The crash that happened in #23566 doesn't happen anymore with the LLVM mergefunc
pass enabled and it hugely reduces code size (for example it shaves off 10% of the
final Servo executable). This patch reenables it.
The LLVM PassManager has a PrepareForThinLTO flag, which is intended
when compilation occurs in conjunction with linking by ThinLTO. The
flag has two effects:
* The NameAnonGlobal pass is run after all other passes, which
ensures that all globals have a name.
* In optimized builds, a number of late passes (mainly related to
vectorization and unrolling) are disabled, on the rationale that
these a) will increase codesize of the intermediate artifacts
and b) will be run by ThinLTO again anyway.
This patch enables the use of PrepareForThinLTO if Thin or ThinLocal
linking is used.
The background for this change is the CI failure in #49479, which
we assume to be caused by the NameAnonGlobal pass not being run.
As this changes which passes LLVM runs, this might have performance
(or other) impact, so we want to land this separately.
In `LLVMRustHasFeature()`, rather than using `MCInfo->getFeatureTable()`
that is specific to Rust's LLVM fork, we can use this in LLVM 6:
/// Check whether the subtarget features are enabled/disabled as per
/// the provided string, ignoring all other features.
bool checkFeatures(StringRef FS) const;
Now rustc using external LLVM can also have `target_feature`.
implement minmax intrinsics
This adds the `simd_{fmin,fmax}` intrinsics, which do a vertical (lane-wise) `min`/`max` for floating point vectors that's equivalent to Rust's `min`/`max` for `f32`/`f64`.
It might make sense to make `{f32,f64}::{min,max}` use the `minnum` and `minmax` intrinsics as well.
---
~~HELP: I need some help with these. Either I should go to sleep or there must be something that I must be missing. AFAICT I am calling the `maxnum` builder correctly, yet rustc/LLVM seem to insert a call to `llvm.minnum` there instead...~~ EDIT: Rust's LLVM version is too old :/