knazarov.com/content/posts/disassembling_functions/note.md

X-Date: 2024-10-06T23:01:40Z X-Note-Id: 710a16d0-f850-4d9b-b054-0e6b1c94b1c1 Subject: Disassembling functions X-Slug: disassembling_functions

In the latest patch to Valeri, I've added support for disassembling arbitrary functions. It can be done both in the REPL and in file execution mode.

When you disassemble a function, you get back the human-readable virtual machine instructions that can be used to see if code generation has been done correctly. This of course can't compare with a fully-featured debugger, but is enough to iron out the basics.

Anyway, here's an example:

;; Function that calculates n!
(fn fact (n)
    (if (<= n 0)
        1
        (* n (fact (- n 1)))))

;; Output the VM bytecode of the function
(println (disassemble fact))

When executed, this program outputs:

constants:
  c0: true
  c1: false
  c2: 0
  c3: 1
  c4: nil

code:
  mov r1 c0
  mov r2 r0
  mov r3 c2
  less-equal r2 r3 0
  mov r1 c1
  equal r1 c0 0
  jump 3
  mov r2 c3
  jump 8
  mov r2 r0
  mov r3 c4
  mov r4 r0
  mov r5 c3
  sub r4 r4 r5
  selfcall r3 r5
  mul r2 r2 r3
  mov r1 r2
  mov r0 r1
  ret r0

As you can see, there are constants and code sections. This is because every function in Valeri is self-contained and the code is not "glued" together with other functions.

Most of the opcodes have their first parameter as the accumulator (destination) and the rest as parameters. For example, mov r1 c0 moves the value of constant c0 into register r1. And sub r4 r4 r5 subtracts value of r5 from r4 and puts the result back to r4.

If you carefully study the opcodes, you'll notice that there are many redundant operations that can be eliminated. This is true, and mostly due to the code generator being naive. In the future when I'll get to optimizations it should be possible to reduce the size of the bytecode.