Why should EDX be 0 before using the DIV instruction?

Question

I noticed when EDX contains some random default value like 00401000, and I then use a DIV instruction like this:

mov eax,10
mov ebx,5
div ebx

Answer 1

What to do

For 32-bit / 32-bit => 32-bit division: zero- or sign-extend the 32-bit dividend from EAX into 64-bit EDX:EAX.
For 16-bit, AX into DX:AX with cwd or xor-zeroing.

• unsigned: XOR EDX,EDX then DIV divisor
• signed: CDQ then IDIV divisor

See also When and why do we sign extend and use cdq with mul/div?

---

Why (TL;DR)

For DIV, the registers EDX and EAX form one single 64 bit value (often shown as EDX:EAX), which is then divided, in this case, by EBX.

So if EAX = 10 or hex A and EDX is, say 20 or hex 14, then together they form the 64 bit value hex 14 0000 000A or decimal 85899345930. If this is divided by 5, the result is 17179869186 or hex
4 0000 0002, which is a value that does not fit in 32 bits.

That is why you get an integer overflow.

If, however, EDX were only 1, you would divide hex 1 0000 000A by 5, which results in hex
3333 3335. That is not the value you wanted, but it does not cause an integer overflow.

To really divide 32 bit register EAX by another 32 bit register, take care that the top of the 64 bit value formed by EDX:EAX is 0.

So, before a single division, you should generally set EDX to 0.

(Or for signed division, cdq to sign extend EAX into EDX:EAX before idiv)

---

But EDX does not have always have to be 0. It can just not be that big that the result causes an overflow.

One example from my BigInteger code:

After a division with DIV, the quotient is in EAX and the remainder is in EDX. To divide something like a BigInteger, which consists of an array of many DWORDS, by 10 (for instance to convert the value to a decimal string), you do something like the following:

    ; ECX contains number of "limbs" (DWORDs) to divide by 10
    XOR     EDX,EDX      ; before start of loop, set EDX to 0
    MOV     EBX,10
    LEA     ESI,[EDI + 4*ECX - 4] ; now points to top element of array
@DivLoop:
    MOV     EAX,[ESI]
    DIV     EBX          ; divide EDX:EAX by EBX. After that,
                         ; quotient in EAX, remainder in EDX
    MOV     [ESI],EAX
    SUB     ESI,4        ; remainder in EDX is re-used as top DWORD... 
    DEC     ECX          ; ... for the next iteration, and is NOT set to 0.
    JNE     @DivLoop

After that loop, the value represented by the entire array (i.e. by the BigInteger) is divided by 10, and EDX contains the remainder of that division.

FWIW, in the assembler I use (Delphi's built-in assembler), labels starting with @ are local to the function, i.e. they don't interfere with equally named labels in other functions.