The original Jonesforth is very cool, and I'm happy it's being ported to RISC-V. However, the in-code documentation[0] clearly has some errors due to blind find-replace:

  > On the right hand side I've written the actual RISC-V machine code.

... followed by a bunch of 5-byte x86 instructions (E8 followed by 4-byte addresses), and '"E8" is the x86 machine code to "CALL" a function'.

  > On an RISC-V machine it turns out that we can write this interpreter rather easily, in just two assembly instructions which turn into just 3 bytes of machine code.

Uncompressed RISC-V instructions are 4 bytes. Compressed instructions are 2 bytes. Without some completely unanticipated extension that goes against the current design, there will be no way to make a 3-byte instruction sequence for RISC-V. From what I can tell, the actual RISC-V implementation is 4 instructions,

  .macro NEXT
    ld a0, 0(s1)
    addi s1, s1, 8
    ld t0, 0(a0)
    jalr t0
 .endm

taking 16 bytes on machines that don't support the compressed instructions extension. If all 4 can be compressed, that's still 8 bytes.

[0]https://github.com/jjyr/jonesforth_riscv/blob/master/jonesfo...