AMD 250 manual Bit Unsigned Remainder Computation, 174

Software Optimization Guide for AMD64 Processors 25112 Rev. 3.06 September 2005

div	ebx		; EAX = quotient_lo
mov	edx,	ecx	; EDX = quotient_hi (quotient in EDX:EAX)
jmp	make_sign		; Make quotient signed.
big_divisor:
sub	esp,	12		; Create three local variables.
mov	[esp], eax			; dividend_lo
mov	[esp+4], ebx			; divisor_lo
mov	[esp+8], edx			; dividend_hi
mov	edi,	ecx		; Save divisor_hi.
shr	edx,	1		; Shift both
rcr	eax,	1		;	divisor and
ror	edi,	1		;	and dividend
rcr	ebx,	1		;	right by 1 bit.
bsr	ecx,	ecx		; ECX = number of remaining shifts
shrd	ebx,	edi, cl		; Scale down divisor and
shrd	eax,	edx, cl		; dividend such that divisor is
shr	edx,	cl		; less than 2^32 (that is, fits in EBX).
rol	edi,	1		; Restore original divisor_hi.
div	ebx			; Compute quotient.
mov	ebx,	[esp]		; dividend_lo
mov	ecx,	eax		; Save quotient.
imul	edi,	eax		; quotient * divisor high word (low only)
mul	DWORD PTR [esp+4]			; quotient * divisor low word
add	edx,	edi		; EDX:EAX = quotient * divisor
sub	ebx,	eax		; dividend_lo - (quot.*divisor)_lo
mov	eax,	ecx		; Get quotient.
mov	ecx,	[esp+8]		; dividend_hi
sbb	ecx,	edx		; Subtract (divisor * quot.) from dividend
sbb	eax,	0		; Adjust quotient if remainder is negative.
xor	edx,	edx		; Clear high word of quotient.
add	esp,	12		; Remove local variables.
make_sign:
xor eax, esi			; If (quotient < 0),
xor edx, esi			; compute 1's complement of result.
sub eax, esi			; If (quotient < 0),
sbb edx, esi			; compute 2's complement of result.
pop edi			; Restore	EDI as per calling convention.
pop esi			; Restore	ESI as per calling convention.
pop ebx			; Restore	EBX as per calling convention.
ret			; Done, return to caller.
_lldiv ENDP

64-Bit Unsigned Remainder Computation

;_ullrem divides two unsigned 64-bit integers and returns the remainder.

; In: [ESP+8]:[ESP+4] = dividend

;[ESP+16]:[ESP+12] = divisor

; Out: EDX:EAX = remainder of division

174 Integer Optimizations Chapter 8