#cuda #api-calls #gpu

cuda-oxide

cuda-oxide provides a high-level, rusty wrapper over CUDA. It provides the best safety one can get when working with hardware.

7 unstable releases (3 breaking)

0.4.0 Jun 16, 2021
0.3.1 Jun 15, 2021
0.2.1 Jun 11, 2021
0.1.1 Jun 9, 2021

#654 in Machine learning

GPL-3.0-or-later

1MB
23K SLoC

cuda-oxide

cuda-oxide is a safe wrapper for CUDA. With cuda-oxide you can execute and coordinate CUDA kernels.

Safety Philosophy

cuda-oxide does not offer any safety on the GPU-side of writing CUDA code. It doesn't compile Rust to PTX. cuda-oxide offers general CPU-level safety working with the CUDA library and best-availability safety for working with GPU buffers and objects.

Examples of things currently considered safe:

  • Reading from an uninitialized GPU buffer into host memory
  • Some invalid libcuda operations that will cause libcuda to stop accepting any API calls
  • Setting various attributes that can have side effects for an entire device
  • Writing to read-only device memory

Supported Features

  • Device Management
  • Context Management
  • Module Management
  • JIT compilation of Modules
  • Stream Management
  • Kernel Execution
  • Device Memory read/write

Unsupported Features

  • Memory Pools
  • Unified Addressing
  • Events & Stream Events
  • Stream State Polling
  • Stream Graph Capturing
  • Stream Batch Memory Operations
  • External Memory
  • Multi-device helper (possible already, but not made easy)
  • Graphs
  • Textures & Surfaces
  • OpenGL/VDPAU/EGL Interoperability

Examples

See the examples directory for usage examples.

Dependencies

~1.5MB
~33K SLoC