o2::its::gpu Namespace Reference

Classes
struct	gpuSpan
struct	gpuSpan< const T >
class	GPUTimer
struct	KernelOccupancy
	Minimum resident blocks per compute unit to request when no per-kernel measurement exists. More...
class	Stream
class	Streams
class	TimeFrameGPU

Functions
constexpr KernelOccupancy	uniformOccupancy (int minBlocks)
	Use the same occupancy floor for every kernel when no per-kernel measurements are available.
constexpr int	gridBlocks (int residentBlocksPerComputeUnit)
	Number of blocks in a grid whose depth is residentBlocksPerComputeUnit blocks per CU.
constexpr int	gridThreads (int residentBlocksPerComputeUnit)
	Number of threads covered by a grid whose depth is residentBlocksPerComputeUnit blocks per CU.
constexpr bool	residentCoversFloor ()
	The grid must provide at least as many blocks per CU as the corresponding occupancy floor.

Variables
constexpr int	ComputeUnits = 60
constexpr int	WarpSize = 64
constexpr int	GPUThreads = 256
constexpr int	DefaultBlocksPerComputeUnit = 4
constexpr int	MaxBlocksPerComputeUnit = 10
constexpr KernelOccupancy	MinBlocks = uniformOccupancy(1)
	NVIDIA: unmeasured.
constexpr KernelOccupancy	ResidentBlocks = uniformOccupancy(DefaultBlocksPerComputeUnit)

Function Documentation

gridBlocks()

constexpr int o2::its::gpu::gridBlocks ( int  residentBlocksPerComputeUnit )

constexpr

Number of blocks in a grid whose depth is residentBlocksPerComputeUnit blocks per CU.

Definition at line 150 of file LaunchGeometry.h.

gridThreads()

constexpr int o2::its::gpu::gridThreads ( int  residentBlocksPerComputeUnit )

constexpr

Number of threads covered by a grid whose depth is residentBlocksPerComputeUnit blocks per CU.

Definition at line 156 of file LaunchGeometry.h.

residentCoversFloor()

constexpr bool o2::its::gpu::residentCoversFloor ( )

constexpr

The grid must provide at least as many blocks per CU as the corresponding occupancy floor.

Definition at line 174 of file LaunchGeometry.h.

uniformOccupancy()

constexpr KernelOccupancy o2::its::gpu::uniformOccupancy ( int  minBlocks )

constexpr

Use the same occupancy floor for every kernel when no per-kernel measurements are available.

Definition at line 101 of file LaunchGeometry.h.

Variable Documentation

ComputeUnits

constexpr int o2::its::gpu::ComputeUnits = 60

constexpr

Definition at line 56 of file LaunchGeometry.h.

DefaultBlocksPerComputeUnit

constexpr int o2::its::gpu::DefaultBlocksPerComputeUnit = 4

constexpr

Definition at line 61 of file LaunchGeometry.h.

GPUThreads

constexpr int o2::its::gpu::GPUThreads = 256

constexpr

Definition at line 60 of file LaunchGeometry.h.

MaxBlocksPerComputeUnit

constexpr int o2::its::gpu::MaxBlocksPerComputeUnit = 10

constexpr

Definition at line 62 of file LaunchGeometry.h.

MinBlocks

constexpr KernelOccupancy o2::its::gpu::MinBlocks = uniformOccupancy(1)

constexpr

NVIDIA: unmeasured.

Definition at line 145 of file LaunchGeometry.h.

ResidentBlocks

constexpr KernelOccupancy o2::its::gpu::ResidentBlocks = uniformOccupancy(DefaultBlocksPerComputeUnit)

constexpr

Definition at line 146 of file LaunchGeometry.h.

WarpSize

constexpr int o2::its::gpu::WarpSize = 64

constexpr

Definition at line 57 of file LaunchGeometry.h.