Run time error if last array value is negative

Question

I have this short piece of code, its purpose is to sum negative values and count positive ones, the vector is initialized else where

;sum all negative values and count positive values 
    ;let eax hold sum and ebx hold number of positives 
    CALL crlf
    mov ecx, VectorSize
    mov eax, 0
    mov ebx, 0
    mov edx, 0
    sumMe: 
        cmp Vector[edx * TYPE Vector], 0
        jge pos
        ;neg
        add eax, Vector[edx * TYPE Vector]
        INC edx
        loop sumMe 
        pos:
            INC ebx 
            INC edx
            loop sumMe

The piece of code works correctly except when a negative value is the last value in the array, even something like -1 -1 -1 -1 5 works. However if I have a negative value in the last location the program crashes.

Any help would be appreciated

With the code you've shown, there's nothing after the loop so your program will always crash when it decodes/executes whatever garbage bytes are next when it eventually leaves the loop. This isn't a [mcve]. — Peter Cordes, Feb 09 '18 at 03:26

score 3 · Accepted Answer · answered Feb 09 '18 at 03:30

3

Assuming there's a ret or something at the end:

What does loop do when ecx reaches zero: execution falls through. If the last element was negative, the loop instruction that makes ecx zero is followed by pos:.

When loop runs with ecx=0 to start with, ecx wraps to 0xFFFFFFFF and the jump is taken. loop is like the bottom of a do{}while(--ecx);.

All of this would be obvious if you used a debugger to single-step that last iteration, because you already know the problem is at the end of the array with a negative element.

answered Feb 09 '18 at 03:30

Peter Cordes

328,167
45
605
847

Changing out the first `loop` to an unconditional jump to the second `loop` did the trick, thanks – Mitch Feb 09 '18 at 03:39
@Mitchel0022: That would make it a bit more like an if/else, but the loop tail-duplication you had before was more efficient. (Although [`loop` itself is slow](https://stackoverflow.com/questions/35742570/why-is-the-loop-instruction-slow-couldnt-intel-have-implemented-it-efficiently) and should generally be avoided). The simple fix is to put a `jmp` *after* the first `loop`, so that exit-from-loop path goes where you want, without adding any instructions inside the loop. Of course, there are plenty of other efficiency improvements I could suggest, like branchless or one load... – Peter Cordes Feb 09 '18 at 03:47

Run time error if last array value is negative

1 Answers1