Build VMC from source

The repository provides build scripts for Linux/macOS shells and Windows PowerShell. This page uses the shell scripts and must be followed from the VMC repository root, where CMakeLists.txt and config.cfg are located.

TODO — UWHPC build environment

Missing from the repository: the supported UWHPC login/build host, exact environment-module names and versions for CMake, GCC/Clang, and CUDA, and any site policy about compiling on login nodes. Replace this block with the UWHPC-wide module and build-host links when that information is available.

Requirements

The build configuration requires:

• CMake 3.24 or newer. CMakeLists.txt enforces 3.24, even though the current README says 3.20.
• A compiler with C++23 support.
• Git when configuring the test build for the first time, because CMake fetches Catch2 from GitHub.
• A CUDA compiler and toolkit only for a CUDA build.

OpenMP is optional. When CMake finds it, the project uses it for SIMD pragmas; otherwise CMake checks for the compiler’s -fopenmp-simd support.

Build the default CPU executable

From the repository root, run:

./scripts/build.sh

The script selects a Release, double-precision, CPU-only build and writes the executable to build/vmc.

The command was run in the documentation workspace. This is its complete output:

-- The CXX compiler identification is GNU 13.3.0
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- OpenMP found (CXX 4.5); using it for '#pragma omp simd'.
-- Configuring done (2.2s)
-- Generating done (0.0s)
-- Build files have been written to: /home/member/docs/Variational-Monte-Carlo/build
[  5%] Building CXX object CMakeFiles/vmc_lib.dir/src/blocking_analysis/blocking_analysis.cpp.o
...
[ 88%] Linking CXX static library libvmc_lib.a
[ 88%] Built target vmc_lib
[ 94%] Building CXX object CMakeFiles/vmc.dir/src/main.cpp.o
[100%] Linking CXX executable vmc
[100%] Built target vmc

Select a build variant

build.sh accepts one positional build type plus three flags:

Command	Result
./scripts/build.sh	CPU, Release, FP64; produces build/vmc
./scripts/build.sh Debug	CPU, Debug, FP64; produces build/vmc
./scripts/build.sh --fp32	CPU, Release, FP32; produces build/vmc
./scripts/build.sh --vmc-fast-math	CPU, Release, FP32 with the project’s fast-math option enabled
./scripts/build.sh --cuda	CUDA, Release, FP64; produces build-cuda/vmc
./scripts/build.sh --cuda --fp32	CUDA, Release, FP32; produces build-cuda/vmc

--vmc-fast-math forces FP32 even if FP64 would otherwise be selected. The shell script explicitly disables CUDA unless --cuda is present.

TODO — verified UWHPC CUDA build

The CUDA variant was not run in this workspace. Missing: the UWHPC GPU model, CUDA module/version, supported compute capability, build node, and a captured successful ./scripts/build.sh --cuda transcript.

Configure manually

The default script is equivalent to configuring the important project options and then building the vmc target:

cmake -S . -B build \
  -DVMC_ENABLE_CUDA=OFF \
  -DFP_64=ON \
  -DVMC_FAST_MATH=OFF \
  -DBUILD_TESTING=OFF \
  -DCMAKE_BUILD_TYPE=Release
cmake --build build --target vmc

Use the scripts unless another tool needs a custom CMake build directory or generator. Continue with Test VMC before running a production calculation.

Source of truth

The behavior documented here comes from CMakeLists.txt, scripts/build.sh, and the observed ./scripts/build.sh run.