MAIN(argc, argv)
{
GOTO_REAL();
SETUP_NUMBER_TASKS(10);
/*
* Tuple for Scalable Data Generation
* stores startVertex, endVertex, long weight and other info
*/
graphSDG* SDGdata;
/*
* The graph data structure for this benchmark - see defs.h
*/
graph* G;
#ifdef ENABLE_KERNEL2
/*
* Kernel 2
*/
edge* maxIntWtList;
edge* soughtStrWtList;
long maxIntWtListSize;
long soughtStrWtListSize;
#endif /* ENABLE_KERNEL2 */
#ifdef ENABLE_KERNEL3
# ifndef ENABLE_KERNEL2
# error KERNEL3 requires KERNEL2
# endif
/*
* Kernel 3
*/
V* intWtVList = NULL;
V* strWtVList = NULL;
Vl** intWtVLList = NULL;
Vl** strWtVLList = NULL;
Vd* intWtVDList = NULL;
Vd* strWtVDList = NULL;
#endif /* ENABLE_KERNEL3 */
double totalTime = 0.0;
/* -------------------------------------------------------------------------
* Preamble
* -------------------------------------------------------------------------
*/
/*
* User Interface: Configurable parameters, and global program control
*/
getUserParameters(argc, (char** const) argv);
SIM_GET_NUM_CPU(THREADS);
TM_STARTUP(THREADS, 0);
P_MEMORY_STARTUP(THREADS);
SETUP_NUMBER_THREADS(THREADS);
thread_startup(THREADS);
double time_total = 0.0;
int repeat = REPEATS;
for (; repeat > 0; --repeat) {
SDGdata = (graphSDG*)malloc(sizeof(graphSDG));
assert(SDGdata);
genScalData_seq(SDGdata);
G = (graph*)malloc(sizeof(graph));
assert(G);
computeGraph_arg_t computeGraphArgs;
computeGraphArgs.GPtr = G;
computeGraphArgs.SDGdataPtr = SDGdata;
TIMER_T start;
TIMER_READ(start);
GOTO_SIM();
thread_start(computeGraph, (void*)&computeGraphArgs);
GOTO_REAL();
TIMER_T stop;
TIMER_READ(stop);
double time_tmp = TIMER_DIFF_SECONDS(start, stop);
PRINT_STATS();
time_total += time_tmp;
}
totalTime += time_total;
#ifdef ENABLE_KERNEL2
/* -------------------------------------------------------------------------
* Kernel 2 - Find Max weight and sought string
* -------------------------------------------------------------------------
*/
printf("\nKernel 2 - getStartLists() beginning execution...\n");
maxIntWtListSize = 0;
soughtStrWtListSize = 0;
maxIntWtList = (edge*)malloc(sizeof(edge));
assert(maxIntWtList);
soughtStrWtList = (edge*)malloc(sizeof(edge));
assert(soughtStrWtList);
getStartLists_arg_t getStartListsArg;
getStartListsArg.GPtr = G;
getStartListsArg.maxIntWtListPtr = &maxIntWtList;
getStartListsArg.maxIntWtListSize = &maxIntWtListSize;
getStartListsArg.soughtStrWtListPtr = &soughtStrWtList;
getStartListsArg.soughtStrWtListSize = &soughtStrWtListSize;
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
getStartLists((void*)&getStartListsArg);
}
#else
thread_start(getStartLists, (void*)&getStartListsArg);
#endif
GOTO_REAL();
TIMER_T stop;
TIMER_READ(stop);
time = TIMER_DIFF_SECONDS(start, stop);
totalTime += time;
printf("\n\tgetStartLists() completed execution.\n");
printf("\nTime taken for kernel 2 is %9.6f sec.\n\n", time);
#endif /* ENABLE_KERNEL2 */
#ifdef ENABLE_KERNEL3
/* -------------------------------------------------------------------------
* Kernel 3 - Graph Extraction
* -------------------------------------------------------------------------
*/
printf("\nKernel 3 - findSubGraphs() beginning execution...\n");
if (K3_DS == 0) {
intWtVList = (V*)malloc(G->numVertices * maxIntWtListSize * sizeof(V));
assert(intWtVList);
strWtVList = (V*)malloc(G->numVertices * soughtStrWtListSize * sizeof(V));
