yakl::LaunchConfig< VL, B4B > Struct Template Reference

This class informs YAKL parallel_for and parallel_outer routines how to launch kernels. More...

#include <YAKL_LaunchConfig.h>

Public Member Functions
	LaunchConfig ()
	set_inner_size() defaults to YAKL_DEFAULT_VECTOR_LEN More...
	LaunchConfig (LaunchConfig &&rhs)
	LaunchConfig objects may be copied or moved. More...
	LaunchConfig (LaunchConfig const &rhs)
	LaunchConfig objects may be copied or moved. More...
	~LaunchConfig ()
void	copyfrom (LaunchConfig const &rhs)
int	get_inner_size () const
	Get the inner loop size for hierarchical parallelism. More...
Stream	get_stream () const
	Get the stream in which this launch will run. More...
LaunchConfig &	operator= (LaunchConfig &&rhs)
	LaunchConfig objects may be copied or moved. More...
LaunchConfig &	operator= (LaunchConfig const &rhs)
	LaunchConfig objects may be copied or moved. More...
LaunchConfig	set_inner_size (int num)
	This sets the actual inner looping size whereas the template parameter VL sets the maximum inner looping size. More...
LaunchConfig	set_stream (Stream stream)
	Set the stream in which this launch will run. More...

Detailed Description

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>
struct yakl::LaunchConfig< VL, B4B >

This class informs YAKL parallel_for and parallel_outer routines how to launch kernels.

It contains two optional template parameters: (1) VL: When passed to parallel_for, this defines the inner looping size on the device (e.g. "block size" for CUDA and HIP. When passed to parallel_outer, this defines the maximum inner looping size on the device. (2) B4B: If this is set to true, then this tells parallel_for and parallel_outer to run the kernel serially (only when the -DYAKL_B4B CPP macro is defined) to enable bitwise determinism when desired for kernels with yakl::atomicAdd in them.

Parameters

VL	For parallel_for, this is the inner looping size. For parallel_outer, this is the maximum inner looping size.
B4B	If the CPP macro YAKL_B4B is also defined, B4B == true will force the kernel to run in serial, typically used for kernels that contain yakl::atomicAdd to maintain bitwise determinism run-to-run. If YAKL_B4B is not defined, the kernel runs normally.

Constructor & Destructor Documentation

LaunchConfig() [1/3]

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

yakl::LaunchConfig< VL, B4B >::LaunchConfig ( )

inline

set_inner_size() defaults to YAKL_DEFAULT_VECTOR_LEN

~LaunchConfig()

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

yakl::LaunchConfig< VL, B4B >::~LaunchConfig ( )

inline

LaunchConfig() [2/3]

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

yakl::LaunchConfig< VL, B4B >::LaunchConfig ( LaunchConfig< VL, B4B > const &  rhs )

inline

LaunchConfig objects may be copied or moved.

LaunchConfig() [3/3]

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

yakl::LaunchConfig< VL, B4B >::LaunchConfig ( LaunchConfig< VL, B4B > &&  rhs )

inline

LaunchConfig objects may be copied or moved.

Member Function Documentation

copyfrom()

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

void yakl::LaunchConfig< VL, B4B >::copyfrom ( LaunchConfig< VL, B4B > const &  rhs )

inline

get_inner_size()

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

int yakl::LaunchConfig< VL, B4B >::get_inner_size ( ) const

inline

Get the inner loop size for hierarchical parallelism.

get_stream()

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

Stream yakl::LaunchConfig< VL, B4B >::get_stream ( ) const

inline

Get the stream in which this launch will run.

operator=() [1/2]

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

LaunchConfig& yakl::LaunchConfig< VL, B4B >::operator= ( LaunchConfig< VL, B4B > &&  rhs )

inline

LaunchConfig objects may be copied or moved.

operator=() [2/2]

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

LaunchConfig& yakl::LaunchConfig< VL, B4B >::operator= ( LaunchConfig< VL, B4B > const &  rhs )

inline

LaunchConfig objects may be copied or moved.

set_inner_size()

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

LaunchConfig yakl::LaunchConfig< VL, B4B >::set_inner_size ( int  num )

inline

This sets the actual inner looping size whereas the template parameter VL sets the maximum inner looping size.

set_stream()

template<int VL = YAKL_DEFAULT_VECTOR_LEN, bool B4B = false>

LaunchConfig yakl::LaunchConfig< VL, B4B >::set_stream ( Stream  stream )

inline

Set the stream in which this launch will run.

---

The documentation for this struct was generated from the following file:

• YAKL/src/YAKL_LaunchConfig.h