int main(void)
{
Item x = 5, y = 2, x2 = 6, y2 = 3;
mat m1, m2, k1, k2;
int i, j;
printf("\nItems: ");
item_print(x);
printf(" ");
item_print(y);
printf("\n");
printf("Item Addition: ");
item_print(item_add(x, y));
printf("\n");
printf("Item Multiplication: ");
item_print(item_mul(x, y));
printf("\n");
printf("Item Substraction: ");
item_print(item_sub(x, y));
printf("\n");
printf("Item Division: ");
item_print(item_div(x, y));
printf("\n\n");
m1 = mat_create(3, 2);
m2 = mat_create(2, 3);
k1 = mat_create(2, 2);
k2 = mat_create(2, 2);
for (i = 0; i < mat_rows(m1); i++)
for (j = 0; j < mat_cols(m1); j++)
mat_add_elem(m1, i, j, x);
for (i = 0; i < mat_rows(m2); i++)
for (j = 0; j < mat_cols(m2); j++)
mat_add_elem(m2, i, j, y);
for (i = 0; i < mat_rows(k1); i++)
for (j = 0; j < mat_cols(k1); j++)
{
mat_add_elem(k1, i, j, x2);
mat_add_elem(k2, i, j, y2);
}
printf("k1 Matrix: \n");
mat_print(k1);
printf("k2 Matrix: \n");
mat_print(k2);
printf("Addition: \n");
mat_print(mat_add(k1, k2));
printf("Substraction: \n");
mat_print(mat_sub(k1, k2));
printf("m1 Matrix: \n");
mat_print(m1);
printf("m2 Matrix: \n");
mat_print(m2);
printf("Multiplication: \n");
mat_print(mat_mul(m1, m2));
return 0;
}
void bidiag_gkl_restart(
int locked, int l, int n,
CAX && Ax, CATX && Atx, CD && D, CE && E, CRho && rho, CP && P, CQ && Q, int s_indx, int t_s_indx) {
// enhancements version from SLEPc
const double eta = 1.e-10;
double t_start = 0.0, t_end = 0.0;
double local_start = 0.0, local_end = 0.0;
double t_total3 = 0.0, t_total4 = 0.0, t_total5 = 0.0, t_total6 = 0.0, t_total7 = 0.0;
int rank, nprocs;
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
// Step 1
int recv_len = (int)P.dim0() * nprocs;
vec_container<double> tmp(Ax.dim0());
vec_container<double> recv_tmp(recv_len);
auto m_Ax = make_gemv_ax(&Ax);
auto m_Atx = make_gemv_ax(&Atx);
m_Ax(Q.col(l), tmp, P.dim0() > 1000);
vec_container<double> send_data(P.dim0(),0);
for(size_t i = s_indx; i < s_indx + Ax.dim0(); ++i)
send_data[i] = tmp.get(i-s_indx);
MPI_Gather(&send_data[0], P.dim0(), MPI_DOUBLE, &recv_tmp[0], P.dim0(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
P.col(l) = 0;
// Generate truly P.col(l)
if(rank == 0) {
local_union(P, recv_tmp, l, nprocs);
// Step 2 & also in rank 0
for (int j = locked; j < l; ++j) {
P.col(l) += -rho(j) * P.col(j);
}
}
MPI_Bcast(&(P.col(0)[0]), P.size(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
//MPI_Bcast(&(P.col(l)[0]), P.size(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
// Main loop
vec_container<double> T(n);
int recv_l = Q.dim0() * nprocs;
vec_container<double> recv_t(recv_l);
for (int j = l; j < n; ++j) {
// Step 3
vec_container<double> tmp2(Atx.dim0());
/* for print */
if(rank == 0)
t_start = currenttime();
local_start = currenttime();
m_Atx(P.col(j), tmp2, Q.dim0() > 1000);
local_end = currenttime();
std::cout << "parallel mv time cost is " << (local_end - local_start) / 1.0e6 << std::endl;
vec_container<double> s_data(Q.dim0(), 0);
for(size_t i = t_s_indx; i < t_s_indx + Atx.dim0(); ++i)
s_data[i] = tmp2[i-t_s_indx];
MPI_Gather(&s_data[0], Q.dim0(), MPI_DOUBLE, &recv_t[0], Q.dim0(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
local_start = currenttime();
std::cout << "parallel mv time cost2 is " << (local_start - local_end) / 1.0e6 << std::endl;
//Q.col(j+1) = 0;
if(rank == 0) {
// Generate truly Q.col(j+1)
