Nikita Popov
Nikita Popov
I can't reproduce this. For me this prints false two times.
@bors r- Probably someone clicked the synchronize button.
@bors try @rust-timer queue
Looks like this update is a very mild regression, but not enough to look into it in more detail, or skip this version.
If there is a problem here, then it is a clang frontend problem. The svst1 is emitted with TBAA metadata for "long", while your other operations use "long long", which...
Hm, I wonder if we should just always do the clone if replaceNodes is disabled, rather than introduce an additional option.
Could you please provide updated compile-time and performance numbers? The implementation has changed a lot in the meantime.
@kasuga-fj Yes, this is valid. The delinearization reasoning based on GEP source element types needs to be completely removed.
> > Could you please provide updated compile-time and performance numbers? The implementation has changed a lot in the meantime. > > Here are the new compile time numbers @nikic...
I think the cl::opt option is mainly useful for use with opt rather than clang. (Ideally we'd not use cl::opt for this and instead allow specifying pipeline tuning options from...