nudles commented on a change in pull request #632: make astype return new tensor
URL: https://github.com/apache/singa/pull/632#discussion_r395954782
##########
File path: src/core/tensor/tensor.cc
##########
@@ -152,29 +152,24 @@ Tensor Resize(const Tensor &in, const Shape &shape) {
}
\
} while (0)
-Tensor &Tensor::AsType(const DataType type) {
+// return new tensor
+Tensor Tensor::AsType(const DataType type) {
Review comment:
Do we have `in-place` for any other tensor functions?
the behavior is
1. in-place = true, then reuse the memory (only works if the bitwidth is the
same for the two data types).
2. in-place= false, return a tensor with a new memory block.
case 1 is a bit confusing for the case that the bitwidths are different for
the two data types although it may save some cost.
so let's just ignore the in-place flag.
is your concern in the same?
----------------------------------------------------------------
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
