bf_jit VS vemips

I think writing a brainfuck JIT is a great hobby project for learning x86 assembly and the basics of how jit compilation of bytecode works. I also wrote an x64 jit for brainfuck in c++ ( https://github.com/ccapitalK/bf_jit ) , it's surprising how similar it ended up being architectured to the code written in part 2 of this post. It's moderately fast (runs mandelbrot in about 1.1 seconds, and runs towers of hanoi in about 0.2 seconds).

Also, if anyone wants to see a hastily-written ELF loader that I haven't cared enough about to go back and clean up in C++, I have one here.

I suggest MIPS.

I actually wrote VeMIPS for this specific purpose.

Unfortunately, I don't remember specifically. It was about when I wrote VeMIPS and was also working on figuring out an alternative ISA for a 3d printer board, and I'd noticed that a specific conversion instruction was missing on it. Maybe it was moving a double-precision register to the general purpose registers on 32-bit? Not sure anymore. I can look over IRC logs maybe to find out.

This is the exact purpose that I wrote VeMIPS for.

Looks similar to vemips. Same rationale as well (embedding within a game or simulation).

Probably wouldn't be too hard to add Chip8 as a backend to VeMIPS, though VeMIPS presently doesn't run on Android (only Win64 and in Javascript using asm.js).

I personally dislike disassembling MIPS, and I wrote VeMIPS!

However, I did end up releasing VeMIPS, which is an embeddable MIPS32r6 emulator/VM which is heavily configurable, has a dynamic recompiler, and has instruction-level execution granularity, meaning that you can tell it "execute 10 instructions and return". This lets you implement things like cost per instruction, virtual CPUs with different performance characteristics, and so forth.