int main(void)
{
double ** matrixA, ** matrixB, ** matrixC, ticks;
int size = 1600;
matrixA = allocMatrix(size,size); randomMatrix(matrixA, size, size);
matrixB = allocMatrix(size,size); randomMatrix(matrixB, size, size);
matrixC = allocMatrix(size,size);
printf("Two random matricies generated, now multiplying...\n\n");
ticks = clock();
multiplyMatrix(matrixA, matrixB, matrixC, size, size, size);
printf("Multiplication of two square matricies of size %d took %g seconds\n",size,(clock() - ticks)/CLOCKS_PER_SEC);
freeMatrix(matrixA);
freeMatrix(matrixB);
freeMatrix(matrixC);
return 0;
}
int main(int argc, char *argv[]) {
unsigned n;
int evt;
double *A;
int i, j;
double checksum = 0;
double startTime, endTime;
long long counter;
if (argc < 2) {
return -1;
}
n = atoi(argv[1]);
evt = (argc > 2) ? atoi(argv[2]) : -1;
A = randomMatrix(n);
assert(A != NULL);
if (evt == -1) {
startTime = dclock();
} else {
papi_init(evt);
papi_start();
}
if (chol(A, n)) {
fprintf(stderr, "Error: matrix is either not symmetric or not positive definite.\n");
} else {
for (i = 0; i < n; i++) {
for (j = i; j < n; j++) {
checksum += A[IDX(i, j, n)];
}
}
printf("Checksum: %f \n", checksum);
}
if (evt == -1) {
endTime = dclock();
fprintf(stderr, "%f\n", endTime - startTime);
} else {
counter = papi_stop();
fprintf(stderr, "%lld\n", counter);
}
free(A);
return 0;
}
int main(int argc, char *argv[])
{
if(argc<3)
{
printf("Usage : %s <taille de la matrice> <filename>\n",argv[0]);
exit(1);
}
srand(time(NULL));
int taille=atoi(argv[1]);
Matrix A=newMatrix(taille,taille);
A=randomMatrix(A);
valeurPropre_gnuplot(A,argv[2]);
deleteMatrix(A);
return 0;
}
static void
check_matrix(generator_t &gen, distribution_t &dist)
{
// check identity
{
mtx44 m;
mtx44Identity(&m);
assert(m == glm::mat4());
}
for(size_t x = 0; x < 10000; ++x)
{
// check multiply
{
mtx44 m1, m2;
randomMatrix(m1, gen, dist);
randomMatrix(m2, gen, dist);
glm::mat4 g1 = loadMatrix(m1);
glm::mat4 g2 = loadMatrix(m2);
mtx44 result;
mtx44Multiply(&result, &m1, &m2);
assert(result == g1*g2);
}
// check translate
{
mtx44 m;
randomMatrix(m, gen, dist);
glm::mat4 g = loadMatrix(m);
glm::vec3 v = randomVector(gen, dist);
mtx44Translate(&m, v.x, v.y, v.z);
assert(m == glm::translate(g, v));
}
// check scale
{
mtx44 m;
randomMatrix(m, gen, dist);
glm::mat4 g = loadMatrix(m);
glm::vec3 v = randomVector(gen, dist);
mtx44Scale(&m, v.x, v.y, v.z);
assert(m == glm::scale(g, v));
}
// check rotate
{
mtx44 m;
randomMatrix(m, gen, dist);
float r = randomAngle(gen, dist);
glm::mat4 g = loadMatrix(m);
glm::vec3 v = randomVector(gen, dist);
mtx44Rotate(&m, (vec3f){ v.x, v.y, v.z }, r);
assert(m == glm::rotate(g, r, v));
}
// check rotate X
{
mtx44 m;
randomMatrix(m, gen, dist);
float r = randomAngle(gen, dist);
glm::mat4 g = loadMatrix(m);
mtx44RotateX(&m, r);
assert(m == glm::rotate(g, r, x_axis));
}
// check rotate Y
{
mtx44 m;
randomMatrix(m, gen, dist);
float r = randomAngle(gen, dist);
glm::mat4 g = loadMatrix(m);
mtx44RotateY(&m, r);
assert(m == glm::rotate(g, r, y_axis));
}
// check rotate Z
{
mtx44 m;
randomMatrix(m, gen, dist);
float r = randomAngle(gen, dist);
glm::mat4 g = loadMatrix(m);
mtx44RotateZ(&m, r);
assert(m == glm::rotate(g, r, z_axis));
}
}
}
void RandomReductionPlugin::randomReduction(std::ofstream &outStream)
{
std::cerr << "[INFO] Random reduction (iteration limit: " << m_iterationLimit << ") ... " << std::endl;
std::set<IndexType> Trand; // random test set
std::set<IndexType> TrandIter; // random test set
std::vector<IndexType> randomizedTests; // stores the shuffled test case id-s
std::vector<IndexType> randomizedIterTests;
// print initial test suite
outStream << "Number of procedures: " << m_nrOfCodeElements << std::endl;
// initialize the remaining set
for(unsigned int i = 0; i < m_nrOfTestCases; i++) {
