int main(int argc, char **argv) {
///******************************************************
///********************** INPUT *************************
///******************************************************
if (argc != 4) {
std::cout << "Invalid number of arguments!" << std::endl;
std::cout << "./compare A.out B.out" << std::endl;
exit(EXIT_FAILURE);
}
//matrix dimensions
int dimM = 0;
int dimN = 0;
int dimO = 0;
//get dimensions
std::ifstream fIn(argv[1]);
if (!fIn) {
std::cout << "Error opening file: " << argv[1] << std::endl;
exit(EXIT_FAILURE);
}
if(!(fIn >> dimM >> dimN))
{
std::cout << "Error in reading matrix entries!" << std::endl;
exit(EXIT_FAILURE);
}
fIn.close();
fIn.open(argv[2]);
if (!fIn) {
std::cout << "Error opening file: " << argv[2] << std::endl;
exit(EXIT_FAILURE);
}
if(!(fIn >> dimN >> dimO))
{
std::cout << "Error in reading matrix entries!" << std::endl;
exit(EXIT_FAILURE);
}
fIn.close();
//calculate minimal matrix size
//all matrices are padded with 0s to this size
//should be power of 2 for efficient block division
//dirty hack...
LD = 64;
if (LD<dimM) LD = dimM;
if (LD<dimN) LD = dimN;
if (LD<dimO) LD = dimO;
LD--;
LD |= LD >> 1;
LD |= LD >> 2;
LD |= LD >> 4;
LD |= LD >> 8;
LD |= LD >> 16;
LD++;
//add useless padding
LD += PADDING;
double* a = (double*) aligned_alloc(ALIGNMENT, sizeof(double) * LD * LD);
double* b = (double*) aligned_alloc(ALIGNMENT, sizeof(double) * LD * LD);
double* c = (double*) aligned_alloc(ALIGNMENT, sizeof(double) * LD * LD);
Matrix A = loadMatrix(argv[1], &a[0]);
Matrix B = loadMatrix(argv[2], &b[0]);
Matrix C(&c[0], nullptr, A.getDimM(), B.getDimN(), 0, 0);
///******************************************************
///********************** CALCULATION *******************
///******************************************************
double time = 0;
#ifdef USE_LIKWID
likwid_markerInit();
likwid_markerStartRegion("dummy");
#endif
siwir::Timer timer;
MMM(A, B, C);
time = timer.elapsed();
std::cout << dimM << "\t" << dimN << "\t" << dimO << "\t" << time << std::endl;
#ifdef USE_LIKWID
likwid_markerStopRegion("dummy");
likwid_markerClose();
#endif
///******************************************************
///********************** OUTPUT ************************
///******************************************************
saveMatrix(argv[3], C);
free(a);
free(b);
free(c);
};
int main(int argc, char** argv)
{
int iter = 100;
uint32_t i;
uint32_t j;
int globalNumberOfThreads = 0;
int optPrintDomains = 0;
int c;
ThreadUserData myData;
bstring testcase = bfromcstr("none");
uint32_t numberOfWorkgroups = 0;
int tmp = 0;
double time;
const TestCase* test = NULL;
Workgroup* currentWorkgroup = NULL;
Workgroup* groups = NULL;
cpuid_init();
numa_init();
affinity_init();
/* Handling of command line options */
if (argc == 1) { HELP_MSG; }
while ((c = getopt (argc, argv, "g:w:t:i:l:aphv")) != -1) {
switch (c)
{
case 'h':
HELP_MSG;
exit (EXIT_SUCCESS);
case 'v':
VERSION_MSG;
exit (EXIT_SUCCESS);
case 'a':
printf(TESTS"\n");
exit (EXIT_SUCCESS);
case 'w':
tmp--;
if (tmp == -1)
{
fprintf (stderr, "More workgroups configured than allocated!\n");
return EXIT_FAILURE;
}
if (!test)
{
fprintf (stderr, "You need to specify a test case first!\n");
return EXIT_FAILURE;
}
testcase = bfromcstr(optarg);
currentWorkgroup = groups+tmp; /*FIXME*/
bstr_to_workgroup(currentWorkgroup, testcase, test->type, test->streams);
bdestroy(testcase);
for (i=0; i< test->streams; i++)
{
if (currentWorkgroup->streams[i].offset%test->stride)
{
fprintf (stderr, "Stream %d: offset is not a multiple of stride!\n",i);
return EXIT_FAILURE;
}
allocator_allocateVector(&(currentWorkgroup->streams[i].ptr),
