MPCGPU CUDA to HIP Porting

This repository contains setup notes, small experiments, and later the MPCGPU CUDA → HIP porting work.

CUDA setup and basic programs

To start working with CUDA, look at:

docs/setup_cuda_wsl.md      # setup and compile/run
experiments/000_cuda_basics/ # real working CUDA examples

HIP setup and HIPIFY

To start working with HIP, look at:

docs/setup_hip_wsl.md        # HIP setup, HIPIFY, compile/run
experiments/000_cuda_basics/ # HIPIFY-generated HIP examples

ROCm AMD backend on WSL

To run HIP code on the AMD GPU/iGPU from WSL, look at:

docs/setup_rocm_amd_wsl.md # AMD ROCm/ROCDXG setup and AMD HIP compile/run

#_amd in ./bin compiled files mean that it was HIP compiled for AMD backend

MPCGPU CUDA/HIP feature tests

To test concrete CUDA API calls, CUDA language features, and their HIPIFY translation, look at:

experiments/001_api_calls_mpcgpu/                # isolated CUDA/HIP feature translation tests
experiments/001_api_calls_mpcgpu/api_collection.md # tested calls/features and results

This folder is for small isolated tests before touching the real MPCGPU code.

Docker setup

We have decided to provide Dockerfiles for configuring both NVIDIA/AMD environment.

BUT:

Docker can package compilers, HIPIFY, ROCm/CUDA userspace, Makefiles, dependencies. Docker CANNOT package the Windows GPU driver, WSL kernel, or /dev/dxg GPU bridge.

Project plan and work separation

The detailed porting plan, ownership split, and scan findings are in:

docs/project_plan/mpcgpu_overview_plan.md              # main detailed project plan
docs/project_plan/final_table_separation.md            # concrete test/feature ownership table
docs/project_plan/tobias_grep_findings/                # grep/static-scan findings for CUDA usage