compute_fft_stage

One stage of a Stockham auto-sort FFT over vec2<f32> (re, im) data. Used by nx.gpu.fft1d / nx.gpu.fft2d, invoked log2(N) times per pass (1D = one pass, 2D = row then col), ping-ponging input/output between stages. Generic over a "line" abstraction so the same shader handles 1D and both 2D passes.

Bindings (see bindings.yaml): input (read), output (read_write) — both array<vec2<f32>> — and params (read array<f32>).

Usage: dispatch by identity with compute.dispatch("@builtin::shaders.compute_fft_stage", ...) — it resolves to the shader's stable guid and auto-compiles on first use (no setup).