forcebru / pyvm Goto Github PK

This is the code used for testing:

import timeit
from io import StringIO

print("Testing...")
t = timeit.repeat(
        "vm.execute_elf('C/bin/bubblesort.elf');vm.execute_elf('C/bin/quicksort.elf');vm.execute_elf('C/bin/insertionsort.elf');vm.execute_elf('C/bin/memcpy_test.elf')",
        "void=StringIO();vm=VM(0x0017801d, void, void, void)",
        globals={'VM': VM.VM, 'StringIO': StringIO}, number=10, repeat=10)

avg = lambda x: sum(x) / len(x)
print(f"NEW: {min(t):.4f}, {avg(t):.4f}, {max(t):.4f}")
# OLD: 26.6493, 28.4966, 29.9140
# NEW: 28.6566, 29.1709, 29.9998

Here the change was introduced here:

Apparently, this doesn't really speed things up, so it should be removed or re-implemented.

TODO list:

• Update README with directory structure (or probably add READMEs to other directories and link to them from the main one?)
• Add more example C programs (maybe some with floating-point calculations?)
• Comment the code

process_ModRM does not expect any args.

Commit 831d87b breaks debug messages.

Many of them use hex(loc) if type else reg_names[loc][sz] to output either the name of the register or the address in hex. First of all, this is super ugly and should be refactored.

Now that type is either vm.reg or vm.mem, it'll always be truthy, so, hex(loc) will be always evaluated, which is incorrect.

ModR/M byte processing is currently only for 32-bit addressing, and there's a whole table of possible values of the ModR/M byte for 16-bit addressing, which isn't supported at all.

Should probably implement that? Cuz stuff like lea ax, [bx] doesn't work since it's 16-bit addressing.

The movzx instruction attempts to access invalid memory from [ebp - 1]. ebp is loaded with some value from memory before that, and this value is weird. The address seems to depend on the first inputted operand.