dist_5D.c

Go to the documentation of this file.

#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include "../ascot5.h"
#include "../consts.h"
#include "../physlib.h"
#include "dist_5D.h"
#include "../particle.h"

size_t dist_5D_index(int i_r, int i_phi, int i_z, int i_ppara, int i_pperp,
                     int i_time, int i_q, size_t step_6, size_t step_5,
                     size_t step_4, size_t step_3, size_t step_2,
                     size_t step_1) {
    return (size_t)(i_r)     * step_6
         + (size_t)(i_phi)   * step_5
         + (size_t)(i_z)     * step_4
         + (size_t)(i_ppara) * step_3
         + (size_t)(i_pperp) * step_2
         + (size_t)(i_time)  * step_1
         + (size_t)(i_q);
}

int dist_5D_init(dist_5D_data* data) {
 
    size_t n_q     = (size_t)(data->n_q);
    size_t n_time  = (size_t)(data->n_time);
    size_t n_pperp = (size_t)(data->n_pperp);
    size_t n_ppara = (size_t)(data->n_ppara);
    size_t n_z     = (size_t)(data->n_z);
    size_t n_phi   = (size_t)(data->n_phi);
    data->step_6 = n_q * n_time * n_pperp * n_ppara * n_z * n_phi;
    data->step_5 = n_q * n_time * n_pperp * n_ppara * n_z;
    data->step_4 = n_q * n_time * n_pperp * n_ppara;
    data->step_3 = n_q * n_time * n_pperp;
    data->step_2 = n_q * n_time;
    data->step_1 = n_q;
 
    data->histogram = calloc(data->step_6 * (size_t)data->n_r, sizeof(real));
    return data->histogram == NULL;
}

void dist_5D_free(dist_5D_data* data) {
    free(data->histogram);
}

void dist_5D_offload(dist_5D_data* data) {
    GPU_MAP_TO_DEVICE(
        data->histogram[0:data->n_r*data->n_phi*data->n_z*data->n_ppara*data->n_pperp*data->n_time*data->n_q]
    )
}

void dist_5D_onload(dist_5D_data* data) {
    GPU_UPDATE_FROM_DEVICE(
        data->histogram[0:data->n_r*data->n_phi*data->n_z*data->n_ppara*data->n_pperp*data->n_time*data->n_q]
        )
}

void dist_5D_update_fo(dist_5D_data* dist, particle_simd_fo* p_f,
                       particle_simd_fo* p_i) {
 
#ifdef GPU
    size_t index;
    real weight;
#else
    size_t index[NSIMD];
    real weight[NSIMD];
#endif
 
    GPU_PARALLEL_LOOP_ALL_LEVELS
    for(int i = 0; i < p_f->n_mrk; i++) {
        if(p_f->running[i]) {
            real i_r = floor((p_f->r[i] - dist->min_r)
                     / ((dist->max_r - dist->min_r)/dist->n_r));
 
            real phi = fmod(p_f->phi[i], 2*CONST_PI);
            if(phi < 0) {
                phi += 2*CONST_PI;
            }
            int i_phi = floor((phi - dist->min_phi)
                       / ((dist->max_phi - dist->min_phi)/dist->n_phi));
 
            int i_z = floor((p_f->z[i] - dist->min_z)
                     / ((dist->max_z - dist->min_z) / dist->n_z));
 
            real ppara = (  p_f->p_r[i]   * p_f->B_r[i]
                        + p_f->p_phi[i] * p_f->B_phi[i]
                        + p_f->p_z[i]   * p_f->B_z[i])
                       / sqrt(  p_f->B_r[i]  * p_f->B_r[i]
                              + p_f->B_phi[i]* p_f->B_phi[i]
                              + p_f->B_z[i]  * p_f->B_z[i]);
            int i_ppara = floor((ppara - dist->min_ppara)
                       / ((dist->max_ppara - dist->min_ppara) / dist->n_ppara));
 
            real pperp = sqrt(
                    p_f->p_r[i]   * p_f->p_r[i]
                  + p_f->p_phi[i] * p_f->p_phi[i]
                  + p_f->p_z[i]   * p_f->p_z[i]
                  - ppara * ppara);
            int i_pperp = floor((pperp - dist->min_pperp)
                         / ((dist->max_pperp - dist->min_pperp) / dist->n_pperp));
 
            int i_time = floor((p_f->time[i] - dist->min_time)
                         / ((dist->max_time - dist->min_time) / dist->n_time));
 
            int i_q = floor((p_f->charge[i]/CONST_E - dist->min_q)
                           / ((dist->max_q - dist->min_q) / dist->n_q));
 
