int main()
{
double *A, *B, *C;
int i,j,r,max_threads,size;
double alpha, beta;
double s_initial, s_elapsed;
printf("Intializing data for matrix multiplication C=A*B for matrix\n\n"
" A(%i*%i) and matrix B(%i*%i)\n",M,P,P,N);
alpha = 1.0;
beta = 0.0;
printf("Allocating memory for matrices aligned on 64-byte boundary for better performance \n\n");
A = ( double *)mkl_malloc(M*P*sizeof( double ),64);
B = ( double *)mkl_malloc(N*P*sizeof( double ),64);
C = ( double *)mkl_malloc(M*N*sizeof( double ),64);
if (A == NULL || B == NULL || C == NULL)
{
printf("Error: can`t allocate memory for matrices.\n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
return 1;
}
printf("Intializing matrix data\n\n");
size = M*P;
for (i = 0; i < size; ++i)
{
A[i] = ( double )(i+1);
}
size = N*P;
for (i = 0; i < size; ++i)
{
B[i] = ( double )(i-1);
}
printf("Finding max number of threads can use for parallel runs \n\n");
max_threads = mkl_get_max_threads();
printf("Running from 1 to %i threads \n\n",max_threads);
for (i = 1; i <= max_threads; ++i)
{
size = M*N;
for (j = 0; j < size; ++j)
{
C[j] = 0.0;
}
printf("Requesting to use %i threads \n\n",i);
mkl_set_num_threads(i);
printf("Measuring performance of matrix product using dgemm function\n"
" via CBLAS interface on %i threads \n\n",i);
s_initial = dsecnd();
for (r = 0; r < LOOP_COUNT; ++r)
{
cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, M, N, P, alpha, A, P, B, N, beta, C, N);
// multiply matrices with cblas_dgemm;
}
s_elapsed = (dsecnd() - s_initial) / LOOP_COUNT;
printf("Matrix multiplication using dgemm completed \n"
" at %.5f milliseconds using %d threads \n\n",
(s_elapsed * 1000),i);
printf("Output the result: \n");
size = M*N;
for (i = 0; i < size; ++i)
{
printf("%i\t",(int)C[i]);
if (i % N == N - 1)
printf("\n");
}
}
printf("Dellocating memory\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
return 0;
}
void solve1()
{
struct st_mesh *q=mshio_create_mesh(fin_base);
//mshio_print_mesh(q,PRINT_INFO_VERBOSE);
struct st_solver_v1 *s=sv1_create_solver(q,ipar,dpar);
sv1_print_solverinfo(s);
//double _Complex *b0=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *x0=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *b1=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *x1=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
//assert(b0);
assert(x0);
assert(b1);
assert(x1);
int nitr;
double eps;
//sv1_gen_b0(s,0.0,b0);
//sv1_solve(s,b0,x0,200,12,1.0E-13,&nitr,&eps);
//sv1_save_solution(s,x0,output_dir);
sv1_gen_b1x0(s,0.0,b1,x0);
sv1_solve(s,b1,x1,200,30,1.0E-13,&nitr,&eps);
sv1_save_solution(s,x1,output_dir);
//mkl_free(b0);
mkl_free(x0);
mkl_free(b1);
mkl_free(x1);
sv1_destroy_solver(s);
mshio_destroy_mesh(q);
}
void GeneticAlgorithm::resetParameters(int nPopulation, double scaleFactor, double crossingProbability){
_scaleFactor = scaleFactor;
_crossingProbability = crossingProbability;
_nPopulation = nPopulation;
delete [] _populationParametersOld;
delete [] _populationParametersNew;
_populationParametersOld = (Parameters::fitParameters *) mkl_malloc(sizeof(Parameters::fitParameters)*nPopulation,16);
_populationParametersNew = (Parameters::fitParameters *) mkl_malloc(sizeof(Parameters::fitParameters)*nPopulation,16);
for(int i = 0; i < _nPopulation; i++){
_populationParametersOld[i].c11 = (randomDouble(0.0,1.0))*pow(10,9);
_populationParametersOld[i].c22 = _populationParametersOld[i].c11;
_populationParametersOld[i].c33 = _populationParametersOld[i].c11;
_populationParametersOld[i].c44 = (randomDouble(0.0,1.0))*pow(10,9);
_populationParametersOld[i].c55 = _populationParametersOld[i].c44;
_populationParametersOld[i].c66 = _populationParametersOld[i].c44;
_populationParametersOld[i].c12 = (randomDouble(0.0,1.0))*pow(10,9);
_populationParametersOld[i].c13 = _populationParametersOld[i].c12;
_populationParametersOld[i].c23 = _populationParametersOld[i].c12;
_populationParametersOld[i].chiSq = 1;
_populationParametersOld[i].chiSq = calculateResidual(&_populationParametersOld[i],0);
}
//delete _integerDistribution;
delete [] ints1;
delete [] ints2;
delete [] ints3;
ints1 = new int[_nPopulation];
ints2 = new int[_nPopulation];
ints3 = new int[_nPopulation];
}
void mexFunction(int nlhs, mxArray *plhs[],
int nrhs, const mxArray *prhs[]){
/*Declarar las variables locales*/
mexPrintf("hios\n"); //Tarea1 termina aqui
double *A, *B, determinante;
int *pivot, info, Nfilas, Ncolumnas;
/*Insertar el código */
if (nrhs != 1){ // nº args diferente de 1
mexErrMsgTxt("Error. myla, Debe tener un arg de entrada");
}
if (!mxIsNumeric(prhs[0])){
mexErrMsgTxt("Error. El argumento de entrada debe ser una matriz");
}
Nfilas = mxGetM(prhs[0]);
Ncolumnas = mxGetN(prhs[0]);
if (Nfilas != Ncolumnas){
mexErrMsgTxt("Error. La matriz debe ser cuadrada");
}
if (Nfilas == 0){
mexErrMsgTxt("Error. La matriz debe no ser vacía");
