2.7.0

Overview

This release modernizes mkl-service with updated Intel MKL instruction set support, improved dependency management, and enhanced code quality. Python 3.9 support has been dropped, and several legacy CBWR branches have been removed in alignment with the latest Intel® oneAPI Math Kernel Library (oneMKL).

Highlights

🎯 New Intel Architecture Support

• Added support for AVX10 and AVX512_E5 instruction set architecture (ISA) constants
• Introduced AVX10 conditional numerical reproducibility (CNR) constants with strict variants

📦 Improved Dependency Management

• MKL is now explicitly declared as a runtime dependency
• Build process simplified with mkl-devel as a formal build dependency

🔧 Code Quality Improvements

• Fixed Cython compilation warnings for cleaner builds
• Removed star imports from package initialization, preventing namespace pollution
• Improved module import hygiene

⚠️ Breaking Changes

• Python 3.9 support dropped - Minimum supported version is now Python 3.10+
• Legacy CBWR branches removed: ssse3, sse4_1, avx, avx512_mic, avx512_mic,strict, and avx512_mic_e1 are no longer supported

Python Version Support

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

Detailed Changes

Added

• Added support for ISA constants "avx10" and "avx512_e5" and CNR constants "avx10" and "avx10,strict" gh-175

Changed

• Added mkl as a runtime dependency and mkl-devel as build dependency in pyproject.toml gh-177

Removed

• Dropped support for Python 3.9 gh-118
• Dropped support for "ssse3", "sse4_1", "avx", "avx512_mic", "avx512_mic,strict", and "avx512_mic_e1" cbwr branches gh-173

Fixed

• Fixed Cython warnings when building mkl-service gh-183
• Removed use of star import in mkl-service initialization, which removes imported module pollution from the namespace gh-178

Contributors

• Anton Volkov (@antonwolfy)
• Evseniia Komarova (@ekomarova)
• Nikita Grigorian (@ndgrigorian)

New Contributors

• Nikolay Petrov (@napetrov)

Full Changelog: 2.6.1...2.7.0