v24.12 Public Major Release
Feat
- Add a build flag to make scheduler object thread_local and make it default in Bazel build
Fix
- CPU regression in Reshape from excess threads
- NEDeconvolutionLayer regression
- Ensure bias type is BF16 for BF16 indirect convolutions
Perf
- Disable mmul kernel selection for fp16 in GPU backend
Documentation (API, build guide, contribution guide, errata, etc.) available here:
https://artificial-intelligence.sites.arm.com/computelibrary/v24.12/index.xhtml