Annotation of /MITgcm_contrib/cg2d_bench/README

#	$Id$	
Benchmarking routine of the CG2D solver in MITgcm (barotropic solve)

To build:

a) Parameterization of SIZE.h:
sNx = size of tile in x-direction (ideally fits in cache, 30-60)
sNy = size of tile in y-direction (ideally fits in cache, 30-60)
OLx = overlap size in x-direction (1 or 3 usually)
OLy = overlap size in y-direction (1 or 3 usually)
b) Compilation
$CC $CFLAGS -c tim.c
$FC $DEFINES $INCLUDES $FCFLAGS -o cg2d *.F tim.o $LIBS -lm

$DEFINES:
1) For single precision add
-DUSE_SINGLE_PRECISION
2) For mixed (single for most ops, double for reductions) precision add
-DUSE_MIXED_PRECISION to -DUSE_SINGLE_PRECISION
3) Parallel (MPI) operation
-DALLOW_MPI -DUSE_MPI_INIT -DUSE_MPI_GSUM -DUSE_MPI_EXCH
4) Use MPI timing routines
-DUSE_MPI_TIME
5) Use of MPI_Sendrecv() instead of MPI_Isend/MPI_Irecv()/MPI_Waitall()
-DUSE_SNDRCV
6) Use of JAM for exchanges (not available without the hardware)
-DUSE_JAM_EXCH
7) Use of JAM for the global sum (not available without the hardware)
-DUSE_JAM_GSUM
8) In order to avoid doing the global sum in MPI do not define
-DUSE_MPI_GSUM
and all processors will see their own residual instead (dangerous)
9) In order to avoid doing the exchanges in MPI do not define
-DUSE_MPI_EXCH
and all processors avoid exchanging shadow regions (dangerous)
10) Performance counters
-DUSE_PAPI_FLOPS	To use PAPI to produce Mflop/s
or
-DUSE_PAPI_FLIPS	To use PAPI to produce Mflip/s
To produce this information for every iteration instead of each "timestep" 
add a -DPAPI_PER_ITERATION to the above
11) Extra (nearest neighbor) exchange steps to stress comms
-DTEN_EXTRA_EXCHS
12) Extra (global) sum steps to stress comms
-DHUNDRED_EXTRA_SUMS
13) 2D (PxQ) vs 1D decomposition
-DDECOMP2D
14) To output the residual every iteration:
-DRESIDUAL_PER_ITERATION

$INCLUDES (if using PAPI)
-I$PAPI_ROOT/include

$LIBS (if using PAPI - depending on the platform extra libs may be needed)
-L$PAPI_ROOT/lib -lpapi

c) Running

1) Allowing the system to choose the PxQ decomposition if setup for it:
mpiexec -n $NPROCS ./cg2d
2) Create a decomp.touse with the P & Q dimensions declared in the first
two lines as two integers, eg.

cat > decomp.touse << EOF
10
20
EOF

mpiexec -n 200 ./cg2d

1	ce107	1.1	# $Id$
2			Benchmarking routine of the CG2D solver in MITgcm (barotropic solve)
3
4			To build:
5
6			a) Parameterization of SIZE.h:
7			sNx = size of tile in x-direction (ideally fits in cache, 30-60)
8			sNy = size of tile in y-direction (ideally fits in cache, 30-60)
9			OLx = overlap size in x-direction (1 or 3 usually)
10			OLy = overlap size in y-direction (1 or 3 usually)
11			b) Compilation
12			$CC $CFLAGS -c tim.c
13			$FC $DEFINES $INCLUDES $FCFLAGS -o cg2d *.F tim.o $LIBS -lm
14
15			$DEFINES:
16			1) For single precision add
17			-DUSE_SINGLE_PRECISION
18			2) For mixed (single for most ops, double for reductions) precision add
19			-DUSE_MIXED_PRECISION to -DUSE_SINGLE_PRECISION
20			3) Parallel (MPI) operation
21			-DALLOW_MPI -DUSE_MPI_INIT -DUSE_MPI_GSUM -DUSE_MPI_EXCH
22			4) Use MPI timing routines
23			-DUSE_MPI_TIME
24			5) Use of MPI_Sendrecv() instead of MPI_Isend/MPI_Irecv()/MPI_Waitall()
25			-DUSE_SNDRCV
26			6) Use of JAM for exchanges (not available without the hardware)
27			-DUSE_JAM_EXCH
28			7) Use of JAM for the global sum (not available without the hardware)
29			-DUSE_JAM_GSUM
30			8) In order to avoid doing the global sum in MPI do not define
31			-DUSE_MPI_GSUM
32			and all processors will see their own residual instead (dangerous)
33			9) In order to avoid doing the exchanges in MPI do not define
34			-DUSE_MPI_EXCH
35			and all processors avoid exchanging shadow regions (dangerous)
36			10) Performance counters
37			-DUSE_PAPI_FLOPS To use PAPI to produce Mflop/s
38			or
39			-DUSE_PAPI_FLIPS To use PAPI to produce Mflip/s
40			To produce this information for every iteration instead of each "timestep"
41			add a -DPAPI_PER_ITERATION to the above
42			11) Extra (nearest neighbor) exchange steps to stress comms
43			-DTEN_EXTRA_EXCHS
44			12) Extra (global) sum steps to stress comms
45			-DHUNDRED_EXTRA_SUMS
46			13) 2D (PxQ) vs 1D decomposition
47			-DDECOMP2D
48			14) To output the residual every iteration:
49			-DRESIDUAL_PER_ITERATION
50
51			$INCLUDES (if using PAPI)
52			-I$PAPI_ROOT/include
53
54			$LIBS (if using PAPI - depending on the platform extra libs may be needed)
55			-L$PAPI_ROOT/lib -lpapi
56
57			c) Running
58
59			1) Allowing the system to choose the PxQ decomposition if setup for it:
60			mpiexec -n $NPROCS ./cg2d
61			2) Create a decomp.touse with the P & Q dimensions declared in the first
62			two lines as two integers, eg.
63
64			cat > decomp.touse << EOF
65			10
66			20
67			EOF
68
69			mpiexec -n 200 ./cg2d
70