Login
You're viewing the techdon.dev public feed.
  • Nov 4, 2026, 5:44 PM

    At a cafe yesterday I decided to try something completely different, and sketch out a microarchitecture implementing the UXN instruction set on an FPGA. After a couple hours of analysis, I reached an unexpected conclusion:

    While these terms are not often used for stack machines, I think there's a strong argument that UXN is "CISC" to my Forth machine's "RISC." This is not intended as a criticism!

    Individual UXN instructions can touch up to twelve stack locations (ROT2k). Device accesses (in Varvara) are blocking and stop the instruction flow for complex operations like traversing a filesystem. An implementation of reasonable complexity would almost certainly want to be microcoded, and device operations would either need to trap into something like the Alpha's PALcode (though the instruction space is fully packed, so it might need to change instruction sets) or invoke an auxiliary core.

    Of course, UXN is designed to be emulated in software, not hardware, and when writing an emulator, it's often useful to have instructions be bigger and more complex to reduce dispatch overhead. This is why my comments shouldn't be taken as criticism -- this property of UXN is fit for purpose. But I suspect the easiest way to implement it in a CPU is to build a specialized CPU running an emulator for it.

    #uxn

    💬 0🔄 0⭐ 6