1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
|
#include <mpi.h>
#include <sys/time.h>
#include <iostream>
#include <functional>
#include <algorithm>
#include <vector>
#include <string>
#include <sstream>
#include <stdint.h>
#include <cmath>
#include "CombBLAS/CombBLAS.h"
#include "Glue.h"
#include "CCGrid.h"
#include "Reductions.h"
#include "Multiplier.h"
#include "SplitMatDist.h"
using namespace std;
using namespace combblas;
double comm_bcast;
double comm_reduce;
double comp_summa;
double comp_reduce;
double comp_result;
double comp_reduce_layer;
double comp_split;
double comp_trans;
double comm_split;
#define ITERS 5
int main(int argc, char *argv[])
{
int provided;
MPI_Init_thread(&argc, &argv, MPI_THREAD_SERIALIZED, &provided);
if (provided < MPI_THREAD_SERIALIZED)
{
printf("ERROR: The MPI library does not have MPI_THREAD_SERIALIZED support\n");
MPI_Abort(MPI_COMM_WORLD, 1);
}
int nprocs, myrank;
MPI_Comm_size(MPI_COMM_WORLD,&nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&myrank);
if(argc < 8)
{
if(myrank == 0)
{
printf("Usage (random): ./mpipspgemm <GridRows> <GridCols> <Layers> <Type> <Scale> <EDGEFACTOR> <algo>\n");
printf("Usage (input): ./mpipspgemm <GridRows> <GridCols> <Layers> <Type=input> <matA> <matB> <algo>\n");
printf("Example: ./mpipspgemm 4 4 2 ER 19 16 outer\n");
printf("Example: ./mpipspgemm 4 4 2 Input matA.mtx matB.mtx column\n");
printf("Type ER: Erdos-Renyi\n");
printf("Type SSCA: R-MAT with SSCA benchmark parameters\n");
printf("Type G500: R-MAT with Graph500 benchmark parameters\n");
printf("algo: outer | column \n");
}
return -1;
}
unsigned GRROWS = (unsigned) atoi(argv[1]);
unsigned GRCOLS = (unsigned) atoi(argv[2]);
unsigned C_FACTOR = (unsigned) atoi(argv[3]);
CCGrid CMG(C_FACTOR, GRCOLS);
int nthreads;
#pragma omp parallel
{
nthreads = omp_get_num_threads();
}
if(GRROWS != GRCOLS)
{
SpParHelper::Print("This version of the Combinatorial BLAS only works on a square logical processor grid\n");
MPI_Barrier(MPI_COMM_WORLD);
MPI_Abort(MPI_COMM_WORLD, 1);
}
int layer_length = GRROWS*GRCOLS;
if(layer_length * C_FACTOR != nprocs)
{
SpParHelper::Print("The product of <GridRows> <GridCols> <Replicas> does not match the number of processes\n");
MPI_Barrier(MPI_COMM_WORLD);
MPI_Abort(MPI_COMM_WORLD, 1);
}
{
SpDCCols<int64_t, double> splitA, splitB;
SpDCCols<int64_t, double> *splitC;
string type;
shared_ptr<CommGrid> layerGrid;
layerGrid.reset( new CommGrid(CMG.layerWorld, 0, 0) );
FullyDistVec<int64_t, int64_t> p(layerGrid); // permutation vector defined on layers
if(string(argv[4]) == string("input")) // input option
{
string fileA(argv[5]);
string fileB(argv[6]);
double t01 = MPI_Wtime();
SpDCCols<int64_t, double> *A = ReadMat<double>(fileA, CMG, true, p);
SpDCCols<int64_t, double> *B = ReadMat<double>(fileB, CMG, true, p);
SplitMat(CMG, A, splitA, false);
SplitMat(CMG, B, splitB, true); //row-split
if(myrank == 0) cout << "Matrices read and replicated along layers : time " << MPI_Wtime() - t01 << endl;
}
else
{
unsigned scale = (unsigned) atoi(argv[5]);
unsigned EDGEFACTOR = (unsigned) atoi(argv[6]);
double initiator[4];
if(string(argv[4]) == string("ER"))
{
initiator[0] = .25;
initiator[1] = .25;
initiator[2] = .25;
initiator[3] = .25;
}
else if(string(argv[4]) == string("G500"))
{
initiator[0] = .57;
initiator[1] = .19;
initiator[2] = .19;
initiator[3] = .05;
EDGEFACTOR = 16;
}
else if(string(argv[4]) == string("SSCA"))
{
initiator[0] = .6;
initiator[1] = .4/3;
initiator[2] = .4/3;
initiator[3] = .4/3;
EDGEFACTOR = 8;
}
else {
if(myrank == 0)
printf("The initiator parameter - %s - is not recognized.\n", argv[5]);
MPI_Abort(MPI_COMM_WORLD, 1);
}
double t01 = MPI_Wtime();
SpDCCols<int64_t, double> *A = GenMat<int64_t,double>(CMG, scale, EDGEFACTOR, initiator, true);
SpDCCols<int64_t, double> *B = GenMat<int64_t,double>(CMG, scale, EDGEFACTOR, initiator, true);
SplitMat(CMG, A, splitA, false);
SplitMat(CMG, B, splitB, true); //row-split
if(myrank == 0) cout << "RMATs Generated and replicated along layers : time " << MPI_Wtime() - t01 << endl;
}
int64_t globalnnzA=0, globalnnzB=0;
int64_t localnnzA = splitA.getnnz();
int64_t localnnzB = splitB.getnnz();
MPI_Allreduce( &localnnzA, &globalnnzA, 1, MPIType<int64_t>(), MPI_SUM, MPI_COMM_WORLD);
MPI_Allreduce( &localnnzB, &globalnnzB, 1, MPIType<int64_t>(), MPI_SUM, MPI_COMM_WORLD);
if(myrank == 0) cout << "After split: nnzA= " << globalnnzA << " & nnzB= " << globalnnzB;
type = string(argv[7]);
if(myrank == 0)
{
printf("\n Processor Grid (row x col x layers x threads): %dx%dx%dx%d \n", CMG.GridRows, CMG.GridCols, CMG.GridLayers, nthreads);
printf(" prow pcol layer thread comm_bcast comm_scatter comp_summa comp_merge comp_scatter comp_result other total\n");
}
if(type == string("outer"))
{
splitB.Transpose(); //locally transpose for outer product
for(int k=0; k<ITERS; k++)
{
splitC = multiply(splitA, splitB, CMG, true, false); // outer product
delete splitC;
}
}
else // default column-threaded
{
for(int k=0; ITERS>0 && k<ITERS-1; k++)
{
splitC = multiply(splitA, splitB, CMG, false, true);
delete splitC;
}
splitC = multiply(splitA, splitB, CMG, false, true);
int64_t nnzC=0;
int64_t localnnzC = splitC->getnnz();
MPI_Allreduce( &localnnzC, &nnzC, 1, MPIType<int64_t>(), MPI_SUM, MPI_COMM_WORLD);
if(myrank == 0) cout << "\n After multiplication: nnzC= " << nnzC << endl << endl;
delete splitC;
}
}
MPI_Finalize();
return 0;
}
|