}
if (nlhs > 2){
mexErrMsgTxt("Error. Debe haber uno o dos args de salida");
}
// copia de las variables
A = mxGetPr(prhs[0]);
B = (double *)mkl_malloc(Nfilas*Ncolumnas*sizeof(double), 64);
memcpy(B, A, Nfilas*Ncolumnas*sizeof(double));
pivot = (int *)mkl_malloc(Nfilas*sizeof(int), 32);
//procesos computacionales
info = LAPACKE_dgetrf(LAPACK_COL_MAJOR, Nfilas, Ncolumnas, B, Ncolumnas, pivot);
determinante = 1.0;
for (int i = 0; i < Nfilas; i++){
if (pivot[i] != (i+1)){
determinante *= -B[i*Ncolumnas + i];
}
else{
determinante *= B[i*Ncolumnas + i];
}
}
// crear los resultados de salida
plhs[0] = mxCreateDoubleScalar(determinante);
if (nlhs == 2){
if (fabs(determinante) < 1.0e-8){
mexWarnMsgTxt("Matriz singular o casi singular");
}
LAPACKE_dgetri(LAPACK_COL_MAJOR, Nfilas, B, Ncolumnas, pivot);
plhs[1] = mxCreateDoubleMatrix(Nfilas, Ncolumnas, mxREAL);
double *C = mxGetPr(plhs[1]);
memcpy(C, B, Nfilas*Ncolumnas*sizeof(double));
}
mkl_free(pivot);
mkl_free(B);
}
void save_2d_image_potential(SimulationData &sim_data, double *potential, const char * fits_file_name) {
double *save_data;
save_data = (double*)mkl_malloc(sim_data.get_num_x() * sim_data.get_num_y() * sizeof(double), 64);
fitsfile *fptr;
int status = 0;
long fpixel = 1, naxis = 2, nelements;
long naxes[2] = {sim_data.get_num_y(), sim_data.get_num_x()};
for (int i = 0; i < sim_data.get_num_x(); ++i) {
for (int j = 0; j < sim_data.get_num_y(); ++j) {
save_data[i * sim_data.get_num_y() + j] = 0;
for (int k = 0; k < sim_data.get_num_z(); ++k) {
save_data[i * sim_data.get_num_y() + j] += potential[i * sim_data.get_num_y() * sim_data.get_num_z() + j * sim_data.get_num_z() + k];
}
}
}
fits_create_file(&fptr, fits_file_name, &status);
fits_create_img(fptr, DOUBLE_IMG, naxis, naxes, &status);
nelements = naxes[0] * naxes[1];
fits_write_img(fptr, TDOUBLE, fpixel, nelements, save_data, &status);
fits_close_file(fptr, &status);
fits_report_error(stderr, status);
mkl_free(save_data);
}
int check_result(double* A, double* BT, double* C, int m, int n, int c, int transposed) {
int err_c = 0; //how many errors found
int i, j, k;
double* C_ref = (double*)mkl_malloc(m * n * sizeof(double), 16); //with zeroed
// for(i = 0; i < m; i ++) {
// for(j = 0; j < n; j++) {
// C_ref[i*n+j] = 0;
// for(k = 0; k < c; k++) {
// C_ref[i*n+j] += A[i*c+k] * BT[j*c+k];
// }
// }
// }
cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasTrans,
m, n, c,
1, A, c, BT, c, 0, C_ref, n);
if(transposed){
for(i = 0; i < m; i++) {
for(j = 0; j < n; j++) {
err_c += (fabs(C[j*m + i] - C_ref[i*n+j]) < 0.0001 ? 0 : 1);
}
}
} else {
//do compare
for(i = 0; i < m * n ; i++) {
err_c += (fabs(C[i] - C_ref[i]) < 0.0001 ? 0 : 1);
}
}
mkl_free(C_ref);
return err_c;
}
// If CUDA is available and in GPU mode, host memory will be allocated pinned,
// using cudaMallocHost. It avoids dynamic pinning for transfers (DMA).
// The improvement in performance seems negligible in the single GPU case,
// but might be more significant for parallel training. Most importantly,
// it improved stability for large models on many GPUs.
inline void CaffeMallocHost(void** ptr, size_t size, bool* use_cuda) {
#ifndef CPU_ONLY
if (Caffe::mode() == Caffe::GPU) {
CUDA_CHECK(cudaMallocHost(ptr, size));
*use_cuda = true;
return;
}
#endif
#ifdef USE_MLSL
if (mn::is_multinode()) {
*ptr = mn::alloc(size ? size : 1, 64);
} else {
#endif /* !USE_MLSL */
#ifdef USE_MKL
*ptr = mkl_malloc(size ? size : 1, 64);
#else
*ptr = malloc(size);
#endif
#ifdef USE_MLSL
}
#endif /* USE_MLSL */
*use_cuda = false;
CHECK(*ptr) << "host allocation of size " << size << " failed";
}
void GeneticAlgorithm2::resetParameters(int nPopulation, double scaleFactor, double crossingProbability){
_scaleFactor = scaleFactor;
_crossingProbability = crossingProbability;
_nPopulation = nPopulation;
delete [] _populationParametersOld;
delete [] _populationParametersNew;
// _populationParametersOld = new Parameters::fitParameters[nPopulation];
// _populationParametersNew = new Parameters::fitParameters[nPopulation];
_populationParametersOld = (Parameters::fitParameters *) mkl_malloc(sizeof(Parameters::fitParameters)*nPopulation,16);
_populationParametersNew = (Parameters::fitParameters *) mkl_malloc(sizeof(Parameters::fitParameters)*nPopulation,16);
for(int i = 0; i < _nPopulation; i++){
_populationParametersOld[i].A1 = randomDouble(0.0,5.0);
_populationParametersOld[i].A2 = randomDouble(0.0,5.0);
_populationParametersOld[i].F1 = randomDouble(440.0,500.0);
_populationParametersOld[i].F2 = randomDouble(500.0,560.0);
_populationParametersOld[i].dF1 = randomDouble(20.0,60.0);
_populationParametersOld[i].dF2 =0;
_populationParametersOld[i].phi1 = randomDouble(0.0,1.0);
_populationParametersOld[i].phi2 = randomDouble(0.0,1.0);
_populationParametersOld[i].Td1 = randomDouble(2.0,10.0);
_populationParametersOld[i].Td2 = randomDouble(2.0,10.0);
