int main(int argc, char *argv[]) {
size_t na = 1000; /* number of atoms */
size_t nr = 10000; /* number of reflections */
int compute_serial = 0;
int times = 1;
TYPE *h; /* h[j,0] == h, h[j,1] == k, h[j,2] == l */
TYPE *E; /* E[j,0] == real part of E, E[j,1] == imag part of E */
TYPE *E1; /* E[j,0] == real part of E, E[j,1] == imag part of E */
TYPE *a; /* a[j,0] == atomic number, a[j,1] == x, a[j,2] == y, a[j,3] == z */
double t0, dt1, dt2;
int i;
int numtask = 1;
if (argc > 1) {
numtask = atoi(argv[1]);
}
if (argc > 2) {
na = atoi(argv[2]);
nr = atoi(argv[3]);
}
if (argc > 4)
times = atoi(argv[4]);
size_t NH = DIM2_H * nr;
size_t NA = DIM2_A * na;
size_t NE = DIM2_E * nr;
/*printf("Computation of crystallographic normalized structure factors\n"
" on the CPU and the GPU\n\n");
printf("Number of atoms: %d\n", na);
printf("Number of reflections: %d\n", nr);
*/
/*
h = (TYPE*) malloc(sizeof(*h) * DIM2_H * nr); // 3*10000 30000
E = (TYPE*) malloc(sizeof(*E) * DIM2_E * nr); // 2*10000 20000
E1 = (TYPE*) malloc(sizeof(*E1) * DIM2_E * nr); // 2*10000 20000
a = (TYPE*) malloc(sizeof(*a) * DIM2_A * na); // 4*1000 4000
*/
posix_memalign((void **) &h, getpagesize(), sizeof(*h) * DIM2_H * nr);
posix_memalign((void **) &E, getpagesize(), sizeof(*E) * DIM2_E * nr);
posix_memalign((void **) &E1, getpagesize(), sizeof(*E1) * DIM2_E * nr);
posix_memalign((void **) &a, getpagesize(), sizeof(*a) * DIM2_A * na);
for (i = 0; i < DIM2_E * nr; i++)
E1[i] = E[i] = 0.0f;
deta(na, a);
deth(nr, h);
int tt;
//printf("Running the GPU code %d times\n", times);
#pragma omp register([NA]a)
#pragma omp register([NH]h)
#pragma omp register([NE]E1)
t0 = omp_get_wtime();
for (tt = 0; tt < times; tt++) {
structfac_gpuss(na, nr, NA, a, NH, h, NE, E1, numtask);
}
#pragma omp taskwait
dt2 = (omp_get_wtime() - t0);/// times;
#if 0
if (compute_serial) {
printf("Cuda: wallclock time seconds:%f\n", dt2);
} else {
printf("computation time (in seconds): %f\n", dt2);
}
#endif
double sumdf = sumdif(E, E1, 2 * nr);
//printf("Cuda: Sumdif: %f mean: %f\n", sumdf, sumdf / nr);
double speed = 1.0e-9 * (NH * NA) / dt2;
printf("%f,%d,%d,%d,%zd,%zd,%f,%.4lf\n", sumdf / nr, times, omp_get_num_threads(), numtask, na, nr, dt2, speed);
return 0;
}
bool gaussseidelMorphed(sData* data, double** s)
{
int curIter=0;
double error;
float tmp;
double a1,a2,a3,a4,a5;
int N=data->dimI-2;
int M=data->dimJ-2;
// allocate memory for derivatives
double ***alpha = new double**[N+2];
double **temp1 = new double*[N+2];
double **temp2 = new double*[N+2];
double **temp3 = new double*[N+2];
double **temp4 = new double*[N+2];
double **temp5 = new double*[N+2];
double **temp6 = new double*[N+2];
double **temp7 = new double*[N+2];
double **temp8 = new double*[N+2];
for (int i=0;i<N+2;i++){
temp1[i] = new double[M+2];
temp2[i] = new double[M+2];
temp3[i] = new double[M+2];
temp4[i] = new double[M+2];
temp5[i] = new double[M+2];
temp6[i] = new double[M+2];
temp7[i] = new double[M+2];
temp8[i] = new double[M+2];
alpha[i] = new double* [M+2];
}
for (int i=0;i<N+2;i++){
for(int j=0;j<M+2;j++){
alpha[i][j] = new double[5];
}
}
// write derivatives
dxi(data,temp1,temp2);
deta(data,temp3,temp4);
ddxi(data,temp5,temp6);
ddeta(data,temp7,temp8);
// calculate alpha
for (int i=1;i<data->dimI-1;i++){
for(int j=1;j<data->dimJ-1;j++){
alpha[i][j][0] = temp1[i][j]*temp1[i][j]+temp2[i][j]*temp2[i][j]; //alpha1
alpha[i][j][1] = temp3[i][j]*temp3[i][j]+temp4[i][j]*temp4[i][j]; //alpha2
alpha[i][j][2] = 2*(temp1[i][j]*temp3[i][j]+temp2[i][j]*temp4[i][j]); //alpha3
alpha[i][j][3] = temp5[i][j]+temp6[i][j]; //alpha4
alpha[i][j][4] = temp7[i][j]+temp8[i][j]; //alpha5
}
}
// free memory
for (int i=0;i<N+2;i++){
delete[] temp1[i];
delete[] temp2[i];
delete[] temp3[i];
delete[] temp4[i];
delete[] temp5[i];
delete[] temp6[i];
delete[] temp7[i];
delete[] temp8[i];
}
delete[] temp1;
delete[] temp2;
delete[] temp3;
delete[] temp4;
delete[] temp5;
delete[] temp6;
delete[] temp7;
delete[] temp8;
while(curIter<data->maxIter) {
/*std::cout << "\r\tGauss-Seidel: Iteration " <<*/ ++curIter;
error =0;
for(int i = 1; i < data->dimI-1; i++)
{
for(int j = 1 ; j < data->dimJ-1; j++)
{
a1 = alpha[i][j][0];
a2 = alpha[i][j][1];
a3 = alpha[i][j][2];
a4 = alpha[i][j][3];
a5 = alpha[i][j][4];
tmp = s[i+1][j+1] * (a3/4.f)
+ s[i+1][j] * (a1+a4/2.f)
+ s[i+1][j-1] * (-a3/4.f)
+ s[i][j+1] * (a2+a5/2.f)
+ s[i][j-1] * (a2-a5/2.f)
+ s[i-1][j+1] * (-a3/4.f)
+ s[i-1][j] * (a1-a4/2.f)
+ s[i-1][j-1] * (a3/4.f);
tmp /=(2*(a1+a2));
error += fabs(tmp-s[i][j]);
s[i][j] = tmp;
}
}
if(error < data->residuum){
std::cout << "Residual r = "<< error << ", after "
<< curIter << "# iterations \t";
return true;
}
}
return true;
}
