHOLMOD Supernodal (cholmod_demo) is placed within the "Demo" subdirectory within CHOLMOD main directory.

   Note: Ensure that you have the GNU fortran compiler installed on your system. Also check the path to the BLAS and LAPACK binaries (lines 164 and 180) within the Makefile SuiteSparse_config.mk.

Execution

1. Running amgmk:

   • Run this benchmark using command "./amgmk"
   • To change the dimensions of the input matrix, modify the values of the variables nx, ny and nz within the routine "test_matvec()" in main.c.

2. Running IncompleteCholesky and SDDMM

   • Place the downloaded input sparse matrix within the sud-directory "input" present in the root directory of each of these benchmarks.
   • Use the following commands for execution" (a) ./sddmm {path to OMP_Parallel or serial}/SDDMM/input/file_name.mtx (b) ./icm7 {path to OMP_Parallel or serial}/IncompleteCholesky/input/file_name.mtx

   where file_name.mtx is the input sparse matrix in the matrix market format.

3. CHOLMOD Supernodal (SuiteSparse)

   • The executable for CHOLMOD Supernodal (cholmod_demo) is placed within the "Demo" subdirectory within CHOLMOD main directory.
   • Execute using command: "./cholmod_demo {path to CHOLMOD}/Demo/Matrix/file_name.mtx"
   • The list of input sparse matrices to be used are mentioned in the LCPC paper linked above.