API Overview

cuda.zig provides a layered architecture: high-level typed abstractions built on top of low-level runtime and driver bindings, all resolving dynamically with zero link-time dependencies.

Top-Level Namespaces

Module	Description
cuda.device	GPU enumeration, properties, selection, and P2P peer access
cuda.memory	Typed buffers (DeviceBuffer, PinnedBuffer, UnifiedBuffer, PoolBuffer) and raw allocators
cuda.stream	Asynchronous execution streams, priorities, and events
cuda.kernel	Driver kernel launch, LaunchConfig, modules, and cuda.occupancy
cuda.tensor	High-level Tensor(T) with cuBLAS & CPU fallback matrix operations
cuda.nvrtc	Runtime CUDA C++ compilation to PTX via NVRTC
cuda.fallback	Automatic CPU fallback execution engine
cuda.profiler	Profiler session markers (start, stop, ProfilerGuard)
cuda.version	Driver & Runtime version detection and capability flags