CUTLASS Examples

Introduction

The CUDA kernel examples using CUTLASS and CuTe abstractions.

Examples

• CuTe Matrix Transpose
• CuTe Vector Copy
• CuTe General Matrix Multiplication

Usages

To download the CUTLASS-Examples repository, please run the following command.

$ git clone --recursive https://github.com/leimao/CUTLASS-Examples
$ cd CUTLASS-Examples
# If you are updating the submodules of an existing checkout.
$ git submodule sync
$ git submodule update --init --recursive

CUTLASS Docker Container

Docker is used to build and run CUTLASS CUDA kernels. The custom Docker container is built based on the NVIDIA NGC CUDA 12.4.1 Docker container.

Please adjust the base Docker container CUDA version if the host computer has a different CUDA version. Otherwise, weird compilation errors and runtime errors may occur.

Build Docker Images

To build the custom Docker image, please run the following command.

$ docker build -f docker/cuda.Dockerfile --no-cache --tag cuda:12.4.1 .

Run Docker Container

To run the custom Docker container, please run the following command.

$ docker run -it --rm --gpus device=0 -v $(pwd):/mnt -w /mnt cuda:12.4.1

To run the custom Docker container with NVIDIA Nsight Compute, please run the following command.

$ xhost +
$ docker run -it --rm --gpus device=0 -v $(pwd):/mnt -w /mnt -e DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix --cap-add=SYS_ADMIN --security-opt seccomp=unconfined --network host cuda:12.4.1
$ xhost -

CUTLASS CMake Examples

Build Examples

To build the CUDA kernels, please run the following commands.

$ export NUM_CMAKE_JOBS=4
$ cmake -B build
$ cmake --build build --config Release --parallel ${NUM_CMAKE_JOBS}

Run Example Unit Tests

To run the unit tests, please run the following command.

$ ctest --test-dir build/ --tests-regex "Test.*" --verbose

Run Example Performance Measurements

To run the performance measurements, please run the following command.

$ ctest --test-dir build/ --tests-regex "Profile.*" --verbose

Performance measurements will run selected CUDA kernels for large problems multiple times and therefore might take a long time to complete.

References

• CuTe Layout Algebra
• Build and Develop CUTLASS CUDA Kernels