Release Notes¶
Version 2.0¶
NumbaPro has been deprecated, and its code generation features have been moved into open-source Numba. The CUDA library functions have been moved into Accelerate, along with some Intel MKL functionality. High-level functions and access to additional native library implementations will be added in future releases of Accelerate, and there will be no further updates to NumbaPro.
A NumbaPro compatibility layer (listed as release 0.22.0 of NumbaPro) provides access to the new Accelerate packages through the old NumbaPro package names. This avoids the need to change any existing code immediately for use with Accelerate. A warning will be generated upon import of the compatibility layer, to highlight the deprecation of the NumbaPro package.
CUDA Libraries¶
CUDA library functionality is equivalent to that in NumbaPro 0.21, with the following packages renamed:
NumbaPro Package | Accelerate package |
---|---|
numbapro.cudalib.cublas |
accelerate.cuda.blas |
numbapro.cudalib.cufft |
accelerate.cuda.fft |
numbapro.cudalib.curand |
accelerate.cuda.rand |
numbapro.cudalib.cusparse |
accelerate.cuda.sparse |
numbapro.cudalib.sorting |
accelerate.cuda.sorting |
Code Generation¶
The vectorize
targets parallel
and cuda
can now be accessed
with Numba, as can the cuda.reduce
decorator. Printing of integers and
floating point values from CUDA kernels is also possible in Numba, and no longer
requires NumbaPro or Accelerate to be imported.
Intel MKL¶
The MKL FFT implementation is available in the package
accelerate.mkl.fftpack
, and MKL service functions are available in the
accelerate.mkl
package.