mirror of
https://gitlab.com/libeigen/eigen.git
synced 2026-04-10 11:34:33 +08:00
fix various typos
This commit is contained in:
committed by
Antonio Sánchez
parent
3753e6a2b3
commit
4b6036e276
@@ -522,10 +522,10 @@ class TensorContractionKernel {
|
||||
* \param rhs_scratch_extract : determines the RHS tile memory. It is either private or local memory based on the
|
||||
* selected contraction_type.
|
||||
*
|
||||
* \param lhs_extract_index: determins the position of each thread on a local memory for lhs input. When private
|
||||
* \param lhs_extract_index: determines the position of each thread on a local memory for lhs input. When private
|
||||
* memory is used this is set to zero as this is not applicable in case of private memory.
|
||||
*
|
||||
* \param rhs_extract_index: determins the position of each thread on a local memory for rhs input. When private
|
||||
* \param rhs_extract_index: determines the position of each thread on a local memory for rhs input. When private
|
||||
* memory is used this is set to zero as this is not applicable in case of private memory.
|
||||
*
|
||||
* \param lhs_scratch_compute : determines the location to load for computation for lhs_local memory. This is the
|
||||
@@ -1236,7 +1236,7 @@ struct GeneralVectorTensor {
|
||||
*
|
||||
* \param out_res: determines the output tensor containing the contraction result
|
||||
*
|
||||
* \param rng: determins the total input data size
|
||||
* \param rng: determines the total input data size
|
||||
*/
|
||||
template <typename OutScalar, typename LhsScalar, typename RhsScalar, typename OutAccessor, typename LhsMapper,
|
||||
typename RhsMapper, typename StorageIndex, bool Vectorizable>
|
||||
|
||||
@@ -98,7 +98,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
|
||||
// context from the heap.
|
||||
//
|
||||
// (*) EvalParallelContext & EvalShardedByInnerDimContext owns all the state
|
||||
// and temporary buffers, requried for executing the tensor contraction.
|
||||
// and temporary buffers, required for executing the tensor contraction.
|
||||
// They are responsible for cleaning it up after contraction is done.
|
||||
static const bool IsEvalInSyncMode =
|
||||
std::is_same<DoneCallback, NoCallback>::value;
|
||||
@@ -700,7 +700,7 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
|
||||
!is_rhs && std::is_same<BlockType, LhsBlock>::value;
|
||||
static const bool kIsRhs =
|
||||
is_rhs && std::is_same<BlockType, RhsBlock>::value;
|
||||
static_assert(kIsLhs || kIsRhs, "Unkown block type");
|
||||
static_assert(kIsLhs || kIsRhs, "Unknown block type");
|
||||
|
||||
using Blocks = ThreadLocalBlocks<BlockType>;
|
||||
|
||||
|
||||
@@ -246,7 +246,7 @@ class QueueInterface {
|
||||
}
|
||||
|
||||
/// The memcpyHostToDevice is used to copy the data from host to device
|
||||
/// The destination pointer could be deleted before the copy happend which is
|
||||
/// The destination pointer could be deleted before the copy happened which is
|
||||
/// why a callback function is needed. By default if none is provided, the
|
||||
/// function is blocking.
|
||||
EIGEN_STRONG_INLINE void memcpyHostToDevice(
|
||||
@@ -274,7 +274,7 @@ class QueueInterface {
|
||||
}
|
||||
|
||||
/// The memcpyDeviceToHost is used to copy the data from device to host.
|
||||
/// The source pointer could be deleted before the copy happend which is
|
||||
/// The source pointer could be deleted before the copy happened which is
|
||||
/// why a callback function is needed. By default if none is provided, the
|
||||
/// function is blocking.
|
||||
EIGEN_STRONG_INLINE void memcpyDeviceToHost(
|
||||
|
||||
@@ -31,7 +31,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T* constCast(const T* data) {
|
||||
}
|
||||
|
||||
// The StorageMemory class is a container of the device specific pointer
|
||||
// used for refering to a Pointer on TensorEvaluator class. While the TensorExpression
|
||||
// used for referring to a Pointer on TensorEvaluator class. While the TensorExpression
|
||||
// is a device-agnostic type and need MakePointer class for type conversion,
|
||||
// the TensorEvaluator class can be specialized for a device, hence it is possible
|
||||
// to construct different types of temproray storage memory in TensorEvaluator
|
||||
|
||||
@@ -142,7 +142,7 @@ template <typename T> class UniformRandomGenerator {
|
||||
// but for SYCL ((CLOCK * 6364136223846793005ULL) + 0xda3e39cb94b95bdbULL) is passed to each thread and each thread adds
|
||||
// the (global_thread_id* 6364136223846793005ULL) for itself only once, in order to complete the construction
|
||||
// similar to CUDA Therefore, the thread Id injection is not available at this stage.
|
||||
//However when the operator() is called the thread ID will be avilable. So inside the opeator,
|
||||
//However when the operator() is called the thread ID will be available. So inside the opeator,
|
||||
// we add the thrreadID, BlockId,... (which is equivalent of i)
|
||||
//to the seed and construct the unique m_state per thead similar to cuda.
|
||||
m_exec_once =false;
|
||||
@@ -252,7 +252,7 @@ template <typename T> class NormalRandomGenerator {
|
||||
//the same for all the thread. As unlike CUDA, the thread.ID, BlockID, etc is not a global function.
|
||||
// and only available on the Operator() function (which is called on the GPU).
|
||||
// Therefore, the thread Id injection is not available at this stage. However when the operator()
|
||||
//is called the thread ID will be avilable. So inside the opeator,
|
||||
//is called the thread ID will be available. So inside the operator,
|
||||
// we add the thrreadID, BlockId,... (which is equivalent of i)
|
||||
//to the seed and construct the unique m_state per thead similar to cuda.
|
||||
m_exec_once =false;
|
||||
|
||||
Reference in New Issue
Block a user