Reference documentation for deal.II version Git 953c9590e9 2020-10-28 17:25:14 -0400
\(\newcommand{\dealvcentcolon}{\mathrel{\mathop{:}}}\) \(\newcommand{\dealcoloneq}{\dealvcentcolon\mathrel{\mkern-1.2mu}=}\) \(\newcommand{\jump}[1]{\left[\!\left[ #1 \right]\!\right]}\) \(\newcommand{\average}[1]{\left\{\!\left\{ #1 \right\}\!\right\}}\)
Namespaces | Classes | Functions | Variables
CUDAWrappers Namespace Reference

Namespaces

 internal
 

Classes

class  FEEvaluation
 
class  MatrixFree
 
class  PreconditionIC
 
class  PreconditionILU
 
struct  SharedData
 
class  SolverDirect
 
class  SparseMatrix
 

Functions

__host__ constexpr unsigned int cells_per_block_shmem (int dim, int fe_degree)
 

Variables

constexpr int block_size = 512
 
constexpr int chunk_size = 1
 
constexpr int warp_size = 32
 
constexpr unsigned int mf_max_elem_degree = 10
 
constexpr unsigned int mf_n_concurrent_objects = 5
 

Detailed Description

Namespace for the CUDA wrappers

Function Documentation

◆ cells_per_block_shmem()

__host__ constexpr unsigned int CUDAWrappers::cells_per_block_shmem ( int  dim,
int  fe_degree 
)

Definition at line 662 of file cuda_matrix_free.h.

Variable Documentation

◆ block_size

constexpr int CUDAWrappers::block_size = 512

Define the size of a block when launching a CUDA kernel. This number can be changed depending on the architecture the code is running on.

Definition at line 29 of file cuda_size.h.

◆ chunk_size

constexpr int CUDAWrappers::chunk_size = 1

Define the size of chunk of data worked on by a thread. This number can be changed depending on the architecture the code is running on.

Definition at line 35 of file cuda_size.h.

◆ warp_size

constexpr int CUDAWrappers::warp_size = 32

Define the number of threads in a warp.

Definition at line 40 of file cuda_size.h.

◆ mf_max_elem_degree

constexpr unsigned int CUDAWrappers::mf_max_elem_degree = 10

Define the largest finite element degree that can be solved using CUDAWrappers::MatrixFree. Changing this number will affect the amount of constant memory being used.

Definition at line 47 of file cuda_size.h.

◆ mf_n_concurrent_objects

constexpr unsigned int CUDAWrappers::mf_n_concurrent_objects = 5

Define the maximum number of valid CUDAWrappers::MatrixFree object. Changing this number will affect the amount of constant memory being used.

Definition at line 53 of file cuda_size.h.