A BrainF*ck-ish compiler in C

Question

-ish because I removed the input functionality from the compiler. And, the compiler does not support nested loops

I've been recently reading up on compilers and how they work. Although this doesn't use most of the things I've learned about (the lexical and parser things), I thought it would be fun to try create a brainf*ck compiler, rather than an interpreter.

bfc.c

#include <stdio.h>

static int i;
static char *code; // these two as static so get_amt_to_change can easily interact

int get_amt_to_change(char c); // so the compiler doesn't write (ie) add di, 1 ten times

int main(int argc, char **argv) {
    code = argv[1];
    puts("xor di, di\n"
         "setup_loop:\n"
         "mov byte [tape + di], 0\n"
         "add di, 1\n"
         "cmp di, 101\n"
         "jne setup_loop\n"
         "xor di, di"); // sets up the tape with all 0's
    int loop_count = 0; // to keep track of asm subroutines for [ and ]
    for(i = 0; code[i] != '\0'; i++) {
        switch(code[i]) {
            case '+':
                printf("add byte [tape + di], %d\n", get_amt_to_change('+'));
                break;
            case '-':
                printf("sub byte [tape + di], %d\n", get_amt_to_change('-'));
                break;
            case '>':
                printf("add di, %d\n", get_amt_to_change('>'));
                break;
            case '<':
                printf("sub di, %d\n", get_amt_to_change('<'));
                break;
            case '.':
                puts("mov ah, 0Eh\n"
                     "mov al, byte [tape +di]\n"
                     "int 10h");
                break;
            case '[':
                printf("cmp byte [tape + di], 0\n"
                       "je end_loop%d\n"
                       "start_loop%d:\n", loop_count, loop_count);
                break;
            case ']':
                printf("cmp byte [tape + di], 0\n"
                       "jne start_loop%d\n"
                       "end_loop%d:\n", loop_count, loop_count);
                loop_count++; // to not repeat subroutine names
                break;
        }
    }
    puts("jmp $\nsection .bss\ntape resb 100"); // a 100 byte tape
    return 0;
}

int get_amt_to_change(char c) {
    int amt;
    for(amt = 0; code[i] == c; amt++, i++);
    i--; // if it wasn't == c, then go back one character and find out what it was equal to in the next call
    return amt;
}

The above code reads Brainf*ck code passed via command line argument and produces an assembly code, which could then be passed into the NASM assembler.

Examples

Purpose: nothing. brainf*ck doesn't have a purpose logs 'd' to output

Brainf*ck

Note: the backslashes are there so the terminal doesn't read the < and > symbols as it normally would.

\>++++++++++[\<++++++++++\>-]\<.

Assembly

xor di, di
setup_loop:
mov byte [tape + di], 0
add di, 1
cmp di, 101
jne setup_loop
xor di, di
add di, 1
add byte [tape + di], 10
cmp byte [tape + di], 0
je end_loop0
start_loop0:
sub di, 1
add byte [tape + di], 10
add di, 1
sub byte [tape + di], 1
cmp byte [tape + di], 0
jne start_loop0
end_loop0:
sub di, 1
mov ah, 0Eh
mov al, byte [tape +di]
int 10h
jmp $
section .bss
tape resb 100

---

Purpose: to show the generated assembly code for each of the symbols

Brainf*ck

++--\>\>\<\<.,[]

Assembly

xor di, di
setup_loop:
mov byte [tape + di], 0
add di, 1
cmp di, 101
jne setup_loop
xor di, di
add byte [tape + di], 2
sub byte [tape + di], 2
add di, 2
sub di, 2
mov ah, 0Eh
mov al, byte [tape +di]
int 10h
cmp byte [tape + di], 0
je end_loop0
start_loop0:
cmp byte [tape + di], 0
jne start_loop0
end_loop0:
jmp $
section .bss
tape resb 100

Questions

• My C code only has one subroutine. Are there any other logical ones to add?

• I did a lot of optimizing on the assembly code output. Could it be further optimized?

• My main problem with compilers I've written in the past is that I've over-complicated things. Is that an issue with this (C) code?

Answer 1

You need to enlarge tape because your initialization zeroes 101 bytes.

 section .bss
 tape resb 101

When in graphics mode, the teletype function uses the BL and BH registers as arguments.

 mov  bx,0007h ;Display page 0 and Color 7
 mov  ah, 0Eh
 mov  al, byte [tape +di]
 int  10h

When in text mode, the teletype function uses the BH register as an argument.

 mov  bh,0     ;Display page 0
 mov  ah, 0Eh
 mov  al, byte [tape +di]
 int  10h

You can optimize the setup code by iterating backwards. It shaves off 2 instructions! It too will leave DI=0.

 mov  di, 101
setup_loop:
 sub di,1
 mov  byte [tape + di], 0
 jnz  setup_loop

I don't know if you would care but the code for [ ] by itself produces an infinite loop if the byte at [tape + di] is anything but zero.