Expand description
§ruda-tensor-host
A fast, portable CPU tensor backend for Ruda.
§Features
- Pure Rust (no C dependencies)
- f16/bf16 support
- SIMD acceleration via macerator (NEON, AVX2/AVX-512/SSE, SIMD128, scalar fallback)
- Zero-copy tensor views
- Thread-safe by design
§Usage
ⓘ
use ruda_tensor_host::Host;
use ruda_tensor::api::Tensor;
let tensor: Tensor<Host, 2> = Tensor::from_data([[1.0, 2.0], [3.0, 4.0]], &Default::default());Structs§
- Host
- The Host backend, a fast, portable CPU backend for Ruda.
- Host
Device - CPU device for the Host backend.
- HostQ
Tensor - Quantized tensor for the Flex backend.
- Host
Tensor - CPU tensor primitive for the Flex backend.
- Layout
- Layout describes how to interpret a linear buffer as an N-dimensional tensor.