Examples

The canonical, runnable examples for cuda.bindings live under samples/cuda_bindings/ in the cuda-python repository. Each sample is a self-contained directory with its own README.md, requirements.txt, and PEP 723 dependency block, and every sample is exercised as part of the cuda.bindings test suite.

The cuda-bindings-flavored samples preserve the same category structure the cuda-samples repository uses, so they are grouped under samples/cuda_bindings/0_Introduction/, samples/cuda_bindings/2_Concepts_and_Techniques/, samples/cuda_bindings/3_CUDA_Features/, samples/cuda_bindings/4_CUDA_Libraries/, and samples/cuda_bindings/extra/.

Introduction

• 0_Introduction/clockNvrtc uses NVRTC-compiled CUDA code and the device clock intrinsic to time a reduction kernel; also demonstrates dynamic shared memory.

• 0_Introduction/simpleCubemapTexture demonstrates cubemap texture sampling, cudaMemcpy3D, and bindless texture objects.

• 0_Introduction/systemWideAtomics exercises every atomic*_system intrinsic on managed memory.

• 0_Introduction/vectorAddDriverApi is the “hello world” of the raw driver API in Python.

• 0_Introduction/vectorAddMmap uses the Virtual Memory Management API (cuMemCreate, cuMemMap, cuMemSetAccess) to stripe an allocation across peer-capable devices.

Concepts and techniques

• 2_Concepts_and_Techniques/streamOrderedAllocation demonstrates cudaMallocAsync / cudaFreeAsync together with memory-pool release thresholds.

CUDA features

• 3_CUDA_Features/globalToShmemAsyncCopy compares asynchronous global-to-shared-memory copy strategies (cuda::memcpy_async + cuda::pipeline / cuda::barrier) in matrix-multiplication kernels.

• 3_CUDA_Features/cudaGraphsManualNodes shows both manual CUDA-graph construction (cudaGraphAdd*Node) and stream-capture-based replay.

Libraries and tools

• 4_CUDA_Libraries/conjugateGradientMultiBlockCG implements a conjugate-gradient solver with cooperative groups and multi-block grid synchronization.

• 4_CUDA_Libraries/nvidiaSmi uses NVML to implement a Python subset of nvidia-smi.

Advanced and interoperability

• extra/isoFdModelling runs isotropic finite-difference wave propagation across multiple GPUs with peer-to-peer halo exchange.

• extra/jitProgram JIT-compiles a SAXPY kernel with NVRTC and launches it through the Driver API – the low-level companion to cuda.core’s jitLtoLinking sample.