junrushao1994 commented on a change in pull request #7082: Sparse Tensor:
request for reviews
URL: https://github.com/apache/incubator-mxnet/pull/7082#discussion_r131105014
##########
File path: include/mxnet/ndarray.h
##########
@@ -93,24 +154,95 @@ class NDArray {
Mkl_mem_ = std::make_shared<MKLMemHolder>();
#endif
}
+
+ /*!
+ * \brief constructing a static NDArray of non-default storage that shares
data with TBlob
+ * Use with caution: allocate ONLY ONE NDArray for each TBlob,
+ * make sure the memory region is available through out the life of NDArray
+ * \param stype the storage type of NDArray
+ * \param shape the shape of NDArray
+ * \param data the memory content of static data
+ * \param aux_data the memory content of static aux data
+ * \param dev_id the device id this tensor sits at
+ */
+ NDArray(const NDArrayStorageType stype, const TShape &shape,
+ const TBlob &data, const std::vector<TBlob> &aux_data, int dev_id)
+ : ptr_(std::make_shared<Chunk>(stype, data, aux_data, dev_id)),
shape_(shape),
+ dtype_(data.type_flag_), entry_({nullptr, 0, 0}) {
+#if MKL_EXPERIMENTAL == 1
+ Mkl_mem_ = std::make_shared<MKLMemHolder>();
+#endif
+ }
+
+
/*!
- * \return the shape of current NDArray
+ * \return the shape of current NDArray.
*/
inline const TShape& shape() const {
return shape_;
}
/*!
+ * \return the shape of underlying chunk which stores the NDArray data/value.
+ * It is only intended for non-default storage. For row-sparse storage, it
is the shape of
+ * the tensor which stores the non-zero values.
+ */
+ inline const TShape &storage_shape() const {
+ CHECK(ptr_ != nullptr);
+ CHECK_NE(storage_type(), kDefaultStorage);
+ return ptr_->storage_shape;
+ }
+
+ /*!
+ * \return the shape of aux data at ith index. If it doesn't exist, return
an empty one.
+ */
+ inline const TShape aux_shape(size_t i) const {
Review comment:
I guess using some name like 'index' is better than 'i'
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
