wkcn edited a comment on issue #11913: Unexpectedly poor copy() performance URL: https://github.com/apache/incubator-mxnet/issues/11913#issuecomment-409904550 @safrooze Yes. I modified the code then reported the result. Here is the code I modified. ``` a = nd.empty((1, 512, 120*120)) nd.waitall() start = time() for _ in range(2000): b = a.copy() nd.waitall() print('\tcopy: elapsed: {:.2f}'.format(time() - start)) start = time() for _ in range(2000): b = a + 0.0 nd.waitall() print('\tcopy via add: elapsed: {:.2f}'.format(time() - start)) ``` And I test the code you provided, the elapsed time of copy is almost invariable. But the elapsed copy via add is around 4. The reason that I changed the code is that the computation of NDArray is lazy. The script `b = a + 0.0` only records this operation but there is no actual computation until `nd.waitall` is called.
---------------------------------------------------------------- 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
