AMD 250 manual 177

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

sub	eax,	esi	; If (remainder < 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.
sr_two_divs:
mov ecx, eax			; Save	dividend_lo in ECX.
mov eax, edx			; Get dividend_hi.
xor edx, edx			; Zero-extend it into EDX:EAX.
div ebx			; EAX = quotient_hi, EDX = intermediate remainder
mov eax, ecx			; EAX = dividend_lo
div	ebx		; EAX = quotient_lo
mov	eax,	edx	; remainder_lo
xor	edx,	edx	; remainder_hi = 0
jmp	sr_makesign		; Make	remainder signed.
sr_big_divisor:
sub	esp,	16	;	Create three local variables.
mov	[esp], eax		;	dividend_lo
mov	[esp+4], ebx		;	divisor_lo
mov	[esp+8], edx		;	dividend_hi
mov	[esp+12], ecx		;	divisor_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	ecx,	[esp+8]	;	dividend_hi
sbb	ecx,	edx	;	Subtract divisor * quot. from dividend.
sbb	eax,	eax	;	remainder < 0 ? 0xffffffff : 0
mov	edx,	[esp+12]	;	divisor_hi
and	edx,	eax	;	remainder < 0 ? divisor_hi : 0
and	eax,	[esp+4]	;	remainder < 0 ? divisor_lo : 0
add	eax,	ebx	;	remainder_lo
add	edx,	ecx	;	remainder_hi
add	esp,	16	;	Remove local variables.

Chapter 8 Integer Optimizations 177