Incorrect iteration of loop in Assembler

Question

Using nasm with 32bit assembler code yields an unexpected result for my current attempts to write a loop which basically swaps elements of a vector. Given that ESI and EDI point to two different vectors (storing double values) and ECX the number n describing to swap the first n elements from [ESI] with the first n elements from [EDI] my attempts so far look like this:

; c-call: dswap(int n, double* dx, int unused1, double* dy, int unused2)
_dswap:
    push ebp
    mov ebp, esp
    pusha
    mov esi, [ebp+12] ; dx
    mov edi, [ebp+20] ; dy
    mov ecx, [ebp+8]  ; n
    ; unused1 and unused2 obviously unused for the moment
mainLoop:
    cmp ecx, 0
    je exitFunction
    mov eax, [esi + 4 * ecx]
    mov ebx, [edi + 4 * ecx]
    mov [esi + 4 * ecx], ebx
    mov [edi + 4 * ecx], eax
    dec ecx
    jmp mainLoop
exitFunction:
    popa
    mov esp, ebp
    pop ebp
    ret

I am getting some unexpected behavior. Calling dswap on (1,2,3) with (4,5,6) and n=3 only swaps the first two elements in both vectors thus making me think what I did wrong here.

jakobbotsch · Accepted Answer · 2014-06-21 13:28:16Z

2

You are dealing with doubles, but you are only multiplying ecx by 4 (the size of a float). Since the size of a double is 8, you should multiply by 8.

Another problem is that you do not decrement ecx before you multiply by it. Assuming n is passed as 3, you will be swapping dx[3] with dy[3] on the first iteration, but that is beyond the ends of the arrays. To fix this, you can decrement ecx before you do the swaps.

answered Jun 21, 2014 at 13:28

jakobbotsch

6,3974 gold badges29 silver badges39 bronze badges

Sign up to request clarification or add additional context in comments.

5 Comments

Christian Ivicevic Over a year ago

After applying your proposed changes my vectors aren't modified at all. They remain unchanged now.

jakobbotsch Over a year ago

Since a double is 8 bytes you need 4 32-bit swaps now. 2 is not sufficient.

Christian Ivicevic Over a year ago

Okay that will take my some time to figure out how to implement this.

jakobbotsch Over a year ago

Any particular reason you are writing it in Assembly? If it is because of speed, then it is likely that your version is less efficient than what a compiler would generate.

Christian Ivicevic Over a year ago

Our assignment specifically states that this function should be implemented in Assembler. And thanks for the hint with the additional swaps - it works now as intended.

Giuseppe Guerrini · Accepted Answer · 2014-06-21 13:30:42Z

0

Double precision values are usually stored in 8-bytes cells. In your loop youe are assuming a 4-bytes cell, that's why only half of array is swapped.

answered Jun 21, 2014 at 13:30

Giuseppe Guerrini

4,45821 silver badges35 bronze badges

Collectives™ on Stack Overflow

Incorrect iteration of loop in Assembler

2 Answers 2

5 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

5 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related