Use rvalue promotion to 'static instead of static items.

[eddyb]

Fixes #44240. Among other things, in crates that do a lot of formatting, this could reduce the number of items, although I haven't measured the performance benefits. If there's a codegen slowdown, that should IMO be solved by caching the output of miri, not by using static.

r? @alexcrichton

[alexcrichton]

Just to confirm: you've verified that the codegen before and after for these callsites is the same? (these were carefully constructed to reduce code bloat)

[eddyb]

@alexcrichton The static data in panic! and format_args! promote in MIR, but the LLVM IR is harder to diff - it looks like there is no difference in the number of allocas (so same stack usage), more static data and less runtime instructions (before optimizations, after both are identical).

[alexcrichton]

Excellent!

@bors: r+

[bors]

📌 Commit 7e0b79d has been approved by alexcrichton

[alexcrichton]

@bors: rollup

[aidanhs]

@bors r- rollup-

Suspect this caused failure in https://travis-ci.org/rust-lang/rust/jobs/273704890, rollup #44460:

[01:42:35]                   ((::fmt::format as
[01:42:35]                        fn(std::fmt::Arguments<'_>) -> std::string::String {std::fmt::format})(((<::std::fmt::Arguments>::new_v1
[01:42:35]                                                                                                    as
[01:42:35]                                                                                                    fn(&[&str], &[std::fmt::ArgumentV1<'_>]) -> std::fmt::Arguments<'_> {std::fmt::Arguments<'_>::new_v1})(({
[01:42:35]                                                                                                                                                                                                                static __STATIC_FMTSTR:
[01:42:35]                                                                                                                                                                                                                       &'static [&'static str]
[01:42:35]                                                                                                                                                                                                                       =
[01:42:35]                                                                                                                                                                                                                    (&([("test"
[01:42:35]                                                                                                                                                                                                                            as
[01:42:35]                                                                                                                                                                                                                            &'static str)]

vs

[01:42:35]                   ((::fmt::format as
[01:42:35]                        fn(std::fmt::Arguments<'_>) -> std::string::String {std::fmt::format})(((<::std::fmt::Arguments>::new_v1
[01:42:35]                                                                                                    as
[01:42:35]                                                                                                    fn(&[&str], &[std::fmt::ArgumentV1<'_>]) -> std::fmt::Arguments<'_> {std::fmt::Arguments<'_>::new_v1})((&([("test"
[01:42:35]                                                                                                                                                                                                                   as
[01:42:35]                                                                                                                                                                                                                   &'static str)]

[eddyb]

Wait, what, the PR builds don't check pretty tests? Ugh...

[eddyb]

Wow, I just ran ./x.py test for nothing... That explains why Travis didn't catch it on the PR though.

[eddyb]

@bors r=alexcrichton p=1 (let's not put it in a rollup, so we can get perf deltas)

[bors]

📌 Commit 10f66bd has been approved by alexcrichton

[bors]

⌛ Testing commit 10f66bd with merge 23aaeb5...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: alexcrichton
Pushing 23aaeb5 to master...