_populationParametersOld[i].ms1 = randomDouble(1,2.6);
_populationParametersOld[i].ms2 = randomDouble(1,2.6);
// _populationParametersOld[i].m1 = randomDouble(1.3,1.9);
// _populationParametersOld[i].m2 = randomDouble(1.3,1.9);
_populationParametersOld[i].m1 = 1.7;
_populationParametersOld[i].m2 = 1.7;
_populationParametersOld[i].dF12 = randomDouble(-15,5);
// _populationParametersOld[i].ms11 = randomDouble(0.0,.2);
// _populationParametersOld[i].ms22 = randomDouble(0.0,.2);
_populationParametersOld[i].T = 4.2;
}
for(int i = 0; i < _nPopulation; i++){
_populationParametersOld[i].chiSq = calculateResidual2(&_populationParametersOld[i],0);
}
//delete _integerDistribution;
delete [] ints1;
delete [] ints2;
delete [] ints3;
// _integerDistribution = new boost::random::uniform_int_distribution<>(0, _nPopulation-1);
ints1 = new int[_nPopulation];
ints2 = new int[_nPopulation];
ints3 = new int[_nPopulation];
}
PotentialData::PotentialData(SimulationData &sim_data) {
this->harmonic_trap = (double*)mkl_malloc(sim_data.num_points * sizeof(double), 64);
#pragma omp parallel for
for (int i = 0; i < sim_data.num_points; ++i) {
harmonic_trap[i] = 0.5 * pow(sim_data.x[i], 2.0);
}
}
int bench_stream_triad()
{
double *A, *B, *C;
double t;
int64_t m, n, k, i, j;
m = SIZE, k = SIZE, n = SIZE;
double scalar=3.14;
A = (double *)mkl_malloc( m*k*sizeof( double ), 64 );
B = (double *)mkl_malloc( k*n*sizeof( double ), 64 );
C = (double *)mkl_malloc( m*n*sizeof( double ), 64 );
#pragma omp parallel for
for (i = 0; i < (m*k); i++) {
A[i] = (double)(i+1);
}
#pragma omp parallel for
for (i = 0; i < (k*n); i++) {
B[i] = (double)(-i-1);
}
#pragma omp parallel for
for (i = 0; i < (m*n); i++) {
C[i] = 0.0;
}
if (A == NULL || B == NULL || C == NULL) {
printf( "\n ERROR: Can't allocate memory for matrices. Aborting... \n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
return 1;
}
t=stoptime();
for (i=0;i<NTIME;i++)
#pragma omp parallel for
for (j=0; j<(m*k); j++)
A[j] = B[j]+scalar*C[j];
t=stoptime()-t;
printf("GB/s : %f\n",(((((m*k)*3)*8)*NTIME)/t)*1E-9);
DPRINTF("\n Deallocating memory \n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
return 0;
}
vector_t::vector_t(size_t _len, double init)
:
data(static_cast<double*>(mkl_malloc(_len * sizeof(double), 64)), std::ptr_fun(mkl_free)),
len(_len), inc(1)
{
stack_assert(data.get() != nullptr);
stack_assert(len > 0);
std::fill(data.get(), data.get() + len, init);
}
vector_t::vector_t(size_t _len, std::mt19937 & gen)
:
data(static_cast<double*>(mkl_malloc(_len * sizeof(double), 64)), std::ptr_fun(mkl_free)),
len(_len), inc(1)
{
stack_assert(data.get() != nullptr);
stack_assert(len > 0);
std::normal_distribution<> d(0,0.1);
std::generate(data.get(), data.get() + len, std::bind(d,gen));
}
inline void* znn_malloc(size_t s)
{
#ifdef ZNN_XEON_PHI
void* r = mkl_malloc(s,64);
#else
void* r = malloc(s);
#endif
if ( !r ) throw std::bad_alloc();
return r;
}
int create_array(const MKL_INT length, T*& x)
{
if (x == nullptr)
{
x = (T*)mkl_malloc(length*sizeof(T),64);
if (x == nullptr)
{
return OUTOFMEMORY;
}
}
return 0;
}
WavefunctionData::WavefunctionData(SimulationData &sim_data) {
this->psi = (MKL_Complex16*)mkl_malloc(sim_data.num_points * sizeof(MKL_Complex16), 64);
this->psi_old = (MKL_Complex16*)mkl_malloc(sim_data.num_points * sizeof(MKL_Complex16), 64);
this->psi_new = (MKL_Complex16*)mkl_malloc(sim_data.num_points * sizeof(MKL_Complex16), 64);
this->conj_psi = (MKL_Complex16*)mkl_malloc(sim_data.num_points * sizeof(MKL_Complex16), 64);
this->psi_tf = (double*)mkl_malloc(sim_data.num_points * sizeof(double), 64);
this->psi_abs2 = (double*)mkl_malloc(sim_data.num_points * sizeof(double), 64);
this->wavefunction_norm = 1;
double expval;
#pragma omp parallel for private(expval)
for (int i = 0; i < sim_data.num_points; ++i) {
expval = exp(-0.05 * pow(sim_data.x[i], 2.0));
this->psi[i].real = expval;
this->psi[i].imag = 0;
this->psi_old[i].real = expval;
this->psi_old[i].imag = 0;
this->psi_new[i].real = 0;
this->psi_new[i].imag = 0;
this->conj_psi[i].real = 0;
this->conj_psi[i].imag = 0;
this->psi_abs2[i] = 0;
this->psi_tf[i] = 0;
}
calc_norm(sim_data, this->psi);
normalize_wf(sim_data, this->psi);
vzAbs(sim_data.num_points, this->psi, this->psi_abs2);
vdMul(sim_data.num_points, this->psi_abs2, this->psi_abs2, this->psi_abs2);
save_data(this->psi_abs2, sim_data, "init_state.bin");
}
void initial_matrix(double** A_addr, double** BT_addr, double** C_addr, int m, int n, int c) {
double* A = (double*)mkl_malloc(m * c * sizeof(double) , 16);
double* BT = (double*)mkl_malloc(n * c * sizeof(double), 16);
double* C = (double*)mkl_malloc(m * n * sizeof(double), 16);
*A_addr = A;
*BT_addr = BT;
*C_addr = C;
//use random number for input
int i,j,k;
srand((unsigned)time(NULL));
for(i = 0; i < m; i++) {
for(k = 0; k < c; k++) {