randomizedTests.push_back(i);
}
unsigned int fullSize = randomizedTests.size();
shuffle(randomizedTests); // make randomized order
randomizedIterTests = randomizedTests;
// print initial test suite
outStream << "Total " << fullSize << " test cases in full test suite." << std::endl;
Trand.clear();
TrandIter.clear();
CReductionData randomIterMatrix(m_data->getCoverage(), m_programName + "-RAND-ITER", m_dirPath);
CReductionData randomMatrix(m_data->getCoverage(), m_programName + "-RAND", m_dirPath);
unsigned int iter = 0;
unsigned int numClasses = 1;
while (iter < m_iterationLimit) {
std::cerr << "[RANDOM] Iteration " << iter + 1 << " (number of classes: " << numClasses << ") " << std::endl;
// fill the random matrix
if (addRandom(randomizedIterTests, TrandIter, numClasses) != (int)numClasses) { // oops
std::cerr << "[RANDOM] random test addition failed (adding 1 test)" << std::endl;
numClasses *= 2;
iter++;
continue;
}
outStream << "Selected tests in the " << iter + 1 << "th iteration: " << std::endl;
outStream << numClasses << " tests added" << std::endl;
// create the random matrix
randomIterMatrix.add(TrandIter);
randomIterMatrix.save(iter + 1);
numClasses *= 2;
iter++;
if (TrandIter.size() >= m_nrOfTestCases) {
std::cerr << "[RANDOM] Full test suite size reached, exiting before iteration " << iter + 1 << " ..." << std::endl;
break;
}
}
for (int i = 0; i < (int)m_reductionSizes.size(); ++i) {
int reductionSize = m_reductionSizes[i];
std::cerr << "[RANDOM] Fixed size " << i << " (number of classes: " << reductionSize << ") " << std::endl;
// fill the random matrix
if (addRandom(randomizedTests, Trand, reductionSize) != (int)reductionSize) {// oops
std::cerr << "[RANDOM] random test addition failed (adding 1 test)" << std::endl;
continue;
}
// create the random matrix
randomMatrix.add(Trand);
randomMatrix.save(i + 1);
if (Trand.size() >= m_nrOfTestCases) {
std::cerr << "[RANDOM] Full test suite size reached, exiting before iteration " << iter + 1 << " ..." << std::endl;
break;
}
}
unsigned int reducedSize = TrandIter.size();
outStream << "Random reduction ended" << std::endl;
// print final test suite:
outStream << "Fixed iteration reduction" << std::endl;
outStream << "Total " << reducedSize << " test cases in reduced test suite." << std::endl;
outStream << "Reduction rate: " << 100.0 * (fullSize - reducedSize) / fullSize << "\%" << std::endl;
reducedSize = Trand.size();
outStream << "Fixed size reduction" << std::endl;
outStream << "Total " << reducedSize << " test cases in reduced test suite." << std::endl;
outStream << "Reduction rate: " << 100.0 * (fullSize - reducedSize) / fullSize << "\%" << std::endl;
std::cerr << "done." << std::endl;
}
int main(int argc, char * argv[]) {
int rank_grid, rank_row, rank_col;
int coordinates[2];
int node_total_size;
int node_dim_size;
int elem_dim_size;
int subelem_dim_size;
int * scatter_sendcount;
int * scatter_displacement;
int gridinit_num_dims = 2;
int gridinit_dims[2] = {0,0};
int gridinit_periods[2] = {0,0};
int gridinit_reorder = 1;
MPI_Comm mpi_comm_grid, mpi_comm_row, mpi_comm_col;
MPI_Datatype mpi_type_submatrix, mpi_type_submatrix_vector;
MPI_Request fox_send_request, fox_recv_request;
int fox_sendto, fox_recfrom, fox_sendtag, fox_rectag;
int fox_broadcaster;
double *mat_a, *mat_b, *mat_c;
double *A_mine, *B_old, *B_new, *C_mine, *A_bcast;
double *mat_verify;
int i, j, k;
int verify = 0;
int verbose = 0;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &node_total_size);
double starttime, endtime;
starttime = MPI_Wtime();
// Set up cartesian coordinate grid
MPI_Dims_create(node_total_size,
gridinit_num_dims,
gridinit_dims);
MPI_Cart_create(MPI_COMM_WORLD,
gridinit_num_dims,
gridinit_dims,
gridinit_periods,
gridinit_reorder,
&mpi_comm_grid);
// ** Get the grid coordinates of this process.
