(git:27c966b)
Loading...
Searching...
No Matches
grid_gpu_collocate.cu File Reference
#include <algorithm>
#include <assert.h>
#include <climits>
#include <cmath>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include "grid_gpu_internal_header.h"
#include "grid_gpu_prepare_pab.h"
#include "grid_gpu_separable.h"

Go to the source code of this file.

Namespaces

namespace  grid_gpu_backend

Functions

template<typename T, bool IS_FUNC_AB>
__device__ __inline__ void grid_gpu_backend::block_to_cab (const kernel_params &params, const smem_task< T > &task, T *cab)
 Decontracts the subblock, going from spherical to cartesian harmonics.
template<typename T, bool IS_FUNC_AB>
__global__ grid_gpu_backend::__launch_bounds__ (64) void calculate_coefficients_two_steps(const kernel_params dev_
 Same as calculate_coefficients, with the spherical to cartesian transformation of block_to_cab done in two steps,.
 grid_gpu_backend::if (number_of_tasks==0) return
 grid_gpu_backend::for (int tk=0;tk< number_of_tasks;tk++)
template<bool distributed__, bool axis_aligned_>
__global__ grid_gpu_backend::__launch_bounds__ (64) void collocate_separable_kernel(const kernel_params dev_
 grid_gpu_backend::if (glb_task.skip_task) return
 grid_gpu_backend::fill_smem_task_reduced (dev_, dev_.first_task+block_index(), task)
 grid_gpu_backend::if (tid==0)
 grid_gpu_backend::__syncthreads ()
 grid_gpu_backend::if (axis_aligned_)
static size_t grid_gpu_backend::collocate_separable_smem_size (const int mc, const int lp_max, const int erows)
static int grid_gpu_backend::collocate_separable_erows (const int mc, const int lp_max, const size_t smem_max)

Variables

__global__ const int grid_gpu_backend::ld
__global__ const int const int grid_gpu_backend::max_nsgf_set
const int grid_gpu_backend::number_of_tasks = dev_.num_tasks_per_block_dev[block_index()]
T * grid_gpu_backend::smem_t = &shared_memory[0]
T * grid_gpu_backend::smem_p = smem_t + max_nsgf_set * ld
const int grid_gpu_backend::tid = thread_global_index()
const int grid_gpu_backend::nthreads = blockDim.x * blockDim.y * blockDim.z
const int grid_gpu_backend::offset = dev_.sorted_blocks_offset_dev[block_index()]
T * grid_gpu_backend::smem_cab
__global__ const int grid_gpu_backend::mc
__global__ const int const int grid_gpu_backend::lp_max
__global__ const int const int const int grid_gpu_backend::erows
__global__ const int const int const int const double *__restrict__ grid_gpu_backend::M
const task_info & grid_gpu_backend::glb_task = dev_.tasks[dev_.first_task + block_index()]
const int grid_gpu_backend::ls = lp_max + 1
__shared__ double grid_gpu_backend::sep_smem []
double * grid_gpu_backend::ex = sep_smem
double * grid_gpu_backend::ey = ex + mc
double * grid_gpu_backend::D = ey + mc
double * grid_gpu_backend::C = D + ls * ls
double * grid_gpu_backend::E = C + ncoset(lp_max)
__shared__ smem_task_reduced< double, double3 > grid_gpu_backend::task
const int grid_gpu_backend::lp = task.lp
const int grid_gpu_backend::ncoef = ncoset(lp)
const double grid_gpu_backend::zetp = task.zetp
const double grid_gpu_backend::r2max = task.radius * task.radius
const bool grid_gpu_backend::trim = orthogonal_ && !task.apply_border_mask
const double *__restrict__ grid_gpu_backend::coef
const int grid_gpu_backend::npairs = (lp + 1) * (lp + 2) / 2
const int grid_gpu_backend::row_group = tid / GRID_SEP_COLLOCATE_ROW_LANES
const int grid_gpu_backend::lane = tid % GRID_SEP_COLLOCATE_ROW_LANES
constexpr int grid_gpu_backend::nrow_groups = 64 / GRID_SEP_COLLOCATE_ROW_LANES