dask: dask.fft does not work with cupy arrays

import numpy as np

import cupy as cp

import dask.array as da

x = cp.random.randn(128, 128)
x = da.from_array(x)

da.fft.fft2(x)

Similar to my last issue. It may be that I’ve mistakenly assumed dask has support for cupy, in which case my apologies. I’ve seen examples with dask and cupy, but do not know how extensive that is.

About this issue

Original URL
State: closed
Created 4 years ago
Comments: 15 (10 by maintainers)

Most upvoted comments

I’m glad to hear it. Thanks for reporting @leofang .

Closing.

mrocklin on Nov 3, 2020

I verified this works with CuPy v8 and Dask v2.30.

leofang on Nov 2, 2020

I tested this with CuPy 8.0.0b5 and it works, but still fails with 7.7.0. Given there has been a lot of changes for CuPy 8 on the cuFFT front, I think it’s unlikely that those changes will be backported to CuPy 7.

@izaid FYI, CuPy 8 is scheduled for release in late September, so by then this should be no problem anymore, if you can’t wait you may considering building it from source to get past this issue.

pentschev on Aug 5, 2020