Skip to content

[FEATURE] Fused NEON WaveNet engine for AArch64 (Apple Silicon) #657

Description

@rikkus

There is an opportunity to reduce further the CPU load of the models on Apple Silicon.

This PR against Core provides a material speedup.

Tested on MacBook Air M2 and iPhone 17.
On M2, a2_fast benchmarks at 29x RT and 'fused' at 50x RT
Non-scientific testing with REAPER yielded around ~1.27× lower load

PR #656 against this Plugin repo adds flags to enable that Core optimisation.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestpriority:lowLow priority issuesunreadThis issue is new and hasn't been seen by the maintainers yet

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions