🏗️ Halide Installation¶
This guide provides clear instructions for obtaining Halide with support for various GPU backends (CUDA, OpenCL, Vulkan, DirectX12) as well as CPU-only configurations.
GPU Backend supported¶
| Backend | Platform | Requirements |
|---|---|---|
| CUDA | NVIDIA GPUs | NVIDIA drivers only, the CUDA Toolkit is NOT required |
| OpenCL | AMD/NVIDIA/Intel | OpenCL runtime (included with GPU drivers) |
| Vulkan | Cross-platform | Vulkan SDK and compatible driver |
| DirectX12 | Windows only | Windows 10/11, compatible GPU |
How to Configure Halide¶
🥇 Method 1 : Pre-built Binary (Recommended & Fastest)¶
The official Halide binaries (CPU + all GPU backends) are distributed as a Python-independent wheel. Python/pip is only used as the downloader.
- Install Python & pip on your machine (Python 3.12+ required).
Via PIP¶
💡 Linux note: on recent distributions, pip refuses to install outside a virtual environment, you can use venv or force install. You can see here [Official docs CMake Halide] (https://github.com/halide/Halide/blob/main/doc/HalideCMakePackage.md)