[LLVM] Fix an assertion when a weak symbol is defined in global_asm.
This change will fix the issue from
https://github.com/japaric/svd2rust/pull/130
cc @japaric
r? @alexcrichton
"Signal termination not considered" -> "Signal termination is not considered"
The first line of the description was rewrapped so it fits into 80 characters.
The docs for the `compiler_builtins_lib` library feature were removed
in #42899. But, though the `compiler_builtins` library has been
migrated out-of-tree, the feature remains, and is needed to use the
stand-alone crate. We reintroduce the docs for the feature, and add a
reference to them when describing how to create a `no_std` executable.
Fix in weak_count in Arc in the case the weak count is locked.
In the case the weak count was locked, the weak_count function could
return usize::MAX. We need to test this condition manually.
trans: Internalize symbols without relying on LLVM
This PR makes the compiler use the information gather by the trans collector in order to determine which symbols/trans-items can be made internal. This has the advantages:
+ of being LLVM independent,
+ of also working in incremental mode, and
+ of allowing to not keep all LLVM modules in memory at the same time.
This is in preparation for fixing issue #39280.
cc @rust-lang/compiler
This PR cuts down on a large number of `#[inline(always)]` and `#[inline]`
annotations in libcore for various core functions. The `#[inline(always)]`
annotation is almost never needed and is detrimental to debug build times as it
forces LLVM to perform inlining when it otherwise wouldn't need to in debug
builds. Additionally `#[inline]` is an unnecessary annoation on almost all
generic functions because the function will already be monomorphized into other
codegen units and otherwise rarely needs the extra "help" from us to tell LLVM
to inline something.
Overall this PR cut the compile time of a [microbenchmark][1] by 30% from 1s to
0.7s.
[1]: https://gist.github.com/alexcrichton/a7d70319a45aa60cf36a6a7bf540dd3a
