Skip to content

Releases: fjarri/reikna

v0.8.0

20 Jan 07:27
Compare
Choose a tag to compare
  • CHANGED: dest_offset renamed to dst_offset to match pyopencl/pycuda API.

  • FIXED: register numpy.bool in addition to bool - it is a separate type now (@perdigao1).

  • FIXED: incompatibility with modern versions of pyopencl.

v0.7.4

26 Jan 05:14
Compare
Choose a tag to compare
  • FIXED: a bug with Array.__setitem__ with the source being a 0-dimensional array.

v0.7.3

26 Jan 05:14
Compare
Choose a tag to compare
  • ADDED: inverse parameter for :py:class:~reikna.fft.FFTShift (contributed by @drtpotter).

  • ADDED: :py:meth:~reikna.core.Type.with_dtype method for :py:class:~reikna.core.Type.

  • ADDED: :py:func:~reikna.transformations.cast transformation.

  • ADDED: :py:meth:~reikna.core.Type.broadcastable_to method for :py:class:~reikna.core.Type.

  • ADDED: added :py:func:~reikna.transformations.copy_broadcasted transformation.

  • ADDED: :py:meth:~reikna.cluda.api.Thread.get_cached_computation method for :py:class:~reikna.cluda.api.Thread.

  • ADDED: arrays now support setting arbitrary slices with scalars or arrays.

  • ADDED: support for get() method for non-contiguous arrays.

  • ADDED: :py:func:~reikna.concatenate for concatenating GPU arrays.

  • ADDED: :py:func:~reikna.roll for GPU arrays and the inplace version :py:meth:~reikna.cluda.api.Array.roll.

  • FIXED: updated the CUDA backend for the change async -> async_ in the new versions of PyCUDA. Bumped PyOpenCL and PyCUDA versions to 2018.1.1.

  • FIXED: an error in the conversion of numpy.int64 to ctype for Windows.

  • FIXED: an unstable type of nbytes in Thread.array(), leading to problems with calling the C++ backend later on.

  • FIXED: a bug where a nonzero offset was ignored when building an accessor macro for an array with a zero-length shape