Qeravio
Canonical AI event

Metal kernels now split into per-dtype libraries.

The update includes support for Ubuntu s390x and Snapdragon's CPU, Adreno GPU, and Hexagon NPU on Linux arm64.

26 Sept 20261 verified claims1 sources2 observations
What happened

The official source reports this update: b11183: metal : split fa kernels into per-dtype libraries.

Why it matters

This official update documents a development concerning b11183: metal : split fa kernels into per-dtype libraries. Its practical significance depends on the scope and evidence stated by the source.

What to watch next

Read the official source update and verify its stated scope, evidence and timing before acting on it.

Connected knowledge

Entities affected by this event

Llama Cppsubject
Continue this topic
infrastructureLegacy Vulkan build issue resolved with cooperativeMatrix API support added.26 Sept 2026infrastructureMetal FWHT kernels now support block widths up to 8192.26 Sept 2026infrastructureHexagon updates DMA for contiguous dim1 CONCAT.26 Sept 2026infrastructureLlama.cpp updates get_alloc_size cache key and floors result at ggml_nbytes26 Sept 2026infrastructureCUDA kernel fusion reduces kernel launches by 96 per batch on Qwen3.8-27B26 Sept 2026infrastructureLlama.cpp adds q5_k quant type support for Hexagon.26 Sept 2026
Evidence trail

Sources behind the event

llama.cpp official releasesb11183: metal : split fa kernels into per-dtype librariesllama.cpp official releasesb11183: metal : split fa kernels into per-dtype libraries
Editorial presentation

Read the public story

Open story