devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Block evaluation for TensorChipping + fixed bugs in TensorPadding and TensorSlicing

Browse Source
This commit is contained in:
Eugene Zhulenev
2019-10-09 12:45:31 -07:00
parent f0a4642bab
commit 33e1746139
9 changed files with 281 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
unsupported/Eigen/CXX11/src/Tensor/TensorBlockV2.h
View File

@@ -418,12 +418,22 @@ class TensorMaterializedBlock {
if (can_use_direct_access) {
const Scalar* block_start = data + desc.offset();
return TensorMaterializedBlock(internal::TensorBlockKind::kView, block_start,
desc.dimensions());
return TensorMaterializedBlock(internal::TensorBlockKind::kView,
block_start, desc.dimensions());
} else {
void* mem = scratch.allocate(desc.size() * sizeof(Scalar));
Scalar* block_buffer = static_cast<Scalar*>(mem);
// Try to reuse destination as an output block buffer.
Scalar* block_buffer = desc.template destination<Scalar, Layout>();
bool materialized_in_output;
if (block_buffer != NULL) {
materialized_in_output = true;
} else {
materialized_in_output = false;
void* mem = scratch.allocate(desc.size() * sizeof(Scalar));
block_buffer = static_cast<Scalar*>(mem);
}
typedef internal::TensorBlockIOV2<Scalar, IndexType, NumDims, Layout>
TensorBlockIO;
@@ -438,8 +448,11 @@ class TensorMaterializedBlock {
TensorBlockIO::Copy(dst, src);
return TensorMaterializedBlock(internal::TensorBlockKind::kMaterializedInScratch,
block_buffer, desc.dimensions());
return TensorMaterializedBlock(
materialized_in_output
? internal::TensorBlockKind::kMaterializedInOutput
: internal::TensorBlockKind::kMaterializedInScratch,
block_buffer, desc.dimensions());
}
}
@@ -1141,7 +1154,7 @@ class TensorBlockAssignment {
it[idx].count = 0;
it[idx].size = target.dims[dim];
it[idx].output_stride = target.strides[dim];
it[idx].output_span = it[i].output_stride * (it[i].size - 1);
it[idx].output_span = it[idx].output_stride * (it[idx].size - 1);
idx++;
}