递归Fibonacci汇编MIPS代码

Question

I have created a code generator for my coursework in a module compilers. It generates code in MIPS assembly code and seems to be working ok (ive tested very simple programs and expressions). I tested the recursive Fibonacci program and it currently loops forever. The base cases, 0 and 1, work ok. But when I try fib(2) or more it keeps looping. Not sure what the problem is, can anyone help me find it?

Here is the code:

main: 
move $fp $sp 
sw $ra 0($sp)
addiu $sp $sp -4 
sw $fp 0($sp) 
addiu $sp $sp -4 
li $a0 2 #testing comment
sw $a0 0($sp) 
addiu $sp $sp -4 
jal fib_entry 
lw $ra 4($sp) 
addiu $sp $sp 8 
lw $fp 0($sp) 
li $v0, 10 
syscall 
fib_entry: 
move $fp $sp 
sw $ra 0($sp)
addiu $sp $sp -4 
lw $a0 4($fp) 
sw $a0 0($sp) 
addiu $sp $sp -4 
li $a0 0 
lw $t1 4($sp) 
addiu $sp $sp 4 
beq $a0 $t1 thenBranch1
elseBranch1: 
lw $a0 4($fp) 
sw $a0 0($sp) 
addiu $sp $sp -4 
li $a0 1 
lw $t1 4($sp) 
addiu $sp $sp 4 
beq $a0 $t1 thenBranch2
elseBranch2: 
sw $fp 0($sp) 
addiu $sp $sp -4 
lw $a0 4($fp) 
sw $a0 0($sp) 
addiu $sp $sp -4 
li $a0 1 
lw $t1 4($sp) 
sub $a0 $t1 $a0 
addiu $sp $sp 4 
sw $a0 0($sp) 
addiu $sp $sp -4 
jal fib_entry 
sw $a0 0($sp) 
addiu $sp $sp -4 
sw $fp 0($sp) 
addiu $sp $sp -4 
lw $a0 4($fp) 
sw $a0 0($sp) 
addiu $sp $sp -4 
li $a0 2 
lw $t1 4($sp) 
sub $a0 $t1 $a0 
addiu $sp $sp 4 
sw $a0 0($sp) 
addiu $sp $sp -4 
jal fib_entry 
lw $t1 4($sp) 
add $a0 $a0 $t1 
addiu $sp $sp 4 
b endIf2 
thenBranch2: 
li $a0 1 
endIf2: 
b endIf1 
thenBranch1: 
li $a0 0 
endIf1: 
lw $ra 4($sp) 
addiu $sp $sp 12 
lw $fp 0($sp) 
jr $ra 

Answer 1

Various problems with that code.

1. you should be decrementing $sp before you write to ($sp) (although this is only by convention)
2. you should be saving $fp before you modify it (this is by common sense ;))
3. MIPS typically uses registers to pass arguments (but you can of course make up your own convention)
4. your stack handling is horribly overcomplicated, and at least in main it's unbalanced
5. it's recommended to return from main and not use exit syscall

You should of course be able to write and debug asm code before you try to generate some.

Given an input structured something like this C implementation:

unsigned fib_entry(unsigned n) {
    unsigned ret;
    unsigned n1;
    unsigned f1;
    unsigned n2;
    unsigned f2;

    if (n <= 1) goto fib_small;

    n1 = n - 1;
    f1 = fib_entry(n1);
    n2 = n - 2;
    f2 = fib_entry(n2);
    ret = f1 + f2;
    goto fib_done;

fib_small:
    ret = n;

fib_done:
    return ret;
}

I would expect a not-too-clever compiler to produce code like this:

fib_entry:
    addiu $sp $sp -28   # we need room for $ra, n, ret, n1, n2, f1, f2
    sw $ra, 24($sp)     # store $ra since not leaf function
    sw $a0, 20($sp)     # store n

    lw $t0, 20($sp)     # load n
    ble $t0 1 fib_small

    lw $t0, 20($sp)     # load n
    addi $t0 $t0 -1     # n - 1
    sw $t0, 12($sp)     # store as n1

    lw $a0, 12($sp)     # pass n1 as argument
    jal fib_entry
    sw $v0 4($sp)       # store into f1

    lw $t0, 20($sp)     # load n
    addi $t0 $t0 -2     # n - 2
    sw $t0, 8($sp)      # store as n2

    lw $a0, 8($sp)      # pass n2 as argument
    jal fib_entry
    sw $v0 ($sp)        # store into f2

    lw $t0 4($sp)       # f1
    lw $t1 ($sp)        # f2
    addu $t0 $t0 $t1    # f1 + f2
    sw $t0 16($sp)      # store into ret

    b fib_done

fib_small:
    lw $t0, 20($sp)     # load n
    sw $t0 16($sp)      # store into ret

fib_done:
    lw $v0 16($sp)      # load return value
    lw $ra 24($sp)      # restore $ra
    addiu $sp $sp 28    # restore stack
    jr $ra              # return

Note I have deliberately left it unoptimized. This kind of code should be simple enough to generate.