PAGE_ALIGNMENT,
currentWorkgroup->size,
currentWorkgroup->streams[i].offset,
test->type,
currentWorkgroup->streams[i].domain);
}
break;
case 'i':
iter = atoi(optarg);
break;
case 'l':
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (biseqcstr(testcase,"none"))
{
fprintf (stderr, "Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
else
{
printf("Name: %s\n",test->name);
printf("Number of streams: %d\n",test->streams);
printf("Loop stride: %d\n",test->stride);
printf("Flops: %d\n",test->flops);
printf("Bytes: %d\n",test->bytes);
switch (test->type)
{
case SINGLE:
printf("Data Type: Single precision float\n");
break;
case DOUBLE:
printf("Data Type: Double precision float\n");
break;
}
}
bdestroy(testcase);
exit (EXIT_SUCCESS);
break;
case 'p':
optPrintDomains = 1;
break;
case 'g':
numberOfWorkgroups = atoi(optarg);
allocator_init(numberOfWorkgroups * MAX_STREAMS);
tmp = numberOfWorkgroups;
groups = (Workgroup*) malloc(numberOfWorkgroups*sizeof(Workgroup));
break;
case 't':
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (biseqcstr(testcase,"none"))
{
fprintf (stderr, "Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
bdestroy(testcase);
break;
case '?':
if (isprint (optopt))
fprintf (stderr, "Unknown option `-%c'.\n", optopt);
else
fprintf (stderr,
"Unknown option character `\\x%x'.\n",
optopt);
return EXIT_FAILURE;
default:
HELP_MSG;
}
}
if (optPrintDomains)
{
affinity_printDomains();
exit (EXIT_SUCCESS);
}
timer_init();
/* :WARNING:05/04/2010 08:58:05 AM:jt: At the moment the thread
* module only allows equally sized thread groups*/
for (i=0; i<numberOfWorkgroups; i++)
{
globalNumberOfThreads += groups[i].numberOfThreads;
}
threads_init(globalNumberOfThreads);
threads_createGroups(numberOfWorkgroups);
/* we configure global barriers only */
barrier_init(1);
barrier_registerGroup(globalNumberOfThreads);
#ifdef PERFMON
printf("Using likwid\n");
likwid_markerInit();
#endif
/* initialize data structures for threads */
for (i=0; i<numberOfWorkgroups; i++)
{
myData.iter = iter;
myData.size = groups[i].size;
myData.test = test;
myData.numberOfThreads = groups[i].numberOfThreads;
myData.processors = (int*) malloc(myData.numberOfThreads * sizeof(int));
myData.streams = (void**) malloc(test->streams * sizeof(void*));
for (j=0; j<groups[i].numberOfThreads; j++)
{
myData.processors[j] = groups[i].processorIds[j];
}
for (j=0; j< test->streams; j++)
{
myData.streams[j] = groups[i].streams[j].ptr;
}
threads_registerDataGroup(i, &myData, copyThreadData);
free(myData.processors);
free(myData.streams);
}
printf(HLINE);
printf("LIKWID MICRO BENCHMARK\n");
printf("Test: %s\n",test->name);
printf(HLINE);
printf("Using %d work groups\n",numberOfWorkgroups);
printf("Using %d threads\n",globalNumberOfThreads);
printf(HLINE);
threads_create(runTest);
threads_destroy();
allocator_finalize();
time = (double) threads_data[0].cycles / (double) timer_getCpuClock();
printf("Cycles: %llu \n", LLU_CAST threads_data[0].cycles);
printf("Iterations: %llu \n", LLU_CAST iter);
printf("Size: %d \n", currentWorkgroup->size );
printf("Vectorlength: %d \n", threads_data[0].data.size);
printf("Time: %e sec\n", time);
printf("MFlops/s:\t%.2f\n",
1.0E-06 * ((double) numberOfWorkgroups * iter * currentWorkgroup->size * test->flops/ time));
printf("MByte/s:\t%.2f\n",
1.0E-06 * ( (double) numberOfWorkgroups * iter * currentWorkgroup->size * test->bytes/ time));
printf("Cycles per update:\t%f\n",
((double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
switch ( test->type )
{
case SINGLE:
printf("Cycles per cacheline:\t%f\n",
(16.0 * (double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
break;
