Re: Any more graceful way to saturate adds?

Terence wrote:

>> I wouldn't do it that way.
>> I would add the two supposedly unsigned 8-bit integers and test the 9
>> th bit for 1, which if present, the addition has overflowed.

>> Note that any non-overflow result of 255 is for your purposes
>> indistinguishable from an overflow,
>> So you just add, test 9th bit, jump around or load or OR 255h.

> Make that last "255h" into 255 or else 0FFh! (Sigh!)

ok. :)

First I thought: 'what's wrong by using the Carry-flag ? ',
but now I may see the flaw in the whole story.
The result can turn out to be FF(h) without an overflow,
so using FF(h) as overflow indication may lead to errors.

;just use carry-flag as indication:
ADD DL,AL
JC error

;or expand it to a true result
MOV DH,0    ;
ADD DL,AL   ;also possible: MOVZX DX,AL
ADC DH,0    ;DH NZ meens overflow but DX holds true value

;or use just bit0 of DH for overflow (a bit shorter/faster):
ADD DL,AL
SETC DH     ;DH NZ means overflow

;or if DL should just be clamped to FF(h):
ADD DL,AL
SETC DH      ;becomes 0 or 1
NEG DH       ;becomes 0 or FF
OR DL,DH     ;DL keeps value or becomes FF(h)

SETcc can avoid branches and gain speed by this,
but it will not work on older CPUs.
(my example suffers from partial register stalls and
dependencies, so it should by used within other pairing
instructions for full speed),

__
wolfgang