A[i*c+k] = ((double)rand()/(double)RAND_MAX);
}
}
for(j = 0; j < n; j++) {
for(k = 0; k < c; k++) {
BT[j*c+k] = ((double)rand()/(double)RAND_MAX);
}
}
}
int main(int argc, char *argv[]){
double inicio, fin = dsecnd();
double *A = (double *)mkl_malloc(N*N*sizeof(double), 64);
double *B = (double *)mkl_malloc(N*sizeof(double), 64);
int *pivot = (int *)mkl_malloc(N*sizeof(int), 32);
// distribucion normal de media 0 y varianza 1
std::default_random_engine generador;
std::normal_distribution<double> aleatorio(0.0, 1.0);
for (int i = 0; i < N*N; i++) A[i] = aleatorio(generador);
for (int i = 0; i < N; i++) B[i] = aleatorio(generador);
// matriz A marcadamente diagonal para evitar riesgo de singularidad
for (int i = 0; i < N; i++) A[i*N + i] += 10.0;
int result;
inicio = dsecnd();
for (int i = 0; i < NTEST; i++)
result = LAPACKE_dgesv(LAPACK_ROW_MAJOR, N, 1, A, N, pivot, B, 1);
fin = dsecnd();
double tiempo = (fin - inicio) / (double)NTEST;
printf("Tiempo: %lf msec\n", tiempo*1.0e3);
mkl_free(A);
mkl_free(B);
std::getchar(); return 0;
}
/*
*struct st_rmsm {
* int status; // internal status
* int size; // dimension, i.e., size of this matrix
* int *pos; // i-th row starts from n[pos[i]] and a[pos[i]]
* int *rsz; // i-th row has rsz[i] non-zero elements, 'rsz' stands for row size
* int **col
* double **data;
* std::vector<intdbl_t> *tmp; // used only when unpacked
*};
*/
struct st_rmsm *rmsm_create(const int size)
{
//fprintf(stderr,"rmsm_create()\n");
struct st_rmsm *m=(struct st_rmsm*)mkl_malloc(sizeof(struct st_rmsm),64);
assert(m);
m->status=0;
m->size = size;
m->pos = (int*)mkl_malloc(sizeof(int)*size,64);
m->rsz = (int*)mkl_malloc(sizeof(int)*size,64);
m->col = NULL;
m->data = NULL;
m->tmp = new std::vector<intdbl_t>[size];
assert(m->pos);
assert(m->rsz);
assert(m->tmp);
m->status=1;
return m;
}
int save_data_real(MKL_Complex16 *data, SimulationData &sim_data, const char * filename) {
double *data2;
data2 = (double*)mkl_malloc(sim_data.num_points * sizeof(double), 64);
for (int i = 0; i < sim_data.num_points; ++i) {
data2[i] = data[i].real;
}
FILE* pFile;
pFile = fopen(filename, "wb");
fwrite(data2, sizeof(double), sim_data.num_points, pFile);
fclose(pFile);
return 0;
}
CMatrix3D::CMatrix3D(MKL_INT r,MKL_INT c, MKL_INT z)
{
//Since we are planning on very large non-sparse arrays therefore we will make this a list of CMatrix pointers
//this will enable the creation of large matrix arrays in different memory regions thus avoid wasting memory at the cost of performance
pMats = (CMatrix*) mkl_malloc(z*sizeof(CMatrix),64);
depth = z;
for(int i =0; i < z;i++)
{
pMats[i].Create(r,c);
}
}
int main()
{
double *A, *B, *C;
int m, n, p, i, j;
double alpha, beta;
printf ("\n This example computes real matrix C=alpha*A*B+beta*C using \n"
" Intel® MKL function dgemm, where A, B, and C are matrices and \n"
" alpha and beta are double precision scalars\n\n");
m = 2000, p = 200, n = 1000;
printf (" Initializing data for matrix multiplication C=A*B for matrix \n"
" A(%ix%i) and matrix B(%ix%i)\n\n", m, p, p, n);
alpha = 1.0; beta = 0.0;
printf (" Allocating memory for matrices aligned on 64-byte boundary for better \n"
" performance \n\n");
A = (double *)mkl_malloc( m*p*sizeof( double ), 64 );
B = (double *)mkl_malloc( p*n*sizeof( double ), 64 );
C = (double *)mkl_malloc( m*n*sizeof( double ), 64 );
if (A == NULL || B == NULL || C == NULL) {
printf( "\n ERROR: Can't allocate memory for matrices. Aborting... \n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
return 1;
}
printf (" Intializing matrix data \n\n");
for (i = 0; i < (m*p); i++) {
A[i] = (double)(i+1);
}
for (i = 0; i < (p*n); i++) {
B[i] = (double)(-i-1);
}
for (i = 0; i < (m*n); i++) {
C[i] = 0.0;
}
printf (" Computing matrix product using Intel® MKL dgemm function via CBLAS interface \n\n");
cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans,
m, n, p, alpha, A, p, B, n, beta, C, n);
printf ("\n Computations completed.\n\n");
printf (" Top left corner of matrix A: \n");
for (i=0; i<min(m,6); i++) {
for (j=0; j<min(p,6); j++) {
printf ("%12.0f", A[i+j*p]);
}
printf ("\n");
}
printf ("\n Top left corner of matrix B: \n");
for (i=0; i<min(p,6); i++) {
for (j=0; j<min(n,6); j++) {
printf ("%12.0f", B[j+i*n]);
}
printf ("\n");
}
printf ("\n Top left corner of matrix C: \n");
for (i=0; i<min(m,6); i++) {
for (j=0; j<min(n,6); j++) {
printf ("%12.5G", C[j+i*n]);
}
printf ("\n");
}
printf ("\n Deallocating memory \n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
printf (" Example completed. \n\n");
return 0;
}
Matrix(const int n): num(n) {
data = (double*)mkl_malloc(sizeof(double)*n*n,64);
assert(data);
}
// X: a MxD matrix, Y: a M vector, W: a M vector
// W0: a M vector
int main(int argc, char ** argv){
if (argc>1 && argv[1][0]=='h') {
printf ("Usage: parSymSGD M D T C lamda r\n");
