mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-21 07:19:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Explicitly construct tensor block dimensions from evaluator dimensions

Browse Source
This commit is contained in:
Eugene Zhulenev 2018-09-14 16:55:05 -07:00
parent 281e631839
commit 4863375723
2 changed files with 5 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h
View File

@ -282,8 +282,9 @@ class TensorExecutor<Expression, ThreadPoolDevice, Vectorizable, /*Tileable*/ tr
TensorOpCost cost = evaluator.costPerCoeff(Vectorizable);
double taskSize = TensorCostModel<ThreadPoolDevice>::taskSize(1, cost);
size_t block_size = static_cast<size_t>(1.0 / taskSize);
TensorBlockMapper block_mapper(evaluator.dimensions(), block_shape,
block_size);
TensorBlockMapper block_mapper(
typename TensorBlockMapper::Dimensions(evaluator.dimensions()),
block_shape, block_size);
block_size = block_mapper.block_dims_total_size();
const size_t aligned_blocksize =
EIGEN_MAX_ALIGN_BYTES *

4
unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
View File

@ -976,7 +976,8 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType, MakePointer_>,
// find that scattered reads are not worth supporting in
// TensorSliceBlockMapper.
TensorSliceBlockMapper block_mapper(
input_tensor_dims, tensor_slice_offsets, tensor_slice_extents,
typename TensorSliceBlockMapper::Dimensions(input_tensor_dims),
tensor_slice_offsets, tensor_slice_extents,
target_input_block_sizes, DimensionList<Index, NumInputDims>());
const Index num_outputs_to_update =
@ -1232,7 +1233,6 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType, MakePointer_>,
} else if (!first_preserved_dim_allocated) {
// TODO(andydavis) Include output block size in this L1 working set
// calculation.
const Index allocated = max_coeff_count - coeff_to_allocate;
const Index alloc_size = numext::maxi(
static_cast<Index>(1), coeff_to_allocate / reducer_overhead);
(*target_input_block_sizes)[dim] =