This commit is an implementation of LLVM's ThinLTO for consumption in rustc
itself. Currently today LTO works by merging all relevant LLVM modules into one
and then running optimization passes. "Thin" LTO operates differently by having
more sharded work and allowing parallelism opportunities between optimizing
codegen units. Further down the road Thin LTO also allows *incremental* LTO
which should enable even faster release builds without compromising on the
performance we have today.
This commit uses a `-Z thinlto` flag to gate whether ThinLTO is enabled. It then
also implements two forms of ThinLTO:
* In one mode we'll *only* perform ThinLTO over the codegen units produced in a
single compilation. That is, we won't load upstream rlibs, but we'll instead
just perform ThinLTO amongst all codegen units produced by the compiler for
the local crate. This is intended to emulate a desired end point where we have
codegen units turned on by default for all crates and ThinLTO allows us to do
this without performance loss.
* In anther mode, like full LTO today, we'll optimize all upstream dependencies
in "thin" mode. Unlike today, however, this LTO step is fully parallelized so
should finish much more quickly.
There's a good bit of comments about what the implementation is doing and where
it came from, but the tl;dr; is that currently most of the support here is
copied from upstream LLVM. This code duplication is done for a number of
reasons:
* Controlling parallelism means we can use the existing jobserver support to
avoid overloading machines.
* We will likely want a slightly different form of incremental caching which
integrates with our own incremental strategy, but this is yet to be
determined.
* This buys us some flexibility about when/where we run ThinLTO, as well as
having it tailored to fit our needs for the time being.
* Finally this allows us to reuse some artifacts such as our `TargetMachine`
creation, where all our options we used today aren't necessarily supported by
upstream LLVM yet.
My hope is that we can get some experience with this copy/paste in tree and then
eventually upstream some work to LLVM itself to avoid the duplication while
still ensuring our needs are met. Otherwise I fear that maintaining these
bindings may be quite costly over the years with LLVM updates!
Don't unwrap work item results as the panic trace is useless
Fixes#43402 now there's no multithreaded panic printouts
Also update a comment
--------
Likely regressed in #43506, where the code was changed to panic in worker threads on error.
Unwrapping gives zero extra information since the stack trace is so short, so we may as well just surface that there was an error and exit the thread properly. Because there are then no multithreaded printouts, I think it should mean the output of the test for #26199 is deterministic and not interleaved (thanks to @philipc https://github.com/rust-lang/rust/issues/43402#issuecomment-333835271 for a hint).
Sadly the output is now:
```
thread '<unnamed>' panicked at 'aborting due to worker thread panic', src/librustc_trans/back/write.rs:1643:20
note: Run with `RUST_BACKTRACE=1` for a backtrace.
error: could not write output to : No such file or directory
error: aborting due to previous error
```
but it's an improvement over the multi-panic situation before.
r? @alexcrichton
incr.comp.: Switch to red/green change tracking, remove legacy system.
This PR finally switches incremental compilation to [red/green tracking](https://github.com/rust-lang/rust/issues/42293) and completely removes the legacy dependency graph implementation -- which includes a few quite costly passes that are simply not needed with the new system anymore.
There's still some documentation to be done and there's certainly still lots of optimizing and tuning ahead -- but the foundation for red/green is in place with this PR. This has been in the making for a long time `:)`
r? @nikomatsakis
cc @alexcrichton, @rust-lang/compiler
Remove mem::transmute used in Box<str> conversions
Given that https://github.com/rust-lang/rust/pull/44877 is failing, I decided to make a separate PR. This is done with the same motivation: to avoid `mem::transmute`-ing non `#[repr(C)]` types.
Overlapping borrows can point to different lvalues.
Overlapping borrows can point to different lvalues.
There's always a basis for the overlap, so instead of removing the assert entirely, I instead pass in the prefix that we found and check that it actually is a prefix of both lvalues.
Fix#44829
make `backtrace = false` compile for windows targets.
when building for windows with `backtrace = false`, `libstd` fails to compile because some modules that use items from `sys_common::backtrace::*` are still included, even though those modules aren't used or referenced by anything.
`sys_common::backtrace` doesn't exist when the backtrace feature is turned off.
--
i've also added `#[cfg(feature = "backtrace")]` to various items that exist exclusively to support `mod backtrace` since the compilation would fail since they would be unused in a configuration with backtraces turned off.
Made `fs::copy` return the length of the main stream
On Windows with the NTFS filesystem, `fs::copy` would return the sum of the
lengths of all streams, which can be different from the length reported by
`metadata` and thus confusing for users unaware of this NTFS peculiarity.
This makes `fs::copy` return the same length `metadata` reports which is the
value it used to return before PR #26751. Note that alternate streams are still
copied; their length is just not included in the returned value.
This change relies on the assumption that the stream with index 1 is always the
main stream in the `CopyFileEx` callback. I could not find any official
document confirming this but empirical testing has shown this to be true,
regardless of whether the alternate stream is created before or after the main
stream.
Resolves#44532
let htmldocck.py check for directories
Since i messed this up during https://github.com/rust-lang/rust/pull/44613, i wanted to codify this into the rustdoc tests to make sure that doesn't happen again.
MIR borrowck: move span_label to `borrowck_errors.rs`
The calls to `span_label` are moved and factorized for:
* E0503 (`cannot_use_when_mutably_borrowed()`)
* E0506 (`cannot_assign_to_borrowed()`)
Additionnally, the error E0594 (`cannot_assign_static()`) has been factorized between `check_loan.rs` and `borrowc_check.rs`.
Part of #44596
Move monomorphize::resolve() to librustc
this moves `monomorphize::resolve(..)` to librustc, and re-enables inlining for some trait methods, fixing #44389
@nikomatsakis I've kept the calls to the new `ty::Instance::resolve(....)` always `.unwrap()`-ing for the moment, how/do you want to add more debugging info via `.unwrap_or()` or something like this?
we still have some related `resolve_*` functions on monomorphize, but I wasn't sure moving them was into the scope for this PR too.
@eddyb mind to take a look too?
There's always a basis for the overlap, so instead of removing the
assert entirely, I instead pass in the prefix that we found and check
that it actually is a prefix of both lvalues.
Fix#44829
Mir pretty print: Add cleanup comment
I found it useful to add a comment indicating whether or not a
BasicBlock is a cleanup block or not. Hopefully you'll find it
useful too.
Docs for size_of::<#[repr(C)]> items.
Most of this info comes from camlorn's blog post on optimizing struct layout and the Rustonomicon.
I don't really like my wording in the first paragraph.
I also cannot find a definition of what `#[repr(C)]` does for enums that have variants with fields. They're allowed, unlike `#[repr(C)] enum`s with no variants.