            if(i_r     >= 0  &&  i_r     <= dist->n_r - 1      &&
               i_phi   >= 0  &&  i_phi   <= dist->n_phi - 1    &&
               i_z     >= 0  &&  i_z     <= dist->n_z - 1      &&
               i_ppara >= 0  &&  i_ppara <= dist->n_ppara - 1  &&
               i_pperp >= 0  &&  i_pperp <= dist->n_pperp - 1  &&
               i_time  >= 0  &&  i_time  <= dist->n_time - 1   &&
               i_q     >= 0  &&  i_q     <= dist->n_q - 1        ) {
#ifdef GPU
                index = dist_5D_index(
                    i_r, i_phi, i_z, i_ppara, i_pperp, i_time,
                    i_q, dist->step_6, dist->step_5, dist->step_4,
                    dist->step_3, dist->step_2, dist->step_1);
                weight = p_f->weight[i] * (p_f->time[i] - p_i->time[i]);
                GPU_ATOMIC
                dist->histogram[index] += weight;
#else
                index[i] = dist_5D_index(
                    i_r, i_phi, i_z, i_ppara, i_pperp, i_time,
                    i_q, dist->step_6, dist->step_5, dist->step_4,
                    dist->step_3, dist->step_2, dist->step_1);
                weight[i] = p_f->weight[i] * (p_f->time[i] - p_i->time[i]);
#endif
            }
        }
    }
 
#ifndef GPU
    for(int i = 0; i < p_f->n_mrk; i++) {
        if(p_f->running[i] && index[i] >= 0 &&
            index[i] < dist->step_6 * dist->n_r) {
            GPU_ATOMIC
            dist->histogram[index[i]] += weight[i];
        }
    }
#endif
}

void dist_5D_update_gc(dist_5D_data* dist, particle_simd_gc* p_f,
                       particle_simd_gc* p_i) {
    real phi[NSIMD];
    real pperp[NSIMD];
 
    int i_r[NSIMD];
    int i_phi[NSIMD];
    int i_z[NSIMD];
    int i_ppara[NSIMD];
    int i_pperp[NSIMD];
    int i_time[NSIMD];
    int i_q[NSIMD];
 
    int ok[NSIMD];
    real weight[NSIMD];
 
    #pragma omp simd
    for(int i = 0; i < NSIMD; i++) {
        if(p_f->running[i]) {
            i_r[i] = floor((p_f->r[i] - dist->min_r)
                     / ((dist->max_r - dist->min_r)/dist->n_r));
 
            phi[i] = fmod(p_f->phi[i], 2*CONST_PI);
            if(phi[i] < 0) {
                phi[i] = phi[i] + 2*CONST_PI;
            }
            i_phi[i] = floor((phi[i] - dist->min_phi)
                       / ((dist->max_phi - dist->min_phi)/dist->n_phi));
 
            i_z[i] = floor((p_f->z[i] - dist->min_z)
                    / ((dist->max_z - dist->min_z) / dist->n_z));
 
            i_ppara[i] = floor((p_f->ppar[i] - dist->min_ppara)
                       / ((dist->max_ppara - dist->min_ppara) / dist->n_ppara));
 
            pperp[i] = sqrt(2 * sqrt(  p_f->B_r[i]   * p_f->B_r[i]
                                     + p_f->B_phi[i] * p_f->B_phi[i]
                                     + p_f->B_z[i]   * p_f->B_z[i] )
                            * p_f->mu[i] * p_f->mass[i]);
            i_pperp[i] = floor((pperp[i] - dist->min_pperp)
                       / ((dist->max_pperp - dist->min_pperp) / dist->n_pperp));
 
            i_time[i] = floor((p_f->time[i] - dist->min_time)
                          / ((dist->max_time - dist->min_time) / dist->n_time));
 
            i_q[i] = floor((p_f->charge[i]/CONST_E - dist->min_q)
                           / ((dist->max_q - dist->min_q) / dist->n_q));
 
            if(i_r[i]     >= 0  &&  i_r[i]     <= dist->n_r - 1      &&
               i_phi[i]   >= 0  &&  i_phi[i]   <= dist->n_phi - 1    &&
               i_z[i]     >= 0  &&  i_z[i]     <= dist->n_z - 1      &&
               i_ppara[i] >= 0  &&  i_ppara[i] <= dist->n_ppara - 1  &&
               i_pperp[i] >= 0  &&  i_pperp[i] <= dist->n_pperp - 1  &&
               i_time[i]  >= 0  &&  i_time[i]  <= dist->n_time - 1   &&
               i_q[i]     >= 0  &&  i_q[i]     <= dist->n_q - 1        ) {
                ok[i] = 1;
                weight[i] = p_f->weight[i] * (p_f->time[i] - p_i->time[i]);
            }
            else {
                ok[i] = 0;
            }
        }
    }
 
    for(int i = 0; i < NSIMD; i++) {
        if(p_f->running[i] && ok[i]) {
            size_t index = dist_5D_index(
                i_r[i], i_phi[i], i_z[i], i_ppara[i], i_pperp[i], i_time[i],
                i_q[i], dist->step_6, dist->step_5, dist->step_4,
                dist->step_3, dist->step_2, dist->step_1);
            #pragma omp atomic
            dist->histogram[index] += weight[i];
        }
    }
}