printf (" M: number of data points, D: dimensions, T: time iterations, C: cores;\n");
printf (" lamda: learning rate, r: panel size in unit of C.\n");
return 1;
}u
// read in the arguments: M, D, I (time iterations), C (cores), r (each panel contains r*C points)
int M = argc>1?atoi(argv[1]):32;
int D = argc>2?atoi(argv[2]):4;
T = argc>3?atoi(argv[3]):10;
int C = argc>4?atoi(argv[4]):4;
float lamda = argc>5?atof(argv[5]):0.01;
int r = argc>6?atoi(argv[6]):1;
///printf("M=%d, D=%d, T=%d, C=%d, lamda=%8.6f, r=%d\n",M,D,T,C,lamda,r);
int max_threads = mkl_get_max_threads(); // get the max number of threads
int rep;
mkl_set_num_threads(1); // set the number of threads to use by mkl
panelSz = C*r;
panels = M/panelSz;
int i,j,k,p,t;
float *Y, *Wreal, *W, *X;
Y = (float *) mkl_malloc(M*sizeof(float),PAGESIZE);
Wreal = (float *) mkl_malloc(D*sizeof(float),PAGESIZE);
W = (float *) mkl_malloc(D*sizeof(float),PAGESIZE);
X = (float *) mkl_malloc(M*D*sizeof(float),PAGESIZE);
float *Ypred = (float*)mkl_malloc(M*sizeof(float),PAGESIZE);
float *Ytmp = (float*)mkl_malloc(M*sizeof(float),PAGESIZE);
float *I = (float*)mkl_malloc(D*D*sizeof(float),PAGESIZE);
float *Z = (float*)mkl_malloc(M*D*sizeof(float),PAGESIZE);
float *B = (float*)mkl_malloc(panels*D*sizeof(float),PAGESIZE);
if (Y==NULL | Wreal==NULL | W==NULL | X==NULL | Ypred==NULL || Ytmp==NULL || Z==NULL || B==NULL || I== NULL){
printf("Memory allocation error.\n");
return 2;
}
initData(Wreal,W,X,Y, M, D,I);
///printf("panelSz=%d, panels=%d\n", panelSz, panels);
for (nt=1; nt<=max_threads && nt<=panelSz; nt*=2){
omp_set_num_threads(nt);// set the number of openMP threads
for (rep=0; rep<REPEATS; rep++){//repeat measurements
double prepTime, gdTime, sInit;
// preprocessing
sInit=dsecnd();
//preprocessSeq(X, Y, Z, B, panelSz, panels, M, D, lamda);
preprocessPar(X, Y, Z, B, panelSz, panels, M, D, lamda);
prepTime = (dsecnd() - sInit);
///dump2("Z",Z,M,D);
///dump2("B",B,panels,D);
// GD
initW(W,D);
///dump1("W (initial)", W, D);
sInit=dsecnd();
float err;
float fixpoint = 0.0;
for (t=0;t<T;t++){
for (p=0;p<panels;p++){
gd(&(X[p*panelSz*D]),&(Z[p*panelSz*D]), &(B[p*D]), panelSz, D, lamda, W, I);
///printf("(t=%d, p=%d) ",t,p);
///dump1("W", W, D);
///err=calErr(X, Ypred, Ytmp, Y, W, M, D);
printf("finish one panels ............................ \n");
}
}
gdTime = (dsecnd() - sInit);
err=calErr(X, Ypred, Ytmp, Y, W, M, D);
fixpoint = err - prev_err;
// print final err. time is in milliseconds
printf("nt=%d\t ttlTime=%.5f\t prepTime=%.5f\t gdTime=%.5f\t error=%.5f\n", nt, (gdTime+prepTime)*1000, prepTime*1000, gdTime*1000, err);
}
}
if (B) mkl_free(B);
if (Z) mkl_free(Z);
if (Ytmp) mkl_free(Ytmp);
if (Ypred) mkl_free(Ypred);
if (Y) mkl_free(Y);
if (Wreal) mkl_free(Wreal);
if (W) mkl_free(W);
if (X) mkl_free(X);
if (I) mkl_free(I);
return 0;
}
int main()
{
double *A, *B, *C;
int m, n, p, i, j, k, r;
double alpha, beta;
double sum;
double s_initial, s_elapsed;
printf ("\n This example measures performance of rcomputing the real matrix product \n"
" C=alpha*A*B+beta*C using a triple nested loop, where A, B, and C are \n"
" matrices and alpha and beta are double precision scalars \n\n");
m = 2000, p = 200, n = 1000;
printf (" Initializing data for matrix multiplication C=A*B for matrix \n"
" A(%ix%i) and matrix B(%ix%i)\n\n", m, p, p, n);
alpha = 1.0; beta = 0.0;
printf (" Allocating memory for matrices aligned on 64-byte boundary for better \n"
" performance \n\n");
A = (double *)mkl_malloc( m*p*sizeof( double ), 64 );
B = (double *)mkl_malloc( p*n*sizeof( double ), 64 );
C = (double *)mkl_malloc( m*n*sizeof( double ), 64 );
if (A == NULL || B == NULL || C == NULL) {
printf( "\n ERROR: Can't allocate memory for matrices. Aborting... \n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
return 1;
}
printf (" Intializing matrix data \n\n");
for (i = 0; i < (m*p); i++) {
A[i] = (double)(i+1);
}
for (i = 0; i < (p*n); i++) {
B[i] = (double)(-i-1);
}
for (i = 0; i < (m*n); i++) {
C[i] = 0.0;
}
printf (" Making the first run of matrix product using triple nested loop\n"
" to get stable run time measurements \n\n");
for (i = 0; i < m; i++) {
for (j = 0; j < n; j++) {
sum = 0.0;
for (k = 0; k < p; k++)
sum += A[p*i+k] * B[n*k+j];
C[n*i+j] = sum;
}
}
printf (" Measuring performance of matrix product using triple nested loop \n\n");
s_initial = dsecnd();
for (r = 0; r < LOOP_COUNT; r++) {
for (i = 0; i < m; i++) {
for (j = 0; j < n; j++) {
sum = 0.0;
for (k = 0; k < p; k++)
sum += A[p*i+k] * B[n*k+j];
C[n*i+j] = sum;
}
}
}
s_elapsed = (dsecnd() - s_initial) / LOOP_COUNT;
printf (" == Matrix multiplication using triple nested loop completed == \n"
" == at %.5f milliseconds == \n\n", (s_elapsed * 1000));
printf (" Deallocating memory \n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
if (s_elapsed < 0.9/LOOP_COUNT) {