MPI_Comm_rank(mpi_comm_grid, &rank_grid);
MPI_Cart_coords(mpi_comm_grid, rank_grid, gridinit_num_dims, coordinates);
// ** Set up column communicators.
MPI_Comm_split(mpi_comm_grid, coordinates[1], coordinates[0], &mpi_comm_col);
MPI_Comm_rank(mpi_comm_col, &rank_col);
// ** Set up row communicators
MPI_Comm_split(mpi_comm_grid, coordinates[0], coordinates[1], &mpi_comm_row);
MPI_Comm_rank(mpi_comm_row, &rank_row);
// Get the number of processors per dimension in grid.
MPI_Comm_size(mpi_comm_row, &node_dim_size);
// ********************************************
// ** CHECK SANITY OF AND SET UP ENVIRONMENT **
// ********************************************
// Check that number of parameters is sane.
if(argc < 2) {
if(rank_grid == 0)
printf("Usage: foxmatrix N\n N = randomize NxN matrices.\n");
MPI_Finalize();
return -1;
}
// Get the number of elements per dimension in matrices from arguments.
elem_dim_size = atoi(argv[1]);
// Check that number of processors is sane.
if(sqrt(node_total_size) != (double) ((int) sqrt(node_total_size))) {
if(rank_grid == 0)
printf("Not a square number of processors.\n");
MPI_Finalize();
return -1;
}
// Check that it is possible to split matrix over the processors.
if(elem_dim_size % node_dim_size != 0) {
if(rank_grid == 0)
printf("Cannot split elements evenly over processors.\n");
MPI_Finalize();
return -1;
}
// Calculate the size (in one dimension) of the submatrices.
subelem_dim_size = elem_dim_size / node_dim_size;
// Check if the user has given the verify/verbose commands.
if(argc == 3 && strcmp(argv[2], "verify") == 0)
verify = 1;
else if(argc == 3 && strcmp(argv[2], "verbose") == 0)
verbose = 1;
else if(argc == 4 &&
strcmp(argv[2], "verbose") == 0 &&
strcmp(argv[3], "verify") == 0) {
verbose = 1;
verify = 1;
} else if(argc == 4 &&
strcmp(argv[2], "verify") == 0 &&
strcmp(argv[3], "verbose") == 0) {
verbose = 1;
verify = 1;
}
// Create datatype used for transmitting submatrices.
// Idea of using vector+struct taken from http://www.mcs.anl.gov/research/projects/mpi/tutorial/mpiexmpl/src4/scatter/C/solution.html.
MPI_Type_vector(subelem_dim_size,
subelem_dim_size,
elem_dim_size,
MPI_DOUBLE,
&mpi_type_submatrix_vector);
int sm_blocklength[2] = {1, 1};
MPI_Aint sm_displacement[2] = {0, subelem_dim_size * sizeof(double)};
MPI_Datatype sm_types[2] = {mpi_type_submatrix_vector, MPI_UB};
MPI_Type_struct(2,
sm_blocklength,
sm_displacement,
sm_types,
&mpi_type_submatrix);
MPI_Type_commit(&mpi_type_submatrix);
// ** CREATE MATRICES AND SET UP SCATTERV/GATHERV VARIABLES **
if(rank_grid == 0) {
// Create matrices on rank 0.
mat_a = (double *) malloc(elem_dim_size * elem_dim_size * sizeof(double));
mat_b = (double *) malloc(elem_dim_size * elem_dim_size * sizeof(double));
mat_c = (double *) malloc(elem_dim_size * elem_dim_size * sizeof(double));
// Randomize matrix contents.
