Instead of writing a traditional 6502/8501/Z80/etc. backend for general-purpose IRs like LLVM, I wonder whether these platforms are constrained enough that a superoptimizer could generate more optimal code in reasonable time
-
Sorry I was being too specific. The general point was back when you had so little memory and no paging so the editor and assembler had to fit into memory WITH the source you were editing, it really paid off to have small source files.