s_elapsed=1.0/LOOP_COUNT/s_elapsed;
i=(int)(s_elapsed*LOOP_COUNT)+1;
printf(" It is highly recommended to define LOOP_COUNT for this example on your \n"
" computer as %i to have total execution time about 1 second for reliability \n"
" of measurements\n\n", i);
}
printf (" Example completed. \n\n");
return 0;
}
int test01(void)
{
int err=0;
printf("TEST01\n");
printf(" |Test solver_v2 workflow\n");
//struct st_mesh *q=mshio_create_mesh("msh/0344/0344");
struct st_mesh *q=mshio_create_mesh("msh/0616/0616");
//mshio_print_mesh(q,PRINT_INFO_VERBOSE);
/*
* ipar[0] = M
* ipar[1] = Nd
* ipar[2] = pad
* ipar[3] = rule1
* ipar[4] = rule2
* ipar[5] = nu
* ipar[6] = nv
* ipar[7] = num_threads in omp
*/
const int ipar[128]={1,3,1, 1,1,5,3, 1};
/*
* dpar[0] = g factor
* dpar[1] = mua (absorption coefficient)
* dpar[2] = mus (scattering coefficient)
*/
const double dpar[128]={0.7,1.0,2.0};
struct st_solver_v2 *s=sv2_create_solver(q,ipar,dpar);
sv2_print_solverinfo(s);
//for (int i = 0; i < s->Nt; i++)
//rmsm_print_row(s->E,i);
double _Complex *b0=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *x0=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *b1=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *x1=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
assert(b0);
assert(x0);
assert(b1);
assert(x1);
int nitr;
double eps;
//sv2_gen_b0(s,0.0,b0);
for (int i = 0; i < s->Ng; i++)
x0[i] = 1.0;
print_vector("x=",5,x0);
sv2_mul(s,x0,b0);
print_vector("b=A.x=",5,b0);
sv2_solve(s,b0,x0,200,12,1.0E-13,&nitr,&eps);
//sv2_gen_b1x0(s,0.0,b1,x0);
//sv2_solve(s,b1,x1,200,12,1.0E-13,&nitr,&eps);
sv2_destroy_solver(s);
mshio_destroy_mesh(q);
printf("END OF TEST01\n");
printf("\n");
return err;
}
int test01(void)
{
int err=0;
printf("TEST01\n");
printf(" |Test solver_v1 workflow\n");
struct st_mesh *q=mshio_create_mesh("msh/0344/0344");
//mshio_print_mesh(q,PRINT_INFO_VERBOSE);
/*
* ipar[0] = M
* ipar[1] = Nd
* ipar[2] = pad
* ipar[3] = rule1
* ipar[4] = rule2
* ipar[5] = nu
* ipar[6] = nv
* ipar[7] = num_threads in omp
*/
//const int ipar[128]={1,3,1, 1,1,5,3, 1};
const int ipar[128]={1,3,1, 1,1,5,3, 8};
//const int ipar[128]={1,30,1, 2,5,5,3, 8};
/*
* dpar[0] = g factor
* dpar[1] = mua (absorption coefficient)
* dpar[2] = mus (scattering coefficient)
*/
const double dpar[128]={0.7,1.0,2.0};
struct st_solver_v1 *s=sv1_create_solver(q,ipar,dpar);
sv1_print_solverinfo(s);
//for (int i = 0; i < s->Ns; i++)
//printf("[%5d] %.5E\n",i,s->E[i]);
double _Complex *b0=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *x0=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *b1=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
double _Complex *x1=(double _Complex*)mkl_malloc(sizeof(double _Complex)*s->Ng,64);
assert(b0);
assert(x0);
assert(b1);
assert(x1);
int nitr;
double eps;
//sv1_gen_b0(s,0.0,b0);
//sv1_solve(s,b0,x0,200,12,1.0E-13,&nitr,&eps);
sv1_gen_b1x0(s,0.0,b1,x0);
sv1_solve(s,b1,x1,200,12,1.0E-13,&nitr,&eps);
char dir[FILENAME_MAX]="SOL";
sv1_save_solution(s,x1,dir);
sv1_destroy_solver(s);
mshio_destroy_mesh(q);
printf("END OF TEST01\n");
printf("\n");
return err;
}
void rmsm_pack(struct st_rmsm *m)
{
//fprintf(stderr,"rmsm_pack()\n");
assert(m->status==1);
//for (int i = 0; i < m->size; i++)
//printf("[%5d] %lu\n",i,m->tmp[i].size());
//const int row=73; // examine this row
//printf("row %d (raw)\n",row);
//for (int i = 0; i < m->tmp[row].size(); i++)
//printf("[%5d] %f\n",m->tmp[row].at(i).i,m->tmp[row].at(i).d);
// sort each row
for (int n = 0; n < m->size; n++)
std::sort(m->tmp[n].begin(),m->tmp[n].end(),cmp);
//printf("(sorted)\n");
//for (int i = 0; i < m->tmp[row].size(); i++)
//printf("[%5d] %f\n",m->tmp[row].at(i).i,m->tmp[row].at(i).d);
// merge
std::vector<int> *vi = new std::vector<int>[m->size];
std::vector<double> *vd = new std::vector<double>[m->size];
for (int n = 0; n < m->size; n++) {
if (m->tmp[n].size()==0) continue;
vi[n].push_back(m->tmp[n].at(0).i);
vd[n].push_back(m->tmp[n].at(0).d);
int curr_col=vi[n].at(0);
for (int j = 1; j < m->tmp[n].size(); j++) {
int col=m->tmp[n].at(j).i;
double val=m->tmp[n].at(j).d;
if (col==curr_col)
vd[n].back() += val;
else {
vi[n].push_back(col);
vd[n].push_back(val);
curr_col = col;
}
}
}
delete [] m->tmp;
//printf("(merged)\n");
//for (int i = 0; i < vi[row].size(); i++)
//printf("[%5d] %f\n",vi[row].at(i),vd[row].at(i));
// pack
{ int ptr=0;
for (int i = 0; i < m->size; i++) {
m->pos[i] = ptr;
ptr += vi[i].size();
m->rsz[i] = vi[i].size();
}
m->length=ptr;
}
m->col =(int*) mkl_malloc(sizeof(int) *(m->length),64);
m->data=(double*)mkl_malloc(sizeof(double)*(m->length),64);
for (int i = 0; i < m->size; i++)
for (int j = 0; j < vi[i].size(); j++) {
m->col [j+m->pos[i]] = vi[i].at(j);
m->data[j+m->pos[i]] = vd[i].at(j);
}
delete [] vi;
delete [] vd;
//printf("(packed)\n");
//for (int i = 0; i < m->rsz[row]; i++)
