News for hackers. Stuff that matters.
Re:Future Directions for Optimizing Compilers by BlackFingolfin on Monday September 10, 2018 @01:01PM (#17953206)
Am I missing something, or is their example at the start of section 3 (bottom of page 7) plain wrong? If x is an uint32_t, then ((x << 31) >> 31) + 1 returns either 1 or 2; while the "optimized" version ~x&1 reutrns 1 or 0 (which matches the semantics they describe, while the initial code does not).
Re:Future Directions for Optimizing Compilers by jcdavis on Monday September 10, 2018 @01:43PM (#17953528)
If you look at the llvm IR, it is using ashr (arithmetic shift), so you can probably assume this isn't targeting unsigned types
Re:Future Directions for Optimizing Compilers by Someone on Monday September 10, 2018 @02:14PM (#17953758)
The text also explicitly states this is about signed integers: ”let’s look at taking ((x << 31) >> 31) + 1, an inefficient idiom for isolating and flipping the low bit of a signed 32-bit integer”

I think there is an error on page 9, though: ”Second, −a+−b can be rewritten as −(a − b)”. I think that should be either a left-hand side of ”-a--b” or a right-hand side of ”-(a + b)”.