MG_SERIAL
NAS Parallel Benchmark MG, Serial Version, FORTRAN77

MG_SERIAL is a FORTRAN77 program which is a serial implementation of the NAS Parallel Benchmark known as MG ("MultiGrid").

MG_SERIAL demonstrates the capabilities of a very simple multigrid solver in computing a three dimensional potential field. This is a simplified multigrid solver in two important respects:

1. it solves only a constant coefficient equation, and that only on a uniform cubical grid;
2. it solves only a single equation, representing a scalar field rather than a vector field.

The NAS Parallel Benchmarks are "paper and pencil" specifications; that is, the proposers have laid down what the programs must do algorithmically, but only very general rules about computer language implementation have been given. Thus, a user or vendor has enormous freedom in implementing the benchmark program in a way that is optimal for a given architecture or parallel programming scheme.

As a starting point for implementors, the proposers have supplied certain initial versions of the benchmark programs.

This directory contains material derived from one such initial benchmark code. Persons interested in the original source of the benchmark codes should NOT work from this material, but should go to the NAS Parallel Benchmark website directly!

The benchmark programs are available in SERIAL, MPI and OPEN_MP versions, and in a sequence of "classes" of increasing size: S, W, A, B, C, D and E.

Statistics for classes S, W, A, B, and C are available, as run on an Apple Intel, using this SERIAL FORTRAN77 version of the program.

Class	Size	Time (sec)	MFLOPS
S	32^3	0.02	404
W	128^3	1.15	423
A	256^3	9.12	426
B	256^3	42.59	456
C	512^3	?	?

This directory contains the serial version of the benchmark, along with include files that allow it to run in some of the smaller sizes.

The web site for the NAS Parallel Benchmarks is http://www.nas.nasa.gov/Resources/Software/npb.html.

Languages:

MG_SERIAL is available in a FORTRAN77 version.

Related Data and Programs:

BT_SERIAL, a FORTRAN77 program which runs the serial version of the NAS Parallel Benchmark known as "BT" (block tridiagonal solver for 3D PDE).

CG_SERIAL, a FORTRAN77 program which is a serial version of the NAS Parallel Benchmark CG (conjugate gradient solver).

EP_SERIAL, a FORTRAN77 program which is a serial version of the NAS Parallel Benchmark EP (embarassingly parallel).

FIRE_SERIAL, a FORTRAN77program which simulates a forest fire over a rectangular array of trees, starting at a single random location. It is intended as a starting point for the development of a parallel version.

IS_SERIAL, a C program which is a serial version of the NAS Parallel Benchmark IS (integer sort).

LINPACK_BENCH, a FORTRAN77 program which measures the time taken by LINPACK to solve a particular linear system.

MATMUL, a FORTRAN77 program which is an interactive matrix multiplication benchmark.

MDBNCH, a FORTRAN77 program which is a benchmark molecular simulation calculation.

MXM_SERIAL, a FORTRAN77 program which sets up a matrix multiplication problem A=B*C, intended as a starting point for implementing a parallel version.

POISSON_SERIAL, a FORTRAN77 program which computes an approximate solution to the Poisson equation in a rectangle, and is intended as the starting point for the creation of a parallel version.

PRIME_SERIAL, a FORTRAN77 program which counts the number of primes between 1 and N, intended as a starting point for the creation of a parallel version.

SEARCH_SERIAL, a FORTRAN77 program which searches the integers from A to B for a value J such that F(J) = C. this version of the program is intended as a starting point for a parallel approach.

SUBSET_SUM_SERIAL, a FORTRAN77 program which seeks solutions of the subset sum problem, in which it is desired to find a subset of a set of integers which has a given sum; this version of the program is intended as a starting point for a parallel approach.

Reference:

1. David Bailey, Eric Barszcz, John Barton, D Browning, Robert Carter, Leonardo Dagum, Rod Fatoohi, Samuel Fineberg, Paul Frederickson, Thomas Lasinski, Robert Schreiber, Horst Simon, V Venkatakrishnan, Sisira Weeratunga,
   The NAS Parallel Benchmarks,
   International Journal of High Performane Computing Applications,
   Volume 5, Number 3, September 1991, pages 63-73.

Source Code:

• mg_serial.f, the source code.
• globals.h, an include file.
• print_results.f, prints the benchmarking results.
• randdp.f, a double precision random number generator.
• timers.f, interfaces to the timing routine.
• wtime.c, a small routine used for timing.
• wtime.h, an include file for WTIME.

Examples and tests:

MG_SERIAL_S runs the benchmark in serial mode with problem size S.

• nbparams_S.h, the include file.
• mg_serial_S.sh, a script which compiles the code.
• mg_serial_S_output.txt, the output file.

MG_SERIAL_W runs the benchmark in serial mode with problem size W.

• nbparams_W.h, the include file.
• mg_serial_W.sh, a script which compiles the code.
• mg_serial_W_output.txt, the output file.

MG_SERIAL_A runs the benchmark in serial mode with problem size A.

• nbparams_A.h, the include file.
• mg_serial_A.sh, a script which compiles the code.
• mg_serial_A_output.txt, the output file.

MG_SERIAL_B runs the benchmark in serial mode with problem size B.

• nbparams_B.h, the include file.
• mg_serial_B.sh, a script which compiles the code.
• mg_serial_B_output.txt, the output file.

MG_SERIAL_C runs the benchmark in serial mode with problem size C.

• nbparams_C.h, the include file.
• mg_serial_C.sh, a script which compiles the code.
• mg_serial_C_output.txt, the output file.

List of Routines:

• MG is the main program.
• BUBBLE does a bubble sort in direction dir
• COMM3 organizes the communication on all borders
• ELAPSED_TIME returns a wall clock reading.
• INTERP adds the coarse grid correction to the current approximation.
• MG3P is the multigrid V-cycle routine.
• NORM2U3 approximates the L2 and uniform norms.
• POWER raises an integer to an integer power.
• PRINT_RESULTS prints benchmarking information.
• PSINV applies an approximate inverse as smoother: u = u + Cr
• RANDLC returns a uniform pseudorandom double precision value.
• REP_NRM reports on the norms.
• RESID computes the residual: r = v - Au
• RPRJ3 projects onto the next coarser grid.
• SETUP does initialization.
• SHOWALL prints a 3D array, or a portion of it.
• TIMER_CLEAR clears the timer for the N-th event.
• TIMER_READ gets the timing of the N-th event.
• TIMER_START starts the timer for the N-th event.
• TIMER_STOP stops the timer for the N-th event.
• VRANLC returns a vector of uniform pseudorandom double precision values.
• ZERO3 zeros out a 3D array.
• ZRAN3 applies a random perturbation.

You can go up one level to the FORTRAN77 source codes.