viennacl::generator Namespace Reference

Provides an OpenCL kernel generator. More...

Namespaces
	autotune
	Provides the implementation for tuning the kernels for a particular device.
	detail
	Contains implementation details of the kernel generator.
	profiles
	Namespace holding the various device-specific parameters for generating the best kernels.
	utils
	Contains various helper routines for kernel generation.

Data Structures
struct	expression_descriptor
	A class for holding meta information such as the type or the underlying scalar type of an expression (such as x = inner_prod(y, z)). More...
class	code_generator
	Class for handling code generation. More...
class	matrix_product
	Kernel generation class for matrix-matrix products. More...
class	profile_base
	Base class for an operation profile. More...
class	vector_saxpy
	OpenCL kernel generation class for vector expressions of AXPY type, i.e. x = alpha * y + beta * z, where the number of summands can in principle be arbitrarily large. More...
class	matrix_saxpy
	OpenCL kernel generation class for matrix expressions of AXPY type, i.e. A = alpha * B + beta * C, where the number of summands can in principle be arbitrarily large. More...
class	scalar_reduction
	OpenCL kernel generation template for scalar reduction operations such as s = norm_2(x). More...
class	vector_reduction
	OpenCL kernel template for reductions resulting in a vector. Example: Computing the row norms of a matrix concurrently. More...

Typedefs
typedef std::pair < expression_type, vcl_size_t >	expression_key_type

Enumerations
enum	expression_type_family {   SCALAR_SAXPY_FAMILY, VECTOR_SAXPY_FAMILY, MATRIX_SAXPY_FAMILY, SCALAR_REDUCE_FAMILY,   VECTOR_REDUCE_FAMILY, MATRIX_PRODUCT_FAMILY, INVALID_EXPRESSION_FAMILY }
enum	expression_type {   SCALAR_SAXPY_TYPE, VECTOR_SAXPY_TYPE, MATRIX_SAXPY_TYPE, SCALAR_REDUCE_TYPE,   VECTOR_REDUCE_Nx_TYPE, VECTOR_REDUCE_Tx_TYPE, MATRIX_PRODUCT_NN_TYPE, MATRIX_PRODUCT_TN_TYPE,   MATRIX_PRODUCT_NT_TYPE, MATRIX_PRODUCT_TT_TYPE, INVALID_EXPRESSION_TYPE }

Functions
void	generate_enqueue_statement (viennacl::scheduler::statement const &s, scheduler::statement_node const &root_node)
	Generate and enqueue a statement plus root_node into the current queue. More...
void	generate_enqueue_statement (viennacl::scheduler::statement const &s)
	Generate and enqueue a statement into the current queue, assumes the root_node is the first node of the statement. More...
const char *	expression_type_to_string (expression_type type)
template<typename KeyT , typename ValueT >
ValueT const &	at (std::map< KeyT, ValueT > const &map, KeyT const &key)
	Emulation of C++11's .at() member for std::map<> More...
viennacl::ocl::program &	get_configured_program (viennacl::generator::code_generator const &generator, std::list< viennacl::ocl::kernel * > &kernels, bool force_recompilation=false)
	Creates the program associated with a generator object and fills the kernels. Checks the context for the program and possibly (re)compile it. More...
void	enqueue (viennacl::generator::code_generator const &generator, bool force_recompilation=false)
	Set the arguments and enqueue a generator object. More...
std::string	get_opencl_program_string (viennacl::scheduler::statement const &s)
	Convenience function to get the OpenCL program string for a single statement. More...
std::string	get_cuda_device_code (viennacl::scheduler::statement const &s)
	Convenience function to get the CUDA device code for a single statement. More...
std::ostream &	operator<< (std::ostream &os, profile_base const &profile)
void	enqueue_custom_op (viennacl::generator::custom_operation &op, viennacl::ocl::command_queue const &queue)

Detailed Description

Provides an OpenCL kernel generator.

Typedef Documentation

typedef std::pair<expression_type, vcl_size_t> expression_key_type

Enumeration Type Documentation

enum expression_type

Enumerator
SCALAR_SAXPY_TYPE
VECTOR_SAXPY_TYPE
MATRIX_SAXPY_TYPE
SCALAR_REDUCE_TYPE
VECTOR_REDUCE_Nx_TYPE
VECTOR_REDUCE_Tx_TYPE
MATRIX_PRODUCT_NN_TYPE
MATRIX_PRODUCT_TN_TYPE
MATRIX_PRODUCT_NT_TYPE
MATRIX_PRODUCT_TT_TYPE
INVALID_EXPRESSION_TYPE

enum expression_type_family

Enumerator
SCALAR_SAXPY_FAMILY
VECTOR_SAXPY_FAMILY
MATRIX_SAXPY_FAMILY
SCALAR_REDUCE_FAMILY
VECTOR_REDUCE_FAMILY
MATRIX_PRODUCT_FAMILY
INVALID_EXPRESSION_FAMILY

Function Documentation

ValueT const& viennacl::generator::at	(	std::map< KeyT, ValueT > const &	map,
		KeyT const &	key
	)

Emulation of C++11's .at() member for std::map<>

void viennacl::generator::enqueue ( viennacl::generator::code_generator const &  generator, bool  force_recompilation = false  )

inline

Set the arguments and enqueue a generator object.

void viennacl::generator::enqueue_custom_op	(	viennacl::generator::custom_operation &	op,
		viennacl::ocl::command_queue const &	queue
	)

const char* viennacl::generator::expression_type_to_string ( expression_type  type )

inline

void generate_enqueue_statement ( viennacl::scheduler::statement const &  s, scheduler::statement_node const &  root_node  )

inline

Generate and enqueue a statement plus root_node into the current queue.

void generate_enqueue_statement ( viennacl::scheduler::statement const &  s )

inline

Generate and enqueue a statement into the current queue, assumes the root_node is the first node of the statement.

viennacl::ocl::program& viennacl::generator::get_configured_program ( viennacl::generator::code_generator const &  generator, std::list< viennacl::ocl::kernel * > &  kernels, bool  force_recompilation = false  )

inline

Creates the program associated with a generator object and fills the kernels. Checks the context for the program and possibly (re)compile it.

Parameters

generator	the generator to work on
kernels	this list will be filled with the kernels associated with the generator
force_recompilation	if true, the program will be recompiled

std::string viennacl::generator::get_cuda_device_code ( viennacl::scheduler::statement const &  s )

inline

Convenience function to get the CUDA device code for a single statement.

std::string viennacl::generator::get_opencl_program_string ( viennacl::scheduler::statement const &  s )

inline

Convenience function to get the OpenCL program string for a single statement.

std::ostream& viennacl::generator::operator<< ( std::ostream &  os, profile_base const &  profile  )

inline