2.2.0

Overview

This release introduces NumPy patching functionality, improves FFT performance consistency, and modernizes dependency management. Python 3.9 support has been dropped, and mkl-service is now an optional dependency.

Highlights

🎯 NumPy FFT Patching

• New API for seamlessly patching NumPy FFT with Intel MKL optimizations
• Context manager support for temporary patching
• Functions to query patch status and control patching explicitly

⚡ Performance Improvements

• Enhanced shape checking in N-D FFT functions for better performance consistency
• More frequent use of faster direct transforms

📦 Improved Dependency Management

• mkl-service is now an optional dependency
• Configurable NumPy dependency in conda recipes

⚠️ Breaking Changes

• Python 3.9 support dropped - Minimum supported version is now Python 3.10+

Python Version Support

• ✅ Python 3.10, 3.11, 3.12, 3.13, 3.14
• ❌ Python 3.9 and earlier

Detailed Changes

Added

• Added mkl_fft patching for NumPy, with mkl_fft context manager, is_patched query, and patch_numpy_fft and restore_numpy_fft calls to replace numpy.fft calls with calls from mkl_fft.interfaces.numpy_fft gh-224, gh-295

Changed

• In mkl_fft.fftn and mkl_fft.ifftn, improved checking of the shape argument s to use faster direct transforms more often. This makes performance more consistent between mkl_fft.fftn/ifftn and mkl.interfaces. gh-283
• Made mkl-service an optional dependency gh-305
• Made conda recipe dependency on numpy configurable through USE_NUMPY_BASE environment variable gh-299

Removed

• Dropped support for Python 3.9 gh-243

Fixed

• Fix TypeError exception raised with empty axes gh-288
• To return input array unchanged when axes=() (ignore out parameter) gh-293

Contributors

• Nikita Grigorian (@ndgrigorian)

New Contributors

• Anton Volkov (@antonwolfy)
• Christopher Hillenbrand (@chillenb)
• Nikolay Petrov (@napetrov)

Full Changelog: 2.1.2...2.2.0