void op_par_loop_bres_calc(char const *name, op_set set,
op_arg arg0,
op_arg arg2,
op_arg arg3,
op_arg arg4,
op_arg arg5 ){
int nargs = 6;
op_arg args[6];
arg0.idx = 0;
args[0] = arg0;
for (int v = 1; v < 2; v++) {
args[0 + v] = op_arg_dat(arg0.dat, v, arg0.map, 2, "float", OP_READ);
}
args[2] = arg2;
args[3] = arg3;
args[4] = arg4;
args[5] = arg5;
int ninds = 4;
int inds[6] = {0,0,1,2,3,-1};
if (OP_diags>2) {
printf(" kernel routine with indirection: bres_calc\n");
}
// get plan
#ifdef OP_PART_SIZE_3
int part_size = OP_PART_SIZE_3;
#else
int part_size = OP_part_size;
#endif
int set_size = op_mpi_halo_exchanges(set, nargs, args);
// initialise timers
double cpu_t1, cpu_t2, wall_t1=0, wall_t2=0;
op_timing_realloc(3);
OP_kernels[3].name = name;
OP_kernels[3].count += 1;
if (set->size >0) {
op_plan *Plan = op_plan_get(name,set,part_size,nargs,args,ninds,inds);
op_timers_core(&cpu_t1, &wall_t1);
// execute plan
int block_offset = 0;
for (int col=0; col < Plan->ncolors; col++) {
if (col==Plan->ncolors_core) op_mpi_wait_all(nargs, args);
int nblocks = Plan->ncolblk[col];
#pragma omp parallel for
for (int blockIdx=0; blockIdx<nblocks; blockIdx++)
op_x86_bres_calc( blockIdx,
(float *)arg0.data,
(float *)arg2.data,
(float *)arg3.data,
(float *)arg4.data,
Plan->ind_map,
Plan->loc_map,
(int *)arg5.data,
Plan->ind_sizes,
Plan->ind_offs,
block_offset,
Plan->blkmap,
Plan->offset,
Plan->nelems,
Plan->nthrcol,
Plan->thrcol,
set_size);
block_offset += nblocks;
}
op_timing_realloc(3);
OP_kernels[3].transfer += Plan->transfer;
OP_kernels[3].transfer2 += Plan->transfer2;
}
// combine reduction data
op_mpi_set_dirtybit(nargs, args);
// update kernel record
op_timers_core(&cpu_t2, &wall_t2);
OP_kernels[3].time += wall_t2 - wall_t1;
}
void op_par_loop_bres_calc(char const *name, op_set set,
op_arg arg0,
op_arg arg1,
op_arg arg2,
op_arg arg3,
op_arg arg4,
op_arg arg5 ){
int nargs = 6;
op_arg args[6] = {arg0,arg1,arg2,arg3,arg4,arg5};
int ninds = 4;
int inds[6] = {0,0,1,2,3,-1};
if (OP_diags>2) {
printf(" kernel routine with indirection: bres_calc \n");
}
// get plan
#ifdef OP_PART_SIZE_3
int part_size = OP_PART_SIZE_3;
#else
int part_size = OP_part_size;
#endif
op_plan *Plan = op_plan_get(name,set,part_size,nargs,args,ninds,inds);
// initialise timers
double cpu_t1, cpu_t2, wall_t1, wall_t2;
op_timers(&cpu_t1, &wall_t1);
// set number of threads
#ifdef _OPENMP
int nthreads = omp_get_max_threads( );
#else
int nthreads = 1;
#endif
// execute plan
int block_offset = 0;
for (int col=0; col < Plan->ncolors; col++) {
int nblocks = Plan->ncolblk[col];
#pragma omp parallel for
for (int blockIdx=0; blockIdx<nblocks; blockIdx++)
op_x86_bres_calc( blockIdx,
(float *)arg0.data, Plan->ind_maps[0],
(float *)arg2.data, Plan->ind_maps[1],
(float *)arg3.data, Plan->ind_maps[2],
(float *)arg4.data, Plan->ind_maps[3],
Plan->loc_maps[0],
Plan->loc_maps[1],
Plan->loc_maps[2],
Plan->loc_maps[3],
Plan->loc_maps[4],
(int *)arg5.data,
Plan->ind_sizes,
Plan->ind_offs,
block_offset,
Plan->blkmap,
Plan->offset,
Plan->nelems,
Plan->nthrcol,
Plan->thrcol);
block_offset += nblocks;
}
// update kernel record
op_timers(&cpu_t2, &wall_t2);
op_timing_realloc(3);
OP_kernels[3].name = name;
OP_kernels[3].count += 1;
OP_kernels[3].time += wall_t2 - wall_t1;
OP_kernels[3].transfer += Plan->transfer;
OP_kernels[3].transfer2 += Plan->transfer2;
}
