PoC: fused FFT stages (radix-2^2)#103

Draft

Shnatsel wants to merge 10 commits intomainfrom

Commits on Apr 2, 2026

Add narrow variants of dit_chunk_n_simd to reduce register pressure; prep for refactoring into radix-2^2
Shnatsel
committed
Initial implementation of radix-2^2-like fused kernel, to reduce the required amount of passes over memory
Shnatsel
committed
Correctly wire up fused kernels to the recursive algorithm
Shnatsel
committed

Commits on Apr 11, 2026