carry/overflow & subtraction in x86

math x86 overflow twos-complement carryflag
Robz picture Robz · Jan 23, 2012 · Viewed 9.4k times · Source

I'm trying to wrap my head around overflow & carry flags in x86.

As I understand it, for addition of signed 2's complement numbers, the flags can only be generated in one of four ways (my examples are 4-bit numbers):

  1. pos+pos = neg (overflow)
    • 0111 + 0001 = 1000 (7 + 1 = -8)
  2. pos+neg = pos (carry)
    • 0011 + 1110 = 0001 (3 + -2 = 1)
  3. neg+neg = neg (carry)
    • 1111 + 1111 = 1110 (-1 + -1 = -2)
  4. neg+neg = pos (overflow & carry)
    • 1000 + 1001 = 0001 (-8 + -7 = 1)

So, in x86 assembly, does subracting B from A generate the same flags as adding A and -B?

Answer

srking picture srking · Jan 24, 2012

Here's a reference table that might help. This shows an example of every possible combination of the 4 arithmetic flags that can result from the ADD and SUB instructions on x86. 'h' 'ud' and 'd' stand for hex, unsigned decimal and signed decimal representations of each value. For example, the first row for SUB says 0xFF - 0xFE = 0x1 with no flags set.

But, I think the short story is that Alex's answer is correct.

 ADD
       A                   B                   A + B              Flags  
 ---------------     ----------------    ---------------      -----------------
 h  |  ud  |   d   | h  |  ud  |   d   | h  |  ud  |   d   | OF | SF | ZF | CF
 ---+------+-------+----+------+-------+----+------+-------+----+----+----+---
 7F | 127  |  127  | 0  |  0   |   0   | 7F | 127  |  127  | 0  | 0  | 0  | 0
 FF | 255  |  -1   | 7F | 127  |  127  | 7E | 126  |  126  | 0  | 0  | 0  | 1
 0  |  0   |   0   | 0  |  0   |   0   | 0  |  0   |   0   | 0  | 0  | 1  | 0
 FF | 255  |  -1   | 1  |  1   |   1   | 0  |  0   |   0   | 0  | 0  | 1  | 1
 FF | 255  |  -1   | 0  |  0   |   0   | FF | 255  |  -1   | 0  | 1  | 0  | 0
 FF | 255  |  -1   | FF | 255  |  -1   | FE | 254  |  -2   | 0  | 1  | 0  | 1
 FF | 255  |  -1   | 80 | 128  | -128  | 7F | 127  |  127  | 1  | 0  | 0  | 1
 80 | 128  | -128  | 80 | 128  | -128  | 0  |  0   |   0   | 1  | 0  | 1  | 1
 7F | 127  |  127  | 7F | 127  |  127  | FE | 254  |  -2   | 1  | 1  | 0  | 0


 SUB
       A                   B                   A - B              Flags  
 ---------------     ----------------    ---------------      -----------------
 h  |  ud  |   d   | h  |  ud  |   d   | h  |  ud  |   d   || OF | SF | ZF | CF
----+------+-------+----+------+-------+----+------+-------++----+----+----+----
 FF | 255  |  -1   | FE | 254  |  -2   | 1  |  1   |   1   || 0  | 0  | 0  | 0
 7E | 126  |  126  | FF | 255  |  -1   | 7F | 127  |  127  || 0  | 0  | 0  | 1
 FF | 255  |  -1   | FF | 255  |  -1   | 0  |  0   |   0   || 0  | 0  | 1  | 0
 FF | 255  |  -1   | 7F | 127  |  127  | 80 | 128  | -128  || 0  | 1  | 0  | 0
 FE | 254  |  -2   | FF | 255  |  -1   | FF | 255  |  -1   || 0  | 1  | 0  | 1
 FE | 254  |  -2   | 7F | 127  |  127  | 7F | 127  |  127  || 1  | 0  | 0  | 0
 7F | 127  |  127  | FF | 255  |  -1   | 80 | 128  | -128  || 1  | 1  | 0  | 1

Related questions

How can I check for NaN values?

float('nan') results in Nan (not a number). But how do I check for it? Should be very easy, but I cannot find it.

Read More
Math.random() explanation

This is a pretty simple Java (though probably applicable to all programming) question: Math.random() returns a number between zero and one. If I want to return an integer between zero and hundred, I would do: (int) Math.floor(Math.…

Read More
Calculate distance between two latitude-longitude points? (Haversine formula)

How do I calculate the distance between two points specified by latitude and longitude? For clarification, I'd like the distance in kilometers; the points use the WGS84 system and I'd like to understand the relative accuracies of the approaches available.

Read More