All examples are shipped within libmathdx package.
This folder demonstrates libmathdx APIs usage. libmathdx is a runtime library for MathDx enabling JIT compilation of MathDx operations. libmathdx exposes a simple C-API interface for ease of adoption in non-C++ projects, and is the backend powering NVIDIA Python projects such as Warp and nvmath-python.
- CMake 3.31 or newer
- Linux or Windows system with installed NVIDIA drivers
- NVIDIA GPU of Turing (SM75) or newer architecture
- CUDA Toolkit 13.X
- The examples use
find_package(libmathdx CONFIG REQUIRED), which consumes thelibmathdx-config.cmakeshipped underlib/cmake/libmathdx/in the libmathdx package. Point CMake at the package with-DCMAKE_PREFIX_PATH=<libmathdx_path>or-Dlibmathdx_ROOT=<libmathdx_path>.
Linux build:
mkdir build && cd build
cmake .. # or cmake -Dlibmathdx_ROOT=<libmathdx_path> ..
make
# Run
LD_LIBRARY_PATH=<cuda_root_directory>/lib64 ctest --parallel 8Windows build:
mkdir build && cd build
cmake .. # or cmake -Dlibmathdx_ROOT=<libmathdx_path> ..
cmake --build . --config Release
REM Run
set PATH=%PATH%;<libmathdx_dll_path>;<cuda_library_path>
ctest -C Release --parallel 8Python wheels: libmathdx is also shipped as downloadable wheels on PyPI:
pip install nvidia-libmathdx-cu13Optional environment variables:
LIBMATHDX_EXAMPLE_VERBOSE- print generated kernel source to stdoutLIBMATHDX_EXAMPLE_DUMP_FILENAME=<path>- write JITed intermediates to a file
See the libmathdx documentation for more details.
For the detailed descriptions of the examples please visit the Examples section of the libmathdx documentation.
| Group | Example | Description |
|---|---|---|
| cuFFTDx Examples | cufftdx | cuFFTDx API introduction example, performs a C2C FFT |
| cufftdx_heuristics | Performs a C2C FFT with queryable tuning knobs | |
| cuBLASDx Examples | cublasdx | cuBLASDx API introduction example, performs a GEMM |
| cublasdx_trsm | Performs a triangular matrix solve using cuBLASDx | |
| cublasdx_tensor | Performs a GEMM with the Opaque Tensor API | |
| cublasdx_pipeline | Performs a GEMM using the pipelining APIs | |
| cublasdx_pipelining_callback | Performs a GEMM using pipelining/TMA instructions with a user callback function | |
| cublasdx_pipeline_rmem_copy | Pipelined GEMM that adds a bias to the output in registers before storing it | |
| cuSolverDx Examples | cusolverdx_trsm | cuSolverDx API introduction example, performs a triangular matrix solve |
| cusolverdx_potrf | Performs Cholesky factorization | |
| cusolverdx_potrs | Performs a linear solve using Cholesky factors | |
| cuRANDDx Examples | curanddx | cuRANDDx API introduction example, generates random numbers |
| nvCOMPDx Examples | nvcompdx | nvCOMPDx API introduction example, builds an LZ4 compress device function |
| nvcompdx_batch | End-to-end LZ4 compress/decompress round-trip across a batch of chunks |