README

# Chroma Build Package for GPU JIT Architectures

This repository provides automated build scripts to simplify compiling **Chroma** on top of **qdp-jit**. It supports building for AMD (ROCm), NVIDIA (CUDA), and Intel (Level Zero) GPU architectures, automatically handling Chroma and all necessary prerequisites.

## Prerequisites

Before beginning the build process, ensure that you have loaded the appropriate software modules for your cluster or environment (e.g., specific GPU drivers, CMake, and your MPI stack). 


## Quick Start Build Guide

Choose the directory corresponding to your target GPU architecture:
* AMD: `rocm-jit-noquda/`
* NVIDIA: `cuda/`
* Intel: `level0-jit-noquda/`

Follow the three steps below to download, configure, and build.


1. Download the Source Code
---------------------------

From the root of the repository, run the download script specific to your hardware to fetch Chroma and its dependencies:

# For AMD GPUs
./download_sources_rocm.sh

# For NVIDIA GPUs
./download_sources_cuda.sh

# For Intel GPUs
./download_sources_intel.sh



2. Configure the Environment
----------------------------

Change to the appropriate architecture directory and modify the environment configuration script to match your system.

# Navigate to your target architecture directory (e.g., CUDA)
cd cuda/

Open env.sh in your preferred text editor. Verify and update the following variables to ensure they point to your MPI compiler wrappers and correct installation paths:

export CXX=mpicxx
export CC=mpicc

Make sure any other cluster-specific paths or modules are also correctly exported here.



3. Run the Build Script
-----------------------

Once env.sh is configured, initiate the automated build process:

./build_all.sh