All Tools
H
Dev ToolsFreeOpen Source
HALIDE
Language for fast portable data-parallel image and tensor kernels
MIT
ABOUT
Hand-tuned SIMD and CUDA kernels couple the algorithm to one hardware target. Halide splits compute from schedule so teams write an image or tensor pipeline once and retarget CPU, GPU, and Hexagon without rewriting the math.
INSTALL
pip install halideINTEGRATION GUIDE
1. Schedule image-processing pipelines for CPU and GPU from one algorithm
2. Generate portable kernels for camera ISP and computational photography
3. Prototype ML operator schedules without rewriting CUDA by hand
4. Compile Halide generators into production C++ or Python runtimes
TAGS
c++pythoncompilersimage-processinggpuschedulingperformance