assert(strWtVList);
findSubGraphs0_arg_t findSubGraphs0Arg;
findSubGraphs0Arg.GPtr = G;
findSubGraphs0Arg.intWtVList = intWtVList;
findSubGraphs0Arg.strWtVList = strWtVList;
findSubGraphs0Arg.maxIntWtList = maxIntWtList;
findSubGraphs0Arg.maxIntWtListSize = maxIntWtListSize;
findSubGraphs0Arg.soughtStrWtList = soughtStrWtList;
findSubGraphs0Arg.soughtStrWtListSize = soughtStrWtListSize;
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
findSubGraphs0((void*)&findSubGraphs0Arg);
}
#else
thread_start(findSubGraphs0, (void*)&findSubGraphs0Arg);
#endif
GOTO_REAL();
TIMER_READ(stop);
} else if (K3_DS == 1) {
intWtVLList = (Vl**)malloc(maxIntWtListSize * sizeof(Vl*));
assert(intWtVLList);
strWtVLList = (Vl**)malloc(soughtStrWtListSize * sizeof(Vl*));
assert(strWtVLList);
findSubGraphs1_arg_t findSubGraphs1Arg;
findSubGraphs1Arg.GPtr = G;
findSubGraphs1Arg.intWtVLList = intWtVLList;
findSubGraphs1Arg.strWtVLList = strWtVLList;
findSubGraphs1Arg.maxIntWtList = maxIntWtList;
findSubGraphs1Arg.maxIntWtListSize = maxIntWtListSize;
findSubGraphs1Arg.soughtStrWtList = soughtStrWtList;
findSubGraphs1Arg.soughtStrWtListSize = soughtStrWtListSize;
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
findSubGraphs1((void*)&findSubGraphs1Arg);
}
#else
thread_start(findSubGraphs1, (void*)&findSubGraphs1Arg);
#endif
GOTO_REAL();
TIMER_READ(stop);
/* Verification
on_one_thread {
for (i=0; i<maxIntWtListSize; i++) {
printf("%ld -- ", i);
currV = intWtVLList[i];
while (currV != NULL) {
printf("[%ld %ld] ", currV->num, currV->depth);
currV = currV->next;
}
printf("\n");
}
for (i=0; i<soughtStrWtListSize; i++) {
printf("%ld -- ", i);
currV = strWtVLList[i];
while (currV != NULL) {
printf("[%ld %ld] ", currV->num, currV->depth);
currV = currV->next;
}
printf("\n");
}
}
*/
} else if (K3_DS == 2) {
intWtVDList = (Vd *) malloc(maxIntWtListSize * sizeof(Vd));
assert(intWtVDList);
strWtVDList = (Vd *) malloc(soughtStrWtListSize * sizeof(Vd));
assert(strWtVDList);
findSubGraphs2_arg_t findSubGraphs2Arg;
findSubGraphs2Arg.GPtr = G;
findSubGraphs2Arg.intWtVDList = intWtVDList;
findSubGraphs2Arg.strWtVDList = strWtVDList;
findSubGraphs2Arg.maxIntWtList = maxIntWtList;
findSubGraphs2Arg.maxIntWtListSize = maxIntWtListSize;
findSubGraphs2Arg.soughtStrWtList = soughtStrWtList;
findSubGraphs2Arg.soughtStrWtListSize = soughtStrWtListSize;
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
findSubGraphs2((void*)&findSubGraphs2Arg);
}
#else
thread_start(findSubGraphs2, (void*)&findSubGraphs2Arg);
#endif
GOTO_REAL();
TIMER_READ(stop);
/* Verification */
/*
on_one_thread {
printf("\nInt weight sub-graphs \n");
for (i=0; i<maxIntWtListSize; i++) {
printf("%ld -- ", i);
for (j=0; j<intWtVDList[i].numArrays; j++) {
printf("\n [Array %ld] - \n", j);
for (k=0; k<intWtVDList[i].arraySize[j]; k++) {
printf("[%ld %ld] ", intWtVDList[i].vList[j][k].num, intWtVDList[i].vList[j][k].depth);
}
}
printf("\n");
}
printf("\nStr weight sub-graphs \n");
for (i=0; i<soughtStrWtListSize; i++) {
printf("%ld -- ", i);
for (j=0; j<strWtVDList[i].numArrays; j++) {
printf("\n [Array %ld] - \n", j);
for (k=0; k<strWtVDList[i].arraySize[j]; k++) {
printf("[%ld %ld] ", strWtVDList[i].vList[j][k].num, strWtVDList[i].vList[j][k].depth);
}
}
printf("\n");
}
}
*/
} else {
assert(0);
}
time = TIMER_DIFF_SECONDS(start, stop);
