Skip to content

Latest commit

 

History

History
166 lines (112 loc) · 5.75 KB

CHANGELOG.md

File metadata and controls

166 lines (112 loc) · 5.75 KB

  • v2.2.3 - April 14, 2024

    • Add KEOPS_CACHE_FOLDER environment variable
    • Now require explicitely python>=3.8 in doc
    • Improve cuda detection with conda
    • RKeOps release

  • v2.2.2 - Feb 9, 2024

    • Fix a utf8 encoding problem with cuda_fp16.h file

  • v2.2.1 - Jan 25, 2024

    • Fix a memory leak introduced with support of forward AD (issue 353)

  • v2.2 - Jan 18, 2024

    • added option to disable fast_math Cuda compiler option
    • added comparison operators between LazyTensor
      • added support for symbolic differentiation of symbolic operations : Grad, Diff, Laplacian, Divergence
      • added support for forward autodiff, compatible with PyTorch forward autodiff tools (torch.jvp, etc.)
      • added support for torch.vmap
      • added support for intermediate variables in formulas (factorize and auto_factorize methods of LazyTensor class)
    • changed cache folder name, now include name of local host
    • added torch.compile comparison to the benchmarks
    • fixed issues 294, 305, 310, 325, 342, ...

  • v2.1.2 - Apr 4, 2023

    • Fixes a memory leaks introduced in version v2.x (issues 284 and 300).
    • The name of the cache folder now include the hostname.
    • Fix a bad memory allocation in tensordot
    • Add Kronecker product

  • v2.1.1 - Jan 6, 2023

    • Fixes for issues 220, 263, 256, 266, 275, 262, 282.
    • Fixed an error caused by a compile warning on recent MacOs systems.
    • Added an official image on DockerHub and updated instructions for Singularity and Docker.

  • v2.1 - Jun 3, 2022

    • Fix for gradient formula of divide operation (issue #246).
    • Fix for special chunked computation mode (issue #243).
    • Fix for the gradient formula for complex multiplication operation (issue #238).
    • Other minor bug fixes (issues #242, #240, #248, #241, #233).
    • Added support for BSpline kernels.

  • v2.0 - Mar 17, 2022

    • Complete rewritting of meta-programming engine: compilation times are divided by 10-100
    • JIT compilation of cuda kernels
    • added support for in-place operations.
    • keopsLab is now deprecated
    • improve unit test framework
    • Many bug fixing

  • v1.5 - Mar 22, 2021

    • Add multiple operations: acos, asin, atan, atan2, sinc, if-else, round, modulo.
    • Complex numbers: full python support, pairs of float/double for other languages.
    • Extensive benchmarks for knn, etc...
    • Extensive documentation improvements.
    • New compilation pipeline for python.

  • v1.4.2 - Nov 27, 2020

    • Refactor pykeops LazyTensor class.
    • Add multi gpu example with pykeops.
    • pykeops' kernel_product module is deprecated.
    • Fix pykeops.set_bin_folder() function.
    • remove Gputils dependency for pykeops.
    • Add Clamp operator in keops.

  • v1.4.1 - Aug 19, 2020

    • Fix compatibility issue with PyTorch 1.5-6.
    • Fix compatibility issue with Cuda 11.0.
    • Improve performances for computations with dimension D>100.

  • v1.4 - Mar 22, 2020

    • Added multiprocessor parallelization in CPU mode.
    • Added support for float16 data type.

  • v1.3 - Jan 16, 2020

    • rKeOps: KeOps binding for R (uploaded in CRAN).
    • Refactor KeOps binders: there is a Cpp namespace called "keops_binder" providing an easy entry point to keops functions (that are in the namespace "keops").
    • Add accuracy options for single precision summations: block summations (used by default now), mixed precision and Kahan compensated scheme.
    • Add a pykeops.clean_pykeops() routine to flush the cache dir.

  • v1.2 - Sep 25, 2019

    • Add TensorDot operation.
    • Refactor Cpp codes.

  • v1.1.2 - Aug 8, 2019

    • Fix bug in batch computations.
    • Add to wheel PyKeOps package licence file.

  • v1.1.1 - Jul 16, 2019

    • Change LazyTensor imports. One should use: from pykeops.{numpy,torch} import LazyTensor.
    • Add to wheel pykeops package a 'devtools' option that install cmake and gcc.

  • v1.1 - Jul 10, 2019

    • Add Support for LazyTensor.
    • Improve the documentation.
    • Refactor KeOps cpp directory structure.
    • Improve build dir structure: builds are done in a subdirectory and builds junks are removed after sucessful compilation. Add keops_hash.log file to decypher .so names.
    • Add support for environment variable (PYKEOS_VERBOSE and PYKEOPS_BUILD_TYPE) to make debug and verbosity easier.

  • v1.0.2 - Apr 21, 2019

    • fix a bug in variable parsing for pykeops bindings.
    • KernelSolve syntax: move alpha kwarg in 'KernelSolve.call'.
    • Improve the documentation.

  • v1.0.1 - Apr 9, 2019

  • v1.0 - Apr 4, 2019

    • Add conjugate gradient solver KernelSolve.
    • Change variable notations in formula: {Vx,Vy} to {Vi,Vj}. Old notations are still compatible (warning message).
    • Change kwarg 'cuda_type' to 'dtype'. Old Kwarg is still valid for backward compatibility.
    • Doc has been improved.

  • v0.1.6 - Mar 13, 2019

  • v0.1.5 - Feb 28, 2019

    • KeOps: the upper bound on CudaBlockSize is now computed at compilation time (and not at runtime). So binaries have to be recompiled when the hardware is changing.
    • PyKeOps: Lock file for the Build dir is now done in python.

  • v0.1.4 - Jan 31, 2019

    • PyKeOps compatibility with PyTorch 1.0.
    • Some refactoring of KeOpslab code (Kernel function is now called keops_kernel and Grad function is called kernel_grad).

  • v0.1.3 - Nov 6, 2018

    • Fix pickle problem with inherited class in pytorch.
    • 'import pykeops' does not try to import torch any more.

  • v0.1.2 - Oct 31, 2018

    • Small fixes on wheel packages.

  • v0.1.0 - Jul 20, 2018

    • PyKeOps is now using Pybind11.
    • Multi GPU support.
    • Min, Max ... reduction added.
    • doc with sphinx.

  • v0.0.12 - Jul 4, 2018

    • Some performance improvements in the formula (Square, Inv, ...).

  • v0.0.10 - May 29, 2018

    • First release of KeOps.