randomMatrixInit();
randomMatrix(mat_a, elem_dim_size);
randomMatrix(mat_b, elem_dim_size);
// Allocate memory for storing scattering information.
scatter_sendcount = (int *) malloc(node_total_size * sizeof(int));
scatter_displacement = (int *) malloc(node_total_size * sizeof(int));
// Set up scatter/gather arguments.
int sit;
for(sit = 0; sit < node_total_size; sit++) {
scatter_sendcount[sit] = 1;
if(sit == 0)
scatter_displacement[sit] = 0;
else {
scatter_displacement[sit] = scatter_displacement[sit - 1] + 1;
if(sit % node_dim_size == 0)
// At end of line, go to start of next submatrix.
scatter_displacement[sit] += node_dim_size * (subelem_dim_size - 1);
}
}
}
A_mine = (double *) malloc(subelem_dim_size*subelem_dim_size*sizeof(double));
A_bcast = (double *) malloc(subelem_dim_size*subelem_dim_size*sizeof(double));
B_old = (double *) malloc(subelem_dim_size*subelem_dim_size*sizeof(double));
B_new = (double *) malloc(subelem_dim_size*subelem_dim_size*sizeof(double));
C_mine = (double *) malloc(subelem_dim_size*subelem_dim_size*sizeof(double));
zeroMatrix(C_mine, subelem_dim_size);
// ** DISTRIBUTE THE SUBMATRICES TO THE GRID NODES **
MPI_Scatterv(mat_a,
scatter_sendcount,
scatter_displacement,
mpi_type_submatrix,
A_mine,
subelem_dim_size * subelem_dim_size,
MPI_DOUBLE,
0,
mpi_comm_grid);
MPI_Scatterv(mat_b,
scatter_sendcount,
scatter_displacement,
mpi_type_submatrix,
B_new,
subelem_dim_size * subelem_dim_size,
MPI_DOUBLE,
0,
mpi_comm_grid);
// ** PERFORM FOX'S ALGORITHM FOR MATRIX MULTIPLICATION **
for(k = 0; k < node_dim_size; k++) {
// **** BROADCAST A **** //
// Decide who broadcasts this iteration.
fox_broadcaster = (k + rank_col) % node_dim_size;
// Copy matrix to the broadcast variable of the node that shall broadcast.
if(rank_row == fox_broadcaster)
copyMatrix(A_bcast, A_mine, subelem_dim_size);
// Perform the broadcasting of the A matrix.
MPI_Bcast(A_bcast,
subelem_dim_size * subelem_dim_size,
MPI_DOUBLE,
fox_broadcaster,
mpi_comm_row);
// **** CREATE COPY OF B **** //
// Wait for everyone to get their new B. If k = 0 everyone has it scattered.
if(k != 0)
MPI_Wait(&fox_recv_request, MPI_STATUS_IGNORE);
// Make a copy of B so we can overwrite the old one.
copyMatrix(B_old, B_new, subelem_dim_size);
// **** SHIFT B **** //
// Find which node to send to, and which to recieve from (B matrix).
fox_recfrom = ((rank_col + 1) % node_dim_size);
fox_sendto = ((rank_col - 1) % node_dim_size);
if(fox_sendto < 0)
fox_sendto = node_dim_size - 1;
fox_sendtag = 1000 + fox_sendto;
fox_rectag = 1000 + rank_col;
// Send the B matrix.
MPI_Isend(B_old,
subelem_dim_size * subelem_dim_size,
MPI_DOUBLE,
fox_sendto,
fox_sendtag,
mpi_comm_col,
&fox_send_request);
// Receive the B matrix.
MPI_Irecv(B_new,
subelem_dim_size * subelem_dim_size,
MPI_DOUBLE,
fox_recfrom,
fox_rectag,
mpi_comm_col,
&fox_recv_request);
// Perform matrix multiplication on the local submatrix.
naiveMatrixMult(A_bcast, B_old, C_mine, subelem_dim_size);
}
// ** GATHER DATA **
MPI_Barrier(MPI_COMM_WORLD);
// Collect C from submatrices.
