Pixels and Bit Twiddling, pt 1

I was discussing bit manipulation with somebody and it reminded me of one of my single favorite pixel munging routines. I've purposely omitted the NAME of the function - but its pretty cool ("One from the book", as Paul Erdos might say). And this one was really originally written (meaning he showed me the trick first) by the prime "PixelBoy" himself, Ben Weiss.

inline uint32 FUNCTION_NAME_OMITTED (uint32 a, uint32 b, uint32 mask=0x01010100)
{
uint32 s = (a + b);
uint32 c = (a ^ b ^ s) & mask;
return (s-c) (c-(c>>8));
}

Any guesses as to what it does?

I'll give you a hint: its a 32 bit pixel operation, but doesn't preserve alpha in this case (if only you could access the carry bit from plain old C ... ).... Just decoding what this did taught me a LOT about graphics coding, back in the day.