Skip to main content

Crate product_os_compute

Crate product_os_compute 

Source
Expand description

§Product OS Compute

Portable compute runtime for Product OS with a stable facade over:

  • shared runtime traits and DeviceBufferF32
  • the implemented CPU SIMD backend
  • optional CUDA and Metal backend crates
  • reserved Phase 2 stub backends for AMD, TPU, and NVPTX

Implemented backends:

  • backend_cpu_simd
  • backend_cuda
  • backend_metal

Reserved stub feature flags:

  • backend_amd
  • backend_tpu
  • backend_nvptx

CPU-only builds do not compile GPU bindgen / native shim build dependencies.

Re-exports§

pub use device_buffer::BlasLevel3BuffersF32;
pub use device_buffer::DeviceBufferF32;
pub use device_buffer::DeviceMemF32;
pub use device_buffer::MatrixLayoutDeviceF32;
pub use device_buffer::MatrixLayoutDeviceF32Mut;
pub use backend::cpu_simd::CpuSimdDevice;

Modules§

backend
Backend namespaces. Concrete compute backend implementations.
device_buffer
Pre-allocated f32 storage for repeated host↔device transfers and in-place GEMM.
runtime
Portable compute runtime: shared errors, device metadata, and traits.