MPI_Gatherv(C_mine,
subelem_dim_size * subelem_dim_size,
MPI_DOUBLE,
mat_c,
scatter_sendcount,
scatter_displacement,
mpi_type_submatrix,
0,
mpi_comm_grid);
// ** PRESENT DATA **
if(verbose && !rank_grid) {
printf("** Will print matrix C from 0:\n");
printMatrix(mat_c, elem_dim_size);
}
// ** VERIFICATION OF CORRECTNESS **
if(verify && !rank_grid) {
// Allocate memory for verification matrix.
mat_verify = (double *)
malloc(elem_dim_size * elem_dim_size * sizeof(double));
// Initialize verification matrix to zeroes.
zeroMatrix(mat_verify, elem_dim_size);
// Do the naive multiplication.
naiveMatrixMult(mat_a, mat_b, mat_verify, elem_dim_size);
// Print the correct matrix.
if(verbose) {
printf("** Print correct matrix from 0:\n");
printMatrix(mat_verify, elem_dim_size);
}
// Check equality between matrices.
if(matrixEqual(mat_c, mat_verify, elem_dim_size))
printf("\n Ok!\n\n");
else
printf("\n FAIL!\n\n");
// Free the memory used by the verification matrix.
free(mat_verify);
}
// ** FINALIZE MPI **
MPI_Barrier(MPI_COMM_WORLD);
if(rank_grid==0)
{
endtime = MPI_Wtime();
printf("%f\n", endtime - starttime);
}
MPI_Finalize();
// ** CLEANUP **
if(A_mine)
free(A_mine);
if(A_bcast)
free(A_bcast);
if(B_old)
free(B_old);
if(B_new)
free(B_new);
if(C_mine)
free(C_mine);
// Local rank 0 cleanup.
if(rank_grid == 0) {
free(mat_a);
free(mat_b);
free(mat_c);
}
return 0;
}
void SpeedTest(FunctionCall fc)
{
int i, min, max, step, sec=0;
int sizeTimeArray, fct;
char fnc[MAXSIZE_FCT];
char foutput[256];
struct sigaction action;
action.sa_handler = handler;
sigemptyset(&action.sa_mask);
struct timeval start, end, result;
double maxtime = 0;
double time;
double usec;
double* timeArray;
Matrix a=NULL, b=NULL, tmp=NULL;
E s;
int n;
TokenizeSpeedTest(fc, fnc, &min, &max, &step, &sec);
if (min>0 && max>0 && step>0)
{
sizeTimeArray = ((max-min)/step)+1;
timeArray = (double*)malloc(sizeTimeArray*sizeof(double));
for (i=0; i<sizeTimeArray; i++) timeArray[i] = 0;
}
usec = (double)sec * 1000000;
sprintf(foutput, "speedtest_%s_%d_%d_%d_%d", fnc, min, max, step, sec);
fct = GetFunction(fnc);
switch (fct)
{
case NMX :
{
printf("\t You must specify another function\n");
return;
}
case ADD :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
b = randomMatrix(i, i, MIN_E, MAX_E);
gettimeofday(&start, NULL);
tmp = addMatricis(a, b);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(b);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case SUB :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
b = randomMatrix(i, i, MIN_E, MAX_E);
gettimeofday(&start, NULL);
tmp = substractMatricis(a, b);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(b);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case MUL :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
b = randomMatrix(i, i, MIN_E, MAX_E);
gettimeofday(&start, NULL);
tmp = mulMatricis(a, b);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(b);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case MSC :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
s = mRand(MIN_SCA, MAX_SCA);
gettimeofday(&start, NULL);
tmp = mult_scal(a, s);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case EXP :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
n = (int)mRand(MIN_EXP, MAX_EXP);
gettimeofday(&start, NULL);
tmp = expo(a, n);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case TRA :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
gettimeofday(&start, NULL);
tmp = transposeMatrix(a);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case DET :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
gettimeofday(&start, NULL);
s = determinant(a);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case DLU :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
b = newMatrix(i, i);
tmp = identityMatrix(i);
gettimeofday(&start, NULL);
decomposition(a, b, tmp);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(b);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case SOL :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
b = randomMatrix(i, 1, MIN_E, MAX_E);
gettimeofday(&start, NULL);
tmp = gauss(a, b);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(b);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case INV :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
gettimeofday(&start, NULL);
tmp = invert(a);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
deleteMatrix(tmp);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case RNK :
{
sigaction(SIGINT, &action, NULL);
for (i=min; i<=max; i+=step)
{
a = randomMatrix(i, i, MIN_E, MAX_E);
gettimeofday(&start, NULL);
n = rank(a);
gettimeofday(&end, NULL);
timersub(&end, &start, &result);
time = (result.tv_sec*1000000)+result.tv_usec;
if (maxtime < time) maxtime = time;
timeArray[(i-min)/step] = time;
printf("\t* %s Size:%5d ;\tTime:%8.0f µs ;\n", fnc, i, time);
deleteMatrix(a);
if (usec >= 1000000) {
if (time >= usec) {
i++;
break;
}
}
if (CTRLC) {
i++;
break;
}
}
CreateGNUPLOTFile(min, i-step, step, timeArray, maxtime, foutput);
break;
}
case VAR : // default
case NOF : // default
default :
{
printf("\t%s : Function Not Implemented\n", fnc);
fni++;
break;
}
}
if (fct!=NOF && fct !=VAR) fni = 0;
free(timeArray);
CTRLC = 0;
sigemptyset(&action.sa_mask);
}
int main(int argc, char *argv[]) {
int numtasks, rank, dest, source, rc, count, tag=1;
MPI_Status Stat;
// Seed random number generator.