//printf("[%5d] %f\n",m->col[i+m->pos[row]],m->data[i+m->pos[row]]);
m->status=2;
}
void GeneticAlgorithm::initializeParameters(double* dataSet, int dataSetLength, int nPopulation, double scaleFactor, double crossingProbability){
_scaleFactor = scaleFactor;
_crossingProbability = crossingProbability;
_dataSetLength = dataSetLength;
_dataSet = dataSet;
_residualArray = new double*[nThreads];
_paramArray = new double*[nThreads];
for(int i = 0; i < nThreads; i++){
_residualArray[i] = new double[_dataSetLength];
_paramArray[i] = new double[nParams];
}
_populationParametersOld = (Parameters::fitParameters *) mkl_malloc(sizeof(Parameters::fitParameters)*nPopulation,16);
_populationParametersNew = (Parameters::fitParameters *) mkl_malloc(sizeof(Parameters::fitParameters)*nPopulation,16);
for(int i = 0; i < _nPopulation; i++){
_populationParametersOld[i].missFreq = new long[_nMissing];
_populationParametersNew[i].missFreq = new long[_nMissing];
/* _populationParametersOld[i].c11 = (randomDouble(196,196.1))*pow(10,9);
_populationParametersOld[i].c22 = _populationParametersOld[i].c11;
_populationParametersOld[i].c33 = (randomDouble(187,187.1))*pow(10,9);
_populationParametersOld[i].c44 = (randomDouble(63.5,63.6))*pow(10,9);
_populationParametersOld[i].c55 = _populationParametersOld[i].c44;
_populationParametersOld[i].c66 = (randomDouble(55.7,55.8))*pow(10,9);
_populationParametersOld[i].c12 = (randomDouble(62.5,62.6))*pow(10,9);
_populationParametersOld[i].c13 = (randomDouble(69.8,69.9))*pow(10,9);
_populationParametersOld[i].c23 = _populationParametersOld[i].c13;
_populationParametersOld[i].chiSq = calculateResidual(&_populationParametersOld[i],0);/// ***Tetragonal PuCoGa5*/
/* _populationParametersOld[i].c11 = (randomDouble(260,300))*pow(10,9);
_populationParametersOld[i].c22 = _populationParametersOld[i].c11;
_populationParametersOld[i].c33 = (randomDouble(290,320))*pow(10,9);
_populationParametersOld[i].c44 = (randomDouble(90,110))*pow(10,9);
_populationParametersOld[i].c55 = _populationParametersOld[i].c44;
_populationParametersOld[i].c66 = (randomDouble(130,150))*pow(10,9);
_populationParametersOld[i].c12 = (randomDouble(140,165))*pow(10,9);
_populationParametersOld[i].c13 = (randomDouble(100,130))*pow(10,9);
_populationParametersOld[i].c23 = _populationParametersOld[i].c13;
_populationParametersOld[i].chiSq = calculateResidual(&_populationParametersOld[i],0);/// ***Tetragonal URu2Si2
*/
_populationParametersOld[i].c11 = (randomDouble(220,260))*pow(10,9);
_populationParametersOld[i].c22 = (randomDouble(210,250))*pow(10,9);
_populationParametersOld[i].c33 = (randomDouble(100,150))*pow(10,9);
_populationParametersOld[i].c44 = (randomDouble(32,38))*pow(10,9);
_populationParametersOld[i].c55 = (randomDouble(48,52))*pow(10,9);
_populationParametersOld[i].c66 = (randomDouble(94,98))*pow(10,9);
_populationParametersOld[i].c12 = (randomDouble(100,150))*pow(10,9);
_populationParametersOld[i].c13 = (randomDouble(25,60))*pow(10,9);
_populationParametersOld[i].c23 = (randomDouble(20,70))*pow(10,9);
_populationParametersOld[i].chiSq = calculateResidual(&_populationParametersOld[i],0);/// ***Orthorhombic YBCO67
//_populationParametersOld[i].c11 = (randomDouble(1,400))*pow(10,9);
//_populationParametersOld[i].c22 = _populationParametersOld[i].c11;
//_populationParametersOld[i].c33 = _populationParametersOld[i].c11;
//
//_populationParametersOld[i].c44 = (randomDouble(1,400))*pow(10,9);
//_populationParametersOld[i].c55 = _populationParametersOld[i].c44;
//_populationParametersOld[i].c66 = _populationParametersOld[i].c44;
//_populationParametersOld[i].c12 = (randomDouble(1,400))*pow(10,9);
//_populationParametersOld[i].c13 = _populationParametersOld[i].c12;
//_populationParametersOld[i].c23 = _populationParametersOld[i].c12;
//_populationParametersOld[i].chiSq = calculateResidual(&_populationParametersOld[i],0);/// ***Cubic Nb
//_populationParametersOld[i].c11 = (randomDouble(120,200))*pow(10,9);
//_populationParametersOld[i].c22 = _populationParametersOld[i].c11;
//
//_populationParametersOld[i].c33 = (randomDouble(120,200))*pow(10,9);
//_populationParametersOld[i].c44 = (randomDouble(10,100))*pow(10,9);
//_populationParametersOld[i].c55 = _populationParametersOld[i].c44;
//
//_populationParametersOld[i].c66 = (randomDouble(10,100))*pow(10,9);
//_populationParametersOld[i].c12 = (randomDouble(10,100))*pow(10,9);
//_populationParametersOld[i].c13 = (randomDouble(10,100))*pow(10,9);
//_populationParametersOld[i].c23 = _populationParametersOld[i].c13;
//_populationParametersOld[i].chiSq = calculateResidual(&_populationParametersOld[i],0);/// ***Tetragonal CeCoIn5
}
_minimumParameters.c11 = 1;
_minimumParameters.c22 = 1;
_minimumParameters.c33 = 1;
_minimumParameters.c44 = 1;
_minimumParameters.c55 = 1;
_minimumParameters.c66 = 1;
_minimumParameters.c12 = 1;
_minimumParameters.c13 = 1;