totalTime += time;
printf("\n\tfindSubGraphs() completed execution.\n");
printf("\nTime taken for kernel 3 is %9.6f sec.\n\n", time);
#endif /* ENABLE_KERNEL3 */
#ifdef ENABLE_KERNEL4
/* -------------------------------------------------------------------------
* Kernel 4 - Graph Clustering
* -------------------------------------------------------------------------
*/
printf("\nKernel 4 - cutClusters() beginning execution...\n");
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
cutClusters((void*)G);
}
#else
thread_start(cutClusters, (void*)G);
#endif
GOTO_REAL();
TIMER_READ(stop);
time = TIMER_DIFF_SECONDS(start, stop);
totalTime += time;
printf("\n\tcutClusters() completed execution.\n");
printf("\nTime taken for Kernel 4 is %9.6f sec.\n\n", time);
#endif /* ENABLE_KERNEL4 */
printf("Time = %9.6f \n", totalTime);
/* -------------------------------------------------------------------------
* Cleanup
* -------------------------------------------------------------------------
*/
P_FREE(G->outDegree);
P_FREE(G->outVertexIndex);
P_FREE(G->outVertexList);
P_FREE(G->paralEdgeIndex);
P_FREE(G->inDegree);
P_FREE(G->inVertexIndex);
P_FREE(G->inVertexList);
P_FREE(G->intWeight);
P_FREE(G->strWeight);
#ifdef ENABLE_KERNEL3
LONGINT_T i;
LONGINT_T j;
Vl* currV;
Vl* tempV;
if (K3_DS == 0) {
P_FREE(strWtVList);
P_FREE(intWtVList);
}
if (K3_DS == 1) {
for (i = 0; i < maxIntWtListSize; i++) {
currV = intWtVLList[i];
while (currV != NULL) {
tempV = currV->next;
P_FREE(currV);
currV = tempV;
}
}
for (i = 0; i < soughtStrWtListSize; i++) {
currV = strWtVLList[i];
while (currV != NULL) {
tempV = currV->next;
P_FREE(currV);
currV = tempV;
}
}
P_FREE(strWtVLList);
P_FREE(intWtVLList);
}
if (K3_DS == 2) {
for (i = 0; i < maxIntWtListSize; i++) {
for (j = 0; j < intWtVDList[i].numArrays; j++) {
P_FREE(intWtVDList[i].vList[j]);
}
P_FREE(intWtVDList[i].vList);
P_FREE(intWtVDList[i].arraySize);
}
for (i = 0; i < soughtStrWtListSize; i++) {
for (j = 0; j < strWtVDList[i].numArrays; j++) {
P_FREE(strWtVDList[i].vList[j]);
}
P_FREE(strWtVDList[i].vList);
P_FREE(strWtVDList[i].arraySize);
}
P_FREE(strWtVDList);
P_FREE(intWtVDList);
}
P_FREE(soughtStrWtList);
P_FREE(maxIntWtList);
#endif /* ENABLE_KERNEL2 */
P_FREE(SOUGHT_STRING);
P_FREE(G);
P_FREE(SDGdata);
TM_SHUTDOWN();
P_MEMORY_SHUTDOWN();
GOTO_SIM();
thread_shutdown();
MAIN_RETURN(0);
}
int main(int argc, char** argv)
{
/* Data structure for storing generated tuples in the
* Scalable Data Generation Stage -- see defs.h */
graphSDG* SDGdata;
/* The graph data structure -- see defs.h */
graph* G;
/* Kernel 2 output */
edge *maxIntWtList;
INT_T maxIntWtListSize;
/* Kernel 4 output */
DOUBLE_T *BC;
DOUBLE_T elapsed_time;
int filter, cutshort;
#ifdef _OPENMP
if (argc < 3) {
fprintf(stderr, "Usage: ./SSCA2 <No. of threads> <SCALE>\n");
exit(-1);
}
NUM_THREADS = atoi(argv[1]);
SCALE = atoi(argv[2]);
omp_set_num_threads(NUM_THREADS);
filter = atoi(argv[3]);
cutshort = atoi(argv[4]);
#else
if (argc < 2) {
fprintf(stderr, "Usage: ./SSCA2 <SCALE>\n");
exit(-1);
}
SCALE = atoi(argv[1]);
filter = atoi(argv[2]);
cutshort = atoi(argv[3]);
#endif
/* ------------------------------------ */
/* Initialization -- Untimed */
/* ------------------------------------ */
fprintf(stderr, "\nHPCS SSCA Graph Analysis Benchmark v2.2\n");
fprintf(stderr, "Running...\n\n");
