TensorFlow NUFFT 0.12.0

Release 0.12.0

This release is compiled against TensorFlow 2.11.

Major Features and Improvements

• Added support for 1D transforms to the GPU kernel.

TensorFlow NUFFT 0.11.0

Release 0.11.0

Major Features and Improvements

• Added new option points_range to control the range supported by the
  algorithm. This option provides a trade-off between flexibility and
  performance.
• Added new option debugging.check_points_range to assert that the input
  points lie within the supported range.

Bug Fixes and Other Changes

• nufft type-1 will now raise an error when source and points have
  incompatible samples dimensions. Previously the computation would have
  proceeded, ignoring any additional samples in source.
• Improved error reporting for invalid grid_shape arguments. nufft will
  now raise an informative error when grid_shape has an invalid length or
  when the user fails to provide it for type-1 transforms. Previously, nufft
  would have behaved erratically or crashed.

TensorFlow NUFFT 0.10.1

Release 0.10.1

Bug Fixes and Other Changes

• Fixed a bug where the options where not being passed to the NUFFT op when
  computing the gradient with respect to the points.

TensorFlow NUFFT 0.10.0

Release 0.10.0

This release bumps the supported TF version to v2.10.x.

TensorFlow NUFFT 0.9.0

Release 0.9.0

Major Features and Improvements

• Added new class Options to be passed to nufft to specify advanced options.
• Added new argument options to nufft to allow specifying advanced options.

Bug Fixes and Other Changes

• FFTW planning rigor now defaults to MEASURE instead of ESTIMATE, but this
  can now be changed via the Options mechanism.

TensorFlow NUFFT 0.8.1

Release 0.8.1

Bug Fixes and Other Changes

• We now use TensorFlow's stream executor to perform FFTs.
• This library should now be fully usable in systems without a CUDA
  installation (CPU only).

TensorFlow NUFFT 0.8.0

Release 0.8.0

This release bumps the supported TensorFlow version to 2.9.0.

Bug Fixes and Other Changes

• Like core TensorFlow, we now compile with _GLIBCXX_USE_CXX11_ABI=1.
• Like core TensorFlow, Python wheels now conform to manylinux2014, an upgrade
  from manylinux2010.
• The op library now links against the static form of the CUDA runtime
  library.
• FINUFFT code is now fully integrated.

Known Caveats

• The op library does not link against the the static cuFFT library, which will
  result in unresolved symbol errors when used in a system without a CUDA
  installation. This will be addressed in a future release (see also issue #24).

TensorFlow NUFFT 0.7.3

Release 0.7.3

Bug Fixes and Other Changes

• Fixed an uninitialized memory issue in the GPU implementation of the
  spread op that would occasionally result in the computation returning
  NaNs or incorrect results.

Thanks to our Contributors

This release contains contributions by chaithyagr_.

.. _chaithyagr: https://github.com/chaithyagr

TensorFlow NUFFT 0.7.2

Release 0.7.2

Bug Fixes and Other Changes

• Fixed a race condition in the GPU implementation that would occasionally
  result in the computation returning NaNs or incorrect results.

Thanks to our Contributors

This release contains contributions by
chaithyagr <https://github.com/chaithyagr>_.

TensorFlow NUFFT 0.7.1

Release 0.7.1

Bug Fixes and Other Changes

• Fixed a bug in computation of the gradient with respect to points that would
  result in incorrect gradients when passed a complex-valued upstream gradient.