News for hackers. Stuff that matters.
Re:ABI Mistakes by galangalalgol on Saturday May 08, 2021 @02:28PM (#27089581)
Seriously, this has been years now. Is this understandable, or does this tell us something bad about llvm?
Re:ABI Mistakes by zozbot234 on Saturday May 08, 2021 @02:52PM (#27089743)
> Is this understandable, or does this tell us something bad about llvm?

LLVM is a large project that's mostly written in pre-modern C++, and "noalias" is a highly non-trivial feature that affects many parts of the compiler in 'cross-cutting' ways. It would be surprising if it did not turn up some initial bugs.

Re:ABI Mistakes by galangalalgol on Saturday May 08, 2021 @03:32PM (#27089999)
Initial, yes, but this was first uncovered in Oct 2015. That seems like long enough to fix it.
Re:ABI Mistakes by masklinn on Saturday May 08, 2021 @05:54PM (#27091097)
It’s not a single bug, it’s a bunch of different bugs in the interactions between noalias and various analysis and optimisation passes.
Re:ABI Mistakes by saagarjha on Saturday May 08, 2021 @06:10PM (#27091217)
Aliasing analysis is a complicated part of the compiler, and it underpins a lot of optimization passes. It’s not an easy thing to bolt on.
Re:ABI Mistakes by masklinn on Sunday May 09, 2021 @04:16AM (#27094223)
TBF the internal API could be designed more pessimistically, as in llvm could drop noalias annotations unless they’re explicitly maintained (/ converted).

This means optimisation phases would need to explicitly opt-in and aliasing optimisations would commonly be missed by default, but it would avoid miscompilations.

Re:ABI Mistakes by swsieber on Saturday May 08, 2021 @02:43PM (#27089690)
This is understandable. Rust really uses restrict semantics in anger compared to any other language I know of. Have you seen restrict used in a c codebase? The LLVM support for restrict just doesn't get exercised much outside of Rust.
Re:ABI Mistakes by galangalalgol on Saturday May 08, 2021 @04:56PM (#27090603)
Also there are a lot of fortran compilers using llvm now. Fortran has the information for noalias as well.
Re:ABI Mistakes by ghoward on Saturday May 08, 2021 @09:36PM (#27092384)
I used it extensively in my bc (https://git.yzena.com/gavin/bc) in the math to tell the compiler that the out BigNum is not aliased by any of the in operands. (See https://git.yzena.com/gavin/bc/src/branch/master/src/num.c#L... .)
Re:ABI Mistakes by galangalalgol on Saturday May 08, 2021 @03:12PM (#27089891)
I use it in c++ on most signal processing stuff. I think eigen will use it if you don't let it use intrinsics for simd. I also use g++ so I wouldn't have encountered it anyway.
Re:ABI Mistakes by khuey on Saturday May 08, 2021 @02:44PM (#27089698)
This is what happens when you're trying to add a hairy feature to a big legacy codebase.
Re:ABI Mistakes by rrdharan on Saturday May 08, 2021 @02:50PM (#27089730)
Referring to the LLVM codebase as legacy makes me feel old…
Re:ABI Mistakes by wnoise on Saturday May 08, 2021 @02:43PM (#27089685)
Or something bad about noalias?