Skip to content

Latest commit

 

History

History
17 lines (14 loc) · 612 Bytes

README.md

File metadata and controls

17 lines (14 loc) · 612 Bytes
Raw

cuda wrapper

This repository contains code for generating a shared library that intercepts calls to the NVIVIA CUDA runtime library.

Quickstart:

mkdir build
cmake ..
make
# edit stubs
make

Components:

  1. make_stubs.py generates CUDA function stubs, for you to fill in with custom functionality
  2. wrap_generate.py combines stubs into a shim library

Notes:

Compile your CUDA program to dynamically link the CUDA library, not statically. This is achieved with the --cudart shared flag to nvcc. To use the shim library: LD_PRELOAD=/path/to/cuda_wrapper.so ./cuda_program