init(SCALE, cutshort);
#ifdef _OPENMP
fprintf(stderr, "\nNo. of threads: %d\n", NUM_THREADS);
#endif
fprintf(stderr, "SCALE: %d\n\n", SCALE);
/* -------------------------------------------- */
/* Scalable Data Generator -- Untimed */
/* -------------------------------------------- */
#ifndef VERIFYK4
fprintf(stderr, "Scalable Data Generator -- ");
fprintf(stderr, "genScalData() beginning execution...\n");
SDGdata = (graphSDG *) malloc(sizeof(graphSDG));
elapsed_time = genScalData(SDGdata);
fprintf(stderr, "\n\tgenScalData() completed execution\n");
fprintf(stderr,
"\nTime taken for Scalable Data Generation is %9.6lf sec.\n\n",
elapsed_time);
#else
fprintf(stderr, "Generating 2D torus for Kernel 4 validation -- ");
fprintf(stderr, "gen2DTorus() beginning execution...\n");
SDGdata = (graphSDG *) malloc(sizeof(graphSDG));
elapsed_time = gen2DTorus(SDGdata);
fprintf(stderr, "\n\tgen2DTorus() completed execution\n");
fprintf(stderr,
"\nTime taken for 2D torus generation is %9.6lf sec.\n\n",
elapsed_time);
#endif
/* ------------------------------------ */
/* Kernel 1 - Graph Construction */
/* ------------------------------------ */
/* From the SDG data, construct the graph 'G' */
fprintf(stderr, "\nKernel 1 -- computeGraph() beginning execution...\n");
G = (graph *) malloc(sizeof(graph));
/* Store the SDG edge lists in a compact representation
* which isn't modified in subsequent Kernels */
elapsed_time = computeGraph(G, SDGdata);
fprintf(stderr, "\n\tcomputeGraph() completed execution\n");
fprintf(stderr, "\nTime taken for Kernel 1 is %9.6lf sec.\n\n",
elapsed_time);
free(SDGdata);
/* ---------------------------------------------------- */
/* Kernel 2 - Find max edge weight */
/* ---------------------------------------------------- */
fprintf(stderr, "\nKernel 2 -- getStartLists() beginning execution...\n");
/* Initialize vars and allocate temp. memory for the edge list */
maxIntWtListSize = 0;
maxIntWtList = (edge *) malloc(sizeof(edge));
elapsed_time = getStartLists(G, &maxIntWtList, &maxIntWtListSize);
fprintf(stderr, "\n\tgetStartLists() completed execution\n\n");
fprintf(stderr, "Max. int wt. list size is %d\n", maxIntWtListSize);
fprintf(stderr, "\nTime taken for Kernel 2 is %9.6lf sec.\n\n",
elapsed_time);
/* ------------------------------------ */
/* Kernel 3 - Graph Extraction */
/* ------------------------------------ */
fprintf(stderr, "\nKernel 3 -- findSubGraphs() beginning execution...\n");
elapsed_time = findSubGraphs(G, maxIntWtList, maxIntWtListSize);
fprintf(stderr, "\n\tfindSubGraphs() completed execution\n");
fprintf(stderr, "\nTime taken for Kernel 3 is %9.6lf sec.\n\n",
elapsed_time);
free(maxIntWtList);
/* ------------------------------------------ */
/* Kernel 4 - Betweenness Centrality */
/* ------------------------------------------ */
fprintf(stderr, "\nKernel 4 -- betweennessCentrality() "
"beginning execution...\n");
BC = (DOUBLE_T *) calloc(N, sizeof(DOUBLE_T));
elapsed_time = betweennessCentrality(G, BC, filter);
fprintf(stderr, "\nTime taken for Kernel 4 is %9.6f sec.\n\n",
elapsed_time);
fprintf(stderr, "TEPS score for Kernel 4 is %lf\n\n",
7*N*(1<<K4approx)/elapsed_time);
free(BC);
/* -------------------------------------------------------------------- */
free(G->numEdges);
free(G->endV);
free(G->weight);
free(G);
return 0;
}
