Kenichi Maehashi comments

Results 160 comments of


Kenichi Maehashi

Add an attribute to FunctionNode to indicate it supports arrays on different devices

This should solve issues like #4144 (GPU mismatch).

ndarray.set() is dramatically slower when the input is a non contiguous array

Thanks for the feedback! When coping CPU memory to GPU, the memory needs to be contiguous, so this is a restriction that is not easy to relax. c.f. #6785

ndarray.set() is dramatically slower when the input is a non contiguous array

> Or if we can at least write a version of ascontiguous which is backed by pinned memory. Yes, I agree pinned memory should be used here. We'll work on...

`compile_with_cache()` is mistakenly exposed to public?

RawKernel introduced in v5 and RawModule introduced in v7. `compile_with_cache` has never been documented API, but it's widely used by major users like Thinc and 3d-ken-burns which need `Raw*` features...

`compile_with_cache()` is mistakenly exposed to public?

Yeah, let's do it in v10rc!

`compile_with_cache()` is mistakenly exposed to public?

Let's keep this issue open as we are in process of deprecation.

MemoryAsyncPool test issues (tracker issue)

@leofang Can we revert #5308 as well?

MemoryAsyncPool test issues (tracker issue)

Thanks! I think this there's no rush 😃

Parallelize Cythonize

/test mini

Parallelize Cythonize

> Question: Is `python setup.py build_ext -j N` expected to work (even before this PR)? I don't need this support, just curious 🙂 No, because we overwrite `self.parallel` (which is...