YOSHIFUJI Naoki

> Where is this called from? You can use graphviz to make call tree from cprofile results.

`copy_from_host()` could report its processing time including other processes'. As you know about OpenCL, `clEnqueueWrite/ReadBuffer()` waits to finish all tasks in the CommandQueue. You need to separate such waiting time...

You can insert clQueueFinish() explicitly before enqueuing read/write.

Can't you get function calling order? The calling order is enough to check what command was enqueued and consumes time.

I tried to upgrade the base version to estimate how difficult it is. * [v2.1.0 -> v2.1.0.1](https://github.com/fixstars/clpy/pull/275) is very easy * [v2.1.0.1 -> v2.2.0](https://github.com/fixstars/clpy/tree/33-v2.2.0) is a little difficult but not...

Note that the above experimental branches would not work well, I've never checked to build and run.

The easiness shown above is from my commits made for the first time as ClPy. In anticipation of the base version update, I firstly renamed files by git mv (f5ff2ee,...

@ybsh I understand that your question is about clpy itself but not about this issue. If you suggest raising an error on `copy_from_device()` (and also on _async), please make an...

This issue would be pending because `_async` functions are not used in cupy tests and even in chainer.

Yes, I understand. It's very difficult on OpenCL to behaive same as CUDA. Therefore, We decide that ClPy won't support `PinndeMemory` (in the near future at least) @nsakabe-fixstars can you...