guanxinq commented on a change in pull request #17569: Adding sparse support to
MXTensor for custom operators
URL: https://github.com/apache/incubator-mxnet/pull/17569#discussion_r387946986
##########
File path: include/mxnet/lib_api.h
##########
@@ -229,20 +241,75 @@ enum MXReturnValue {
MX_SUCCESS = 1,
};
+// For sparse input, read/write the data from NDarray via pointers.
+struct MXInSparse {
+ // Pointer to data.
+ void *data{nullptr};
+ // length of (non-zero) data.
+ int64_t data_len;
+
+ // To store aux data for sparse.
+ // For CSR, indices stores the col index of non-zero values.
+ // For row sparse, indices store row index of rows which have non-zero
values.
+ int64_t* indices;
+ int64_t indices_len;
+
+ // For CSR, indptr gives the start and end index of data for each row.
+ // For row sparse, indptr is empty.
+ int64_t* indptr;
+ int64_t indptr_len;
+
+ void set(void *data_ptr, const int64_t* dims, int ndims, void *idx,
+ int64_t num_idx, void *idx_ptr = nullptr, int64_t num_idx_ptr = 0) {
+ data = data_ptr;
+ // If CSR, num of non-zero value is num_idx,
+ // If row sparse, num of value is num_idx * width.
+ data_len = idx_ptr ? num_idx : num_idx * dims[1];
Review comment:
You are right. Will update.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to 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
