This is the release note of v3.0.1

Changelog

Problem Fixes

• Fixed incorrect execution results of nlcpy.fft.fftshift() and nlcpy.fft.ifftshift() that are returned when the number of array dimensions is larger than 1 and the input parameter of axis is not equal to 0.

• Fixed incorrect execution results of nlcpy.linalg.norm() that are returned when the number of VE threads is larger than 1 and the memory layout of input array is not contiguous.

• Fixed a problem where nlcpy.diff() raises unexpected RuntimeWarning.

This is the release note of v3.0.0

Changelog

New Features

• Newly added shared libraries for VE30.

This is the release note of v2.2.0

Changelog

New Features

• Newly added VE device management routines, which enable you to select execution VE device from Python script.
• Newly added environment variable VE_NLCPY_MEMPOOL_SIZE which controls amount of memory pool.
• Newly supports a feature that NLCPy automatically replaces almost functions and methods that implemented not yet to the NumPy’s one.

Performance Enhancements

• Improved the performance of nlcpy.power() and nlcpy.floor_divide() when input data types are complex.

This is the release note of v2.1.1

Changelog

New Features

• Newly added environment variable VE_NLCPY_WARMUP which enables/disables Python scripts to perform some warming up when importing NLCPy.
• Minor updates for mpi4py-ve support.

Performance Enhancements

• Improved the performance of nlcpy.linalg.norm() if the Frobenius norm is specified.
• Improved the performance of some functions if the memory layout is C-contiguous or F-contiguous.

This is the release note of v2.1.0

Note

NLCPy v2.1.0 requires to install the RPM version later than or equal to nec-blas-ve-2.3.0-2.5.

Changelog

New Features

• Newly added Just-In-Time compilation functionality, which enables your Python scripts to call VE functions written by C/C++/Fortran.
• Newly added Masked Array operations.

Performance Enhancements

• Improved the performance of nlcpy.ndarray.__getitem__() with list, tuple, or nlcpy.ndarray elements.
• Improved the performance of nlcpy.linalg.eigvalsh().

This is the release note of v2.0.0

Changelog

New Features

• Newly added SCA Interface, which enables Python scripts to call SCA functions of NEC Numeric Library Collection.

• Newly added the profiling routines with FTRACE.

• Newly added the following functions.

  • nlcpy.save()
  • nlcpy.savetxt()
  • nlcpy.savez()
  • nlcpy.savez_compressed()
  • nlcpy.roll()
  • nlcpy.swapaxes()
  • nlcpy.unique()
  • nlcpy.count_nonzero()
  • nlcpy.broadcast_arrays()
  • nlcpy.diag_indices()
  • nlcpy.diagflat()
  • nlcpy.tri()
  • nlcpy.tril()
  • nlcpy.triu()
  • nlcpy.fill_diagonal()
  • nlcpy.hsplit()
  • nlcpy.vsplit()
  • nlcpy.split()
  • nlcpy.atleast_1d()
  • nlcpy.atleast_2d()
  • nlcpy.atleast_3d()
  • nlcpy.block()
  • nlcpy.clip()
  • nlcpy.select()
  • nlcpy.nanargmax()
  • nlcpy.nanargmin()

Performance Enhancements

• Improved the overall performance of NLCPy by applying asynchronous VE function calls.
• Improved the performance of reduction operations such as nlcpy.sum(), nlcpy.amax(), and nlcpy.amin().

This is the release note of v1.0.1

Changelog

Problem Fixes

• Fixed abnormal termination that could occur when a Python script using NLCPy is executed by a batch scheduler.

• Fixed a problem that stalls when a multi-dimensional ndarray is specified as a parameter to nlcpy.sort() .

Function Enhancements

• Improved a part of the memory allocation logic during the lazy evaluation.

This is the release note of v1.0.0

Changelog

New Features

Newly added the following functions and methods.

Descrete Fourier Transform

• nlcpy.fft.fft()
• nlcpy.fft.ifft()
• nlcpy.fft.fft2()
• nlcpy.fft.ifft2()
• nlcpy.fft.fftn()
• nlcpy.fft.ifftn()
• nlcpy.fft.rfft()
• nlcpy.fft.irfft()
• nlcpy.fft.rfft2()
• nlcpy.fft.irfft2()
• nlcpy.fft.rfftn()
• nlcpy.fft.irfftn()
• nlcpy.fft.hfft()
• nlcpy.fft.ihfft()
• nlcpy.fft.fftfreq()
• nlcpy.fft.rfftfreq()
• nlcpy.fft.fftshift()
• nlcpy.fft.ifftshift()

Linear Algebra

• nlcpy.inner()
• nlcpy.outer()
• nlcpy.linalg.svd()
• nlcpy.linalg.cholesky()
• nlcpy.linalg.qr()
• nlcpy.linalg.eig()
• nlcpy.linalg.eigh()
• nlcpy.linalg.eigvals()
• nlcpy.linalg.eigvalsh()
• nlcpy.linalg.norm()
• nlcpy.linalg.solve()
• nlcpy.linalg.lstsq()
• nlcpy.linalg.inv()
• nlcpy.linalg.LinAlgError()

Array Creation Routines

• nlcpy.meshgrid()
• nlcpy.logspace()

Array Manipulation Routines

• nlcpy.copyto()
• nlcpy.hstack()
• nlcpy.stack()
• nlcpy.vstack()
• nlcpy.append()
• nlcpy.delete()
• nlcpy.insert()
• nlcpy.fromfile()
• nlcpy.loadtxt()
• nlcpy.load()
• nlcpy.flip()
• nlcpy.fliplr()
• nlcpy.flipud()
• nlcpy.repeat()

Statistics Routines

• nlcpy.ptp()
• nlcpy.percentile()
• nlcpy.nanpercentile()
• nlcpy.quantile()
• nlcpy.nanquantile()
• nlcpy.nanmedian()
• nlcpy.correlate()
• nlcpy.histogram()
• nlcpy.histogram2d()
• nlcpy.histogramdd()
• nlcpy.histogram_bin_edges()
• nlcpy.bincount()
• nlcpy.digitize()

Mathematical Functons

• nlcpy.prod()

Ndarray Methods

• nlcpy.ndarray.ptp()
• nlcpy.ndarray.mean()
• nlcpy.ndarray.var()
• nlcpy.ndarray.std()
• nlcpy.ndarray.conj()
• nlcpy.ndarray.conjugate()
• nlcpy.ndarray.cumsum()
• nlcpy.ndarray.dot()
• nlcpy.ndarray.prod()

Problem Fixes

Fixed abnormal termination that could occur when a large amount of memory were allocated.

This is the release note of v1.0.0b2

Changelog

Performance enhancements

• Improved performance of universal functions with multi-dimensional arrays.
• Improved performance of broadcasting operations from a scalar value to an ndarray.
• Improved performance of nlcpy.arccos() and nlcpy.arctanh().
• Improved performance of random number generators by changing the number of threads to be execeted on VE.

Bug fixes

• Fixed a bug in nlcpy.sort() that might cause a freeze of Python interpreter at the end of the program.
• Fixed a bug in nlcpy.matmul() that returned invalid results when input ndarrays (a and b) fulfill any of the following conditions:
  • a.flags.c_congituous is False and a.flags.f_contiguous is False
  • b.flags.c_congituous is False and b.flags.f_contiguous is False
• Fixed a bug in nlcpy.prof.print_run_stats() that might showed negative elapsed time.