fkl_python

Add a new FKL operation to the fkl-python front-end (operations.py descriptor, codegen, tests). Use when wrapping a C++ FKL Operation struct for Python, when adding a new op to fkl_python, or when codegen produces invalid C++ for a new op.

Use the fkl-python front-end to compose and run fused GPU kernels from Python with zero overhead. Covers compose(), all operations (arithmetic, casts, vector ops, color conversion, Crop/Resize), Vertical Fusion, Backwards Vertical Fusion, Horizontal Fusion (batch crops), torch/cupy zero-copy interop, and the compile cache. Use when writing Python code that uses the fkl package or when asked how to build a fused pipeline from Python.

Run and write tests for fkl-python (fused GPU kernel front-end). Covers the test harness, per-fusion-mode suites, CPU reference patterns, backend matrix runs (nvcc + clang), and debugging compile failures from cached .cu files. Use when running fkl_python tests, adding tests for a new op, or diagnosing test failures.