0.22.0

New Backend support

clEsperanto now comes with support for CUDA 12 and Metal. While OpenCL remain the default backend, you can now install pyclesperanto with additional backend

pip install pyclesperanto          // only install opencl
pip install pyclesperanto[cuda]    // not available in MacOS
pip install pyclesperanto[metal]   // only available in MacOS
pip install pyclesperanto[all]     // install all possible configuration

and you can select and choose you backend at the start of the import:

import pyclesperanto as cle
cle.list_avaiable_backend()
>>> ['opencl', 'cuda']
cle.select_backend("cuda")

DLPack support

And now the Array object support DLPack which allows both way copy-free interoperability with other GPU library that uses same backend and support the same protocol (e.g. cupy, torch, ...)

import pyclesperanto as cle
import cupy as cp
cle.select_backend("cuda")
cp_array = cp.random.random((10,10), dtype=cp.float32)
cle_array = cle.from_dlpack(cp_array)

Degraded

This release came with a reshape of the project architecture to allow multiple backend to coexist. As a result, the conda-recipe in conda-forge is broken.

Full Changelog: 0.21.4...0.22.0

v0.21.4

Default pyclesperanto ship with OpenCL
Option pyclesperanto[cuda] or pyclesperanto[all] will now also provide CUDA as a backend (for unix and windows only!)

>>> import pyclesperanto as cle
>>> cle.list_available_backends()
['opencl', 'cuda']
>>> cle.select_backend("cuda")
>>> cle.get_device()
(CUDA) Tesla T4 (driver 13.0)
[...]

Full Changelog: 0.21.1...0.21.4

v0.21.1

Full Changelog: 0.21.0...0.21.1

v0.21.0

Introducing multiple backend (#379)

• Enable multiple backends from CLIc to coexists
  • pyclesperanto is now pure python
  • Hard dependency with pyclesperanto-opencl
  • Optional dependency with pyclesperanto-cuda

Updates and new functions (#363, #369, #381)

• Provide evalueate() for quick elementwise arithmetic operation on arrays
• Restructure tests organisation
• Add new function for image processing (e.g. threshold_yen)

Documentation

• Update demos (still WIP) by @StRigaud in #365

Broken

• This new version should break the conda recipe. Installation using pip is advised

pip install pyclesperanto
pip install pyclesperanto[all]  <- to test CUDA backend

Full Changelog: 0.19.0...0.21.0

v0.19.0

What's Changed

• Update to CLIc@0.19.0 by @github-actions[bot] in #356

Full Changelog: 0.18.3...0.19.0

0.18.3

What's Changed

• add a runtime-cache for program to avoid rebuild time and possible memory leak related to OpenCL drivers on MacOS

Full Changelog: 0.18.2...0.18.3

0.18.2

What's Changed

• This patch fixes a memory leak at Buffer Allocation (more info here)

Full Changelog: 0.18.0...0.18.2

0.18.0

What's Changed

See full changelog

Full Changelog: 0.17.1...0.18.0

0.17.1

What's Changed

• Update to CLIc@0.17.1 by @github-actions in #321

Full Changelog: 0.17.0...0.17.1

0.17.0

What's Changed

• Introduce FFT operation adapted from clij2-fft
• Add CLAHE filter
• Fix minor issues

See Changelog for more information on this release

Full Changelog: 0.16.1...0.17.0