Skip to content

Latest commit

 

History

History
86 lines (67 loc) · 3.62 KB

Developing.md

File metadata and controls

86 lines (67 loc) · 3.62 KB

Getting Started with Developing CUDA Quantum

This document contains guidelines for contributing to the code in this repository. This document is relevant primarily for contributions to the nvq++ compiler, the CUDA Quantum runtime, or to the integrated simulation backends. If you would like to contribute applications and examples that use the CUDA Quantum platform, please follow the instructions for installing CUDA Quantum instead.

Quick start guide

Before getting started with development, please create a fork of this repository if you haven't done so already and make sure to check out the latest version on the main branch. After following the instruction for setting up your development environment and building CUDA Quantum from source, you should be able to confirm that you can run the tests and examples using your local build. If you edit this file to add a print statement

std::cout << "Custom registration of " << #NAME << "\n" << std::endl;

to the definition of the NVQIR_REGISTER_SIMULATOR macro, you should see this line printed when you build the code and run an example using the command

bash "$CUDAQ_REPO_ROOT/scripts/build_cudaq.sh" && \
nvq++ "$CUDAQ_REPO_ROOT/docs/sphinx/examples/cpp/algorithms/grover.cpp" -o grover.out && \
./grover.out

When working on compiler internals, it can be useful to look at intermediate representations for CUDA Quantum kernels.

To see how the kernels in this example are translated, you can run

cudaq-quake $CUDAQ_REPO_ROOT/docs/sphinx/examples/cpp/algorithms/grover.cpp

to see its representation in the Quake MLIR dialect. To see its translation to QIR, you can run

cudaq-quake $CUDAQ_REPO_ROOT/docs/sphinx/examples/cpp/algorithms/grover.cpp |
cudaq-opt --canonicalize --add-dealloc |
quake-translate --convert-to=qir

Code style

With regards to code format and style, we distinguish public APIs and CUDA Quantum internals. Public APIs should follow the style guide of the respective language, specifically this guide for C++, and the this guide for Python. The CUDA Quantum internals on the other hand follow the MLIR/LLVM style guide. Please ensure that your code includes comprehensive doc comments as well as a comment at the top of the file to indicating its purpose.

Testing and debugging

CUDA Quantum tests are categorized as unit tests on runtime library code and FileCheck tests on compiler code output. All code added under the runtime libraries should have an accompanying test added to the appropriate spot in the unittests folder. All code that directly impacts compiler code should have an accompanying FileCheck test. These tests are located in the test folder.

When running a CUDA Quantum executable locally, the verbosity of the output can be configured by setting the CUDAQ_LOG_LEVEL environment variable. Setting its value to info will enable printing of informational messages, and setting its value to trace generates additional messages to trace the execution. These logs can be directed to a file by setting the CUDAQ_LOG_FILE variable when invoking the executable, e.g.

CUDAQ_LOG_FILE=grover_log.txt CUDAQ_LOG_LEVEL=info grover.out