case DOUBLE:
printf("Cycles per cacheline:\t%f\n",
(8.0 * (double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
break;
}
printf(HLINE);
#ifdef PERFMON
likwid_markerClose();
#endif
return EXIT_SUCCESS;
}
likwid_markerclose_(void)
{
likwid_markerClose();
}
int
main()
{
int quantum, checktick();
int BytesPerWord;
register int j, k;
double scalar, t, times[4][NTIMES];
/* --- SETUP --- determine precision and check timing --- */
printf(HLINE);
printf("STREAM version $Revision: 5.8 $\n");
printf(HLINE);
BytesPerWord = sizeof(double);
printf("This system uses %d bytes per DOUBLE PRECISION word.\n",
BytesPerWord);
printf(HLINE);
printf("Array size = %d, Offset = %d\n" , N, OFFSET);
printf("Total memory required = %.1f MB.\n",
(3.0 * BytesPerWord) * ( (double) N / 1048576.0));
printf("Each test is run %d times, but only\n", NTIMES);
printf("the *best* time for each is used.\n");
#ifdef PERFMON
printf("Using likwid\n");
likwid_markerInit();
#endif
#ifdef _OPENMP
printf(HLINE);
#pragma omp parallel
{
#pragma omp master
{
k = omp_get_num_threads();
printf ("Number of Threads requested = %i\n",k);
}
printf ("Thread %d running on processor %d ....\n",omp_get_thread_num(),threadGetProcessorId());
}
#endif
/* Get initial value for system clock. */
#pragma omp parallel for
for (j=0; j<N; j++) {
a[j] = 1.0;
b[j] = 2.0;
c[j] = 0.0;
}
printf(HLINE);
if ( (quantum = checktick()) >= 1)
printf("Your clock granularity/precision appears to be "
"%d microseconds.\n", quantum);
else {
printf("Your clock granularity appears to be "
"less than one microsecond.\n");
quantum = 1;
}
t = mysecond();
#pragma omp parallel for
for (j = 0; j < N; j++)
a[j] = 2.0E0 * a[j];
t = 1.0E6 * (mysecond() - t);
printf("Each test below will take on the order"
" of %d microseconds.\n", (int) t );
printf(" (= %d clock ticks)\n", (int) (t/quantum) );
printf("Increase the size of the arrays if this shows that\n");
printf("you are not getting at least 20 clock ticks per test.\n");
printf(HLINE);
printf("WARNING -- The above is only a rough guideline.\n");
printf("For best results, please be sure you know the\n");
printf("precision of your system timer.\n");
printf(HLINE);
/* --- MAIN LOOP --- repeat test cases NTIMES times --- */
scalar = 3.0;
for (k=0; k<NTIMES; k++)
{
times[0][k] = mysecond();
#pragma omp parallel
{
START_PERFMON("copy")
#pragma omp for
for (j=0; j<N; j++)
c[j] = a[j];
STOP_PERFMON("copy")
}
times[0][k] = mysecond() - times[0][k];
times[1][k] = mysecond();
#pragma omp parallel
{
START_PERFMON("scale")
#pragma omp for
for (j=0; j<N; j++)
b[j] = scalar*c[j];
STOP_PERFMON("scale")
}
times[1][k] = mysecond() - times[1][k];
times[2][k] = mysecond();
#pragma omp parallel
{
START_PERFMON("add")
#pragma omp for
for (j=0; j<N; j++)
c[j] = a[j]+b[j];
STOP_PERFMON("add")
}
times[2][k] = mysecond() - times[2][k];
times[3][k] = mysecond();
#pragma omp parallel
{
START_PERFMON("triad")
#pragma omp for
for (j=0; j<N; j++)
a[j] = b[j]+scalar*c[j];
STOP_PERFMON("triad")
}
times[3][k] = mysecond() - times[3][k];
}
/* --- SUMMARY --- */
for (k=1; k<NTIMES; k++) /* note -- skip first iteration */
{
for (j=0; j<4; j++)
{
avgtime[j] = avgtime[j] + times[j][k];
mintime[j] = MIN(mintime[j], times[j][k]);
maxtime[j] = MAX(maxtime[j], times[j][k]);
}
}
printf("Function Rate (MB/s) Avg time Min time Max time\n");
for (j=0; j<4; j++) {
avgtime[j] = avgtime[j]/(double)(NTIMES-1);
printf("%s%11.4f %11.4f %11.4f %11.4f\n", label[j],
1.0E-06 * bytes[j]/mintime[j],
avgtime[j],
mintime[j],
maxtime[j]);
}
printf(HLINE);
/* --- Check Results --- */
checkSTREAMresults();
printf(HLINE);
#ifdef PERFMON
likwid_markerClose();
#endif
return 0;
}
