New features
- Support for Arm NN 24.08.
Public API changes
- None.
Other changes
- Fix a cyclic dependency issue when using depthwise convolution split across channels.
- Align the SRAM buffer allocations to account for PLE address rounding.
- Improve the multiplication PLE kernel for signed inputs.
- Improve the algorithm detecting padding over stripe boundaries.
- Improve the rules determining plans for depthwise convolution in the middle of a cascade.