randomMatrixInit();
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
double *A;
double *B;
if(argc==1)
{
printf("Specify matrix size\n");
return 1;
}
int size=atoi(argv[1]);
A = (double*)malloc(sizeof(double)*size*size);
B = (double*)malloc(sizeof(double)*size*size);
randomMatrix(A,size);
randomMatrix(B,size);
// int Atest[]={1,2,3,4,5,6,7,8,9};
// int Btest[]={3,2,1,6,5,4,4,3,2};
// A=Atest;
// B=Btest;
if(numtasks>size*size)
numtasks=size*size;
int minCells=size*size/(numtasks);
int extra=size*size-minCells*(numtasks);
int pad=0;
if (rank == 0) {
dest = 1;
source = 1;
double *C;
C = (double*)malloc(sizeof(double)*size*size);
double *ret = (double*)malloc(sizeof(double)*size*size/(numtasks)+1);
for(int cell=0; cell<minCells; cell++)
{
C[cell]=0;
for(int j=0; j<size; j++)
C[cell]+=A[(cell/size)*size+j]*B[j*size+cell%size];
// printf("Calculating job %d. C[%d]=%d\n", rank, cell, C[cell]);
}
for(int i=1; i<numtasks; i++)
{
int noCells=minCells;
if(i>=numtasks-extra)
{
pad=i-numtasks+extra;
noCells++;
}
rc = MPI_Recv(ret, noCells, MPI_DOUBLE, i, tag, MPI_COMM_WORLD, &Stat);
for(int j=0; j<noCells; j++)
{
int cell = minCells*i+pad+j;
// if(cell>size*size-1)
// printf(" KUK!\n");
C[cell]=ret[j];
// printf("C[%d]=ret[%d]=%d\n",cell, j, ret[j]);
}
}
// printMatrix(A, size);
// printf("*\n");
// printMatrix(B, size);
// printf("=\n");
// printMatrix(C, size);
// printf("\nControl:\n");
double *D;
if(argc==3 && strcmp(argv[2],"verify") == 0)
{
D = (double*)malloc(sizeof(double)*size*size);
naiveMatrixMult(A,B,D,size);
// printMatrix(D, size);
if(matrixEqual(C,D,size))
printf("\n Ok!\n\n");
else
printf("\n FAIL!\n\n");
free(D);
}
free(A);
free(B);
free(C);
}
else if (rank+1 <= size*size) {
int noCells=minCells;
if(rank>=numtasks-extra)
{
pad=rank-numtasks+extra;
noCells++;
}
// printf("noCells: %d extra: %d rank: %d\n", noCells, extra, rank);
double *ret = (double*)calloc(noCells,sizeof(double));
for(int i=0; i<noCells; i++)
{
ret[i] = 0;
int cell=minCells*(rank)+pad+i;
for(int j=0; j<size; j++)
ret[i]+=A[(cell/size)*size+j]*B[j*size+cell%size];
// printf("Calculating job %d. C[%d]=%d\n", rank, cell, ret[i]);
}
rc = MPI_Send(ret, noCells, MPI_DOUBLE, 0, tag, MPI_COMM_WORLD);
}
else {
}
MPI_Finalize();
return 0;
}