ThreadProperties is a template class that provides each thread's properties within a workgroup. Please see the sycl-1.2.1 specification (https://www.khronos.org/registry/SYCL/specs/sycl-1.2.1.pdf) for the workgroup, work-items. More...

#include <TensorContractionSycl.h>

Public Member Functions
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE	ThreadProperties (const StorageIndex linearLocalThreadId_, const StorageIndex kGroupId_, const StorageIndex mGroupOffset_, const StorageIndex nGroupOffset_, const StorageIndex kGroupOffset_, const StorageIndex mLocalOffset_, const StorageIndex nLocalOffset_, const StorageIndex mGlobalOffset_, const StorageIndex nGlobalOffset_, StorageIndex kSize_, const bool is_internal_)

Public Attributes
const bool	is_internal

const StorageIndex	kGroupId

const StorageIndex	kGroupOffset

StorageIndex	kSize

const StorageIndex	linearLocalThreadId

const StorageIndex	mGlobalOffset

const StorageIndex	mGroupOffset

const StorageIndex	mLocalOffset

const StorageIndex	nGlobalOffset

const StorageIndex	nGroupOffset

const StorageIndex	nLocalOffset

Detailed Description

template<typename StorageIndex>
struct Eigen::TensorSycl::internal::ThreadProperties< StorageIndex >

ThreadProperties is a template class that provides each thread's properties within a workgroup. Please see the sycl-1.2.1 specification (https://www.khronos.org/registry/SYCL/specs/sycl-1.2.1.pdf) for the workgroup, work-items.

Template Parameters

StorageIndex determines the StorageIndex Type

Parameters

linearLocalThreadId	determines the linearized location of a thread within a work-group
kGroupId	determines the logical group id in a k dimension of the flattened tensor. It will be > 1 when tall/skinny algorithm is used
mGroupOffset	determines the logical start position of all thread within a workgroup for the m dimension of the flattened tensor.
kGroupOffset	determines the logical start position of all thread within a workgroup for the k dimension of the flattened tensor. It will be > 1 when tall/skinny algorithm is used.
mLocalOffset	determines the logical start position of each thread within a workgroup for the m dimension of a flattened tensor. The position determines the distance of each thread within the workgroup from each other independent from their global position.
nLocalOffset	determines the logical start position of each thread within a workgroup for the n dimension of a flattened tensor. The position determines the distance of each thread within the workgroup from each other independent from their global position.
mGlobalOffset	determines the logical start position of each thread a thread for the m dimension on a flattened tensor
nGlobalOffset	determines the logical start position of each thread a thread for the n dimension on a flattened tensor
kSize	: determine the number of the k elements of the flattened Tensor to be processed by each thread for the given tensor block. This is !=K dimension of Flattened Tensor when Tall/Skinny matrix is used.
is_internal	: this will determined if the thread within the work-group computes an internal block of tensor or the edge blocks. When it is internal, there is no need to check the boundaries and all the if stantement can be resolve by compiler.

Definition at line 369 of file TensorContractionSycl.h.