Qeravio
Canonical AI event

CUDA FWHT now supports F16 input type.

The CUDA FWHT now supports F16 input directly, eliminating the need for conversion.

27 Sept 20261 verified claims1 sources1 observations
What happened

The official source reports this update: b11203: cuda: add F16 input to the FWHT. cuda: add F16 input to the FWHT ( #29096 ) cuda: add F16 input to the FWHT The CUDA FWHT accepts F32 input only. This makes the source type a template parameter, so the kernel reads an F16 source directly instead of requiring a converted copy. The F32 path is unchanged. supports_op accepts an F16 src1 against an F32 src0 for the Hadamard hint. Every other F16 src1 against a non-F16 src0 is still refused. ggml_cuda_op_mul_mat_use_fwht is the single predicate both supports_op and the dispatch call now share, checking contiguity and same-shape(src1, dst) in addition to the type/hint conditions above.

Why it matters

This official update documents a development concerning b11203: cuda: add F16 input to the FWHT. 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
infrastructureWindows wake_fd warning fixed in server software27 Sept 2026infrastructureLlama.cpp adds sameas test for multiple platforms27 Sept 2026infrastructureJinja fixes compile error in llama.cpp across multiple platforms.27 Sept 2026infrastructureLlama.cpp adds A8 Q8_0 non-MoE dp4a binary kernel for OpenCL27 Sept 2026infrastructureLegacy Vulkan build issue resolved with cooperativeMatrix API support added.26 Sept 2026infrastructureMetal FWHT kernels now support block widths up to 8192.26 Sept 2026
Evidence trail

Sources behind the event

llama.cpp official releasesb11203: cuda: add F16 input to the FWHT
Editorial presentation

Read the public story

Open story