Stream: t-compiler/wg-llvm

Topic: noalias annotations

rkruppe (Mar 27 2019 at 12:24, on Zulip):

In case people here haven't seen it yet:

rkruppe (Mar 27 2019 at 12:24, on Zulip):

I would love to contribute, but I haven't even found the time to read Jeroen Dobbelaere's proposal, and probably won't until after EuroLLVM

Nikita Popov (Apr 19 2019 at 14:00, on Zulip):

Finally got around to reading that RFC... unfortunately I can't say that I really understood any of it. Seems to be very convoluted and I suspect that it will make noalias practically unusable for rust (because unlike C restrict, pretty much everything is noalias for us).

nagisa (Apr 19 2019 at 15:41, on Zulip):

It was hard to follow, indeed.

nagisa (Apr 19 2019 at 15:41, on Zulip):

which is exactly why it is hard to oppose it or contirbute to it

nagisa (Apr 19 2019 at 15:41, on Zulip):

my understanding, however, was that the proposal does not get rid of the currently available alias annotations

rkruppe (Apr 19 2019 at 17:31, on Zulip):

They hosted a round table at EuroLLVM and I attended. Even in such an interactive discussion I didn't really manage to a good idea of what the IR semantics of their proposal would be (or even how it would be used by AA passes), they kept talking in C terms. I got the impression they want to map C restrict rules 1:1 to IR. That's fair in a way, but probably indeed makes it useless for Rust.

Last update: Jul 03 2020 at 16:35UTC