_minimumParameters.c23 = 1;
_minimumParameters.chiSq = std::numeric_limits<double>::infinity();
_minimumParameters.missFreq = new long[_nMissing];
}
void GeneticAlgorithm2::initializeParameters(double** dataSet, int dataSetLength, int nPopulation, double scaleFactor, double crossingProbability){
_scaleFactor = scaleFactor;
_crossingProbability = crossingProbability;
_dataSetLength = dataSetLength;
_dataSet = dataSet;
_residualArray = new double*[nThreads];
_paramArray = new double*[nThreads];
for(int i = 0; i < nThreads; i++){
_residualArray[i] = new double[_dataSetLength];
_paramArray[i] = new double[nParams];
// xVals[i] = new double[4];
}
_populationParametersOld = (Parameters::fitParameters *) mkl_malloc(sizeof(Parameters::fitParameters)*nPopulation,16);
_populationParametersNew = (Parameters::fitParameters *) mkl_malloc(sizeof(Parameters::fitParameters)*nPopulation,16);
for(int i = 0; i < _nPopulation; i++){
_populationParametersOld[i].A1 = randomDouble(0.0,1.0);
_populationParametersOld[i].A2 = randomDouble(0.0,1.0);
_populationParametersOld[i].F1 = randomDouble(460,480);
_populationParametersOld[i].F2 = randomDouble(520,540);
_populationParametersOld[i].dF1 = randomDouble(25.,45.0);
_populationParametersOld[i].dF2 = 0;
_populationParametersOld[i].phi1 = randomDouble(0.0,1.0);
_populationParametersOld[i].phi2 = randomDouble(0.0,1.0);
_populationParametersOld[i].Td1 = randomDouble(4., 8.);
_populationParametersOld[i].Td2 = randomDouble(4., 8.);
// _populationParametersOld[i].ms1 = randomDouble(1,2.2);
// _populationParametersOld[i].ms2 = randomDouble(1.5,1.7);
_populationParametersOld[i].ms1 = randomDouble(1.0,1.4);
_populationParametersOld[i].ms2 = randomDouble(1.5,1.7);
// _populationParametersOld[i].m1 = randomDouble(1.0,3);
// _populationParametersOld[i].m2 = randomDouble(1.0,3);
_populationParametersOld[i].m1 = 1.7;
_populationParametersOld[i].m2 = 1.7;
// _populationParametersOld[i].dF12 = randomDouble(-15,5);
// _populationParametersOld[i].ms11 = randomDouble(0.0,.2);
// _populationParametersOld[i].ms22 = randomDouble(0.0,.2);
_populationParametersOld[i].T = 4.2;
_populationParametersOld[i].chiSq = calculateResidual2(&_populationParametersOld[i],0);
}
for(int i = 0; i < _nPopulation; i++){
_populationParametersNew[i].T = 0;
_populationParametersNew[i].m1 = 0;
_populationParametersNew[i].m2 = 0;
}
_minimumParameters.A1 = 1;
_minimumParameters.A2 = 1;
_minimumParameters.F1 = 0;
_minimumParameters.F2 = 0;
_minimumParameters.dF1 = 0;
_minimumParameters.dF2 = 0;
_minimumParameters.phi1 = 0;
_minimumParameters.phi2 = 0;
_minimumParameters.Td1 = 0;
_minimumParameters.Td2 = 0;
_minimumParameters.ms1 = 0;
_minimumParameters.ms2 = 0;
_minimumParameters.m1 = 0;
_minimumParameters.m2 = 0;
// _minimumParameters.dF12 = 0;
// _minimumParameters.ms11 = 0;
// _minimumParameters.ms22 = 0;
_minimumParameters.T = 0;
_minimumParameters.chiSq = INFINITE;
}
int bench_dgemm()
{
double *A, *B, *C;
int m, n, k, i, j;
double alpha, beta;
double t;
m = SIZE, k = SIZE, n = SIZE;
DPRINTF(" Initializing data for matrix multiplication C=A*B for matrix \n"
" A(%ix%i) and matrix B(%ix%i)\n\n", m, k, k, n);
alpha = 1.0; beta = 0.0;
DPRINTF(" Allocating memory for matrices aligned on 64-byte boundary for better \n"
" performance \n\n");
A = (double *)mkl_malloc( m*k*sizeof( double ), 64 );
B = (double *)mkl_malloc( k*n*sizeof( double ), 64 );
C = (double *)mkl_malloc( m*n*sizeof( double ), 64 );
if (A == NULL || B == NULL || C == NULL) {
printf( "\n ERROR: Can't allocate memory for matrices. Aborting... \n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
return 1;
}
DPRINTF(" Intializing matrix data \n\n");
#pragma omp parallel for
for (i = 0; i < (m*k); i++) {
A[i] = (double)(i+1);
}
#pragma omp parallel for
for (i = 0; i < (k*n); i++) {
B[i] = (double)(-i-1);
}
#pragma omp parallel for
for (i = 0; i < (m*n); i++) {
C[i] = 0.0;
}
DPRINTF(" Computing matrix product using Intel(R) MKL dgemm function via CBLAS interface \n\n");
t=stoptime();
cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans,
m, n, k, alpha, A, k, B, n, beta, C, n);
t=stoptime()-t;
printf("calculation time : %f\n",t);
printf("gflops/s : %f\n",((2.0*m*n*k)*1E-9)/t);
DPRINTF("\n Computations completed.\n\n");
DPRINTF(" Top left corner of matrix A: \n");
for (i=0; i<min(m,6); i++) {
for (j=0; j<min(k,6); j++) {
DPRINTF("%12.0f", A[j+i*k]);
}
DPRINTF("\n");
}
DPRINTF("\n Top left corner of matrix B: \n");
for (i=0; i<min(k,6); i++) {
for (j=0; j<min(n,6); j++) {
DPRINTF("%12.0f", B[j+i*n]);
}
DPRINTF("\n");
}
DPRINTF("\n Top left corner of matrix C: \n");
for (i=0; i<min(m,6); i++) {
for (j=0; j<min(n,6); j++) {
DPRINTF("%12.5G", C[j+i*n]);
}
DPRINTF("\n");
}
DPRINTF("\n Deallocating memory \n\n");
mkl_free(A);
mkl_free(B);
mkl_free(C);
DPRINTF(" Example completed. \n\n");
return 0;
}
