TNL::Algorithms::Segments::LaunchConfiguration Struct Reference

Launch configuration for segment operations. More...

#include <TNL/Algorithms/Segments/LaunchConfiguration.h>

Inheritance diagram for TNL::Algorithms::Segments::LaunchConfiguration:

Collaboration diagram for TNL::Algorithms::Segments::LaunchConfiguration:

Public Member Functions
	LaunchConfiguration ()=default
	Constructs a configuration with the Default mapping.
	LaunchConfiguration (ThreadsToSegmentsMapping threadsToSegmentsMapping, int threadsPerSegmentCount, int blockSize=1)
	Constructs a configuration with the given mapping and threads per segment count.
const Backend::LaunchConfiguration &	getBackendLaunchConfiguration () const
	Returns the backend part of the launch configuration with the grid and block sizes.
int	getThreadsPerSegmentCount () const
	Returns the number of threads per segment.
ThreadsToSegmentsMapping	getThreadsToSegmentsMapping () const
	Returns the mapping of threads to segments.
void	setThreadsPerSegmentCount (int threadsPerSegmentCount)
	Sets the number of threads per segment.
void	setThreadsPerSegmentCountToWarpSize ()
	Sets the number of threads per segment to the warp size of the current GPU.
void	setThreadsToSegmentsMapping (ThreadsToSegmentsMapping threadsToSegmentsMapping)
	Sets the mapping of threads to segments.
Public Member Functions inherited from TNL::Backend::LaunchConfiguration
constexpr	LaunchConfiguration (const LaunchConfiguration &)=default
constexpr	LaunchConfiguration (dim3 gridSize, dim3 blockSize, std::size_t dynamicSharedMemorySize=0U, Backend::stream_t stream=0, bool blockHostUntilFinished=true)
constexpr	LaunchConfiguration (LaunchConfiguration &&)=default

Protected Attributes
int	threadsPerSegmentCount = 1
ThreadsToSegmentsMapping	threadsToSegmentsMapping = ThreadsToSegmentsMapping::Default

Additional Inherited Members
Public Attributes inherited from TNL::Backend::LaunchConfiguration
bool	blockHostUntilFinished = true
	indicates whether host execution is blocked until the kernel execution is finished
dim3	blockSize
	kernel block dimensions (in threads)
std::size_t	dynamicSharedMemorySize = 0U
	size of dynamic shared memory (in bytes per block)
dim3	gridSize
	kernel grid dimensions (in blocks)
Backend::stream_t	stream = 0
	stream handle

Detailed Description

Launch configuration for segment operations.

This class encapsulates the configuration for launching segment operations, including the mapping of threads to segments and the number of threads per segment. The default configuration requests the Default mapping, i.e. the best native strategy of the segments format. A segments format which does not support the requested mapping or threads per segment count throws std::invalid_argument.

Constructor & Destructor Documentation

LaunchConfiguration()

TNL::Algorithms::Segments::LaunchConfiguration::LaunchConfiguration ( ThreadsToSegmentsMapping threadsToSegmentsMapping, int threadsPerSegmentCount, int blockSize = 1 )

inline

Constructs a configuration with the given mapping and threads per segment count.

Parameters

threadsToSegmentsMapping	The mapping of threads to segments.
threadsPerSegmentCount	The number of threads per segment, which must be a positive power of two.
blockSize	The block size of the backend launch configuration.

Member Function Documentation

setThreadsPerSegmentCount()

void TNL::Algorithms::Segments::LaunchConfiguration::setThreadsPerSegmentCount ( int threadsPerSegmentCount )

inline

Sets the number of threads per segment.

Its meaning depends on the mapping, see ThreadsToSegmentsMapping.

Parameters

threadsPerSegmentCount

The number of threads per segment.

Exceptions

std::invalid_argument

if threadsPerSegmentCount is not a positive power of two. The current count is not changed in that case.

---

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

• src/TNL/Algorithms/Segments/LaunchConfiguration.h