News for hackers. Stuff that matters.
Re:An Introduction to Wait-Free Programming and Concurrency ... by NobodyNada on Wednesday December 18, 2024 @12:48AM (#42448399)
You'd need that fetch_sub operation to saturate at zero instead of wrapping to 0xFFFF, and I'm not aware of a processor that provides saturating atomic arithmetic operations.
Re:An Introduction to Wait-Free Programming and Concurrency ... by kevmo314 on Wednesday December 18, 2024 @01:11AM (#42448465)
Why? If we define zero as 0xFFFF it will saturate by rolling over the first bit, which increment will not change?
Re:An Introduction to Wait-Free Programming and Concurrency ... by NobodyNada on Wednesday December 18, 2024 @01:21AM (#42448493)
I don't understand what you mean. Decrementing from 0 gives you 0xFFFF, incrementing from 0xFFFF gives you zero. (It doesn't matter whether you use unsigned or signed increment as those are the same operation, at least on a two's-complement processor, where the bit pattern 0xFFFF represents -1.)

You could do this trick on a machine that uses signed-magnitude representation, since a signed increment of 0xFFFF (INT_MIN) would result in 0xFFFE (INT_MIN + 1). But all modern processors use two's complement.