That seems like the right approach for CPU-targeted code. Has an APL descendant ever been created to target a GPGPU compute-kernel, or even to compile to an FPGA netlist description of a DSP?
Aaron Hsu’s co-dfns is a compiler of a subset of APL, written in APL, which compiles to GPU code.
Do a HN Algolia search for his username “arcfide” to find a lot of discussion, and there’s a couple of YouTube video talks, one of him talking through the codebase on a livestream, another a recording of a conference style tall introducing it to people.
It needs Dyalog APL, but that’s now easily available for non commercial use.