local_union(Q, recv_t, j + 1, nprocs);
local_end = currenttime();
t_end = currenttime();
std::cout << "parallel mv time cost3 is " << (local_end - local_start) / 1.0e6 << std::endl;
std::cout << "time of step 3 is : " << (t_end - t_start) / 1.0e6 << std::endl;
t_total3 += (t_end - t_start) / 1.0e6;
}
// Step 4
for(size_t aa = 0; aa < Q.dim0(); ++aa) // row
MPI_Bcast(&(Q.row(aa)[0]), j + 2, MPI_DOUBLE, 0, MPI_COMM_WORLD);
// MPI_Bcast(&(Q.col(0)[0]), Q.size(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
if(rank == 0)
t_end = currenttime();
auto Qj = mat_cols(Q, 0, j + 1);
auto Tj = make_vec(&T, j + 1);
//Tj.assign(gemv(Qj.trans(), Q.col(j + 1)), j >= 3);
parallel_gemv_task(Qj.trans(), Q.col(j+1), Tj);
if(rank == 0) {
t_start = currenttime();
t_total4 += (t_start - t_end) / 1.0e6;
std::cout << "time of step 4 is : " << (t_start - t_end) / 1.0e6 << std::endl;
}
// Step 5
if(rank == 0) {
double r = Q.col(j + 1).norm2();
D[j] = vec_unit(P.col(j));
Q.col(j + 1).scale(1. / D[j]);
Tj = Tj / D[j];
r /= D[j];
Q.col(j + 1).plus_assign(- gemv(Qj, Tj), Q.dim0() > 1000);
t_end = currenttime();
t_total5 += (t_end - t_start) / 1.0e6;
std::cout << "time of step 5 is : " << (t_end - t_start) / 1.0e6 << std::endl;
// Step 6
double beta = r * r - Tj.square_sum();
if (beta < eta * r * r) {
Tj.assign(gemv(Qj.trans(), Q.col(j + 1)), Q.dim0() > 1000);
r = Q.col(j + 1).square_sum();
Q.col(j + 1).plus_assign(-gemv(Qj, Tj), Q.dim0() > 1000);
beta = r * r - Tj.square_sum();
}
beta = std::sqrt(beta);
E[j] = beta;
Q.col(j + 1).scale(1. / E[j]);
t_start = currenttime();
t_total6 += (t_start - t_end) / 1.0e6;
std::cout << "time of step 6 is : " << (t_start - t_end) / 1.0e6 << std::endl;
}
// Step 7
// MPI_Bcast(&(Q.col(j+1)[0]), Q.size(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
// MPI_Bcast(&(Q.col(0)[0]), Q.size(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
for(size_t aa = 0; aa < Q.dim0(); ++aa)
MPI_Bcast(&(Q.col(j+1)[aa]), 1, MPI_DOUBLE, 0, MPI_COMM_WORLD);
if (j + 1 < n) {
if(rank == 0)
t_start = currenttime();
vec_container<double> tmp3(Ax.dim0());
vec_container<double> se_data(P.dim0(), 0);
m_Ax(Q.col(j + 1), tmp3, P.dim0() > 1000);
for(size_t k1 = s_indx; k1 < s_indx + Ax.dim0(); ++k1)
se_data[k1] = tmp3[k1-s_indx];
MPI_Gather(&se_data[0], P.dim0(), MPI_DOUBLE, &recv_tmp[0], P.dim0(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
// P.col(j+1) = 0;
if(rank == 0) {
local_union(P, recv_tmp, j + 1, nprocs);
P.col(j + 1).plus_assign(- E[j] * P.col(j), P.dim0() > 1000);
}
/* for print */
if(rank == 0) {
t_end = currenttime();
t_total7 += (t_end - t_start) / 1.0e6;
std::cout << "time of step 7 is : " << (t_end - t_start) / 1.0e6 << std::endl;
}
// MPI_Bcast(&(P.col(l)[0]), P.size(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
// MPI_Bcast(&(P.col(0)[0]), P.size(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
for(size_t aa = 0; aa < P.dim0(); ++aa)
MPI_Bcast(&(P.col(j+1)[aa]), 1, MPI_DOUBLE, 0, MPI_COMM_WORLD);
} // end if
} // end while
/* for print time of each step. */
if(rank == 0) {
std::cout << "total step 3 time is : " << t_total3 << std::endl;
std::cout << "total step 4 time is : " << t_total4 << std::endl;
std::cout << "total step 5 time is : " << t_total5 << std::endl;
std::cout << "total step 6 time is : " << t_total6 << std::endl;
std::cout << "total step 7 time is : " << t_total7 << std::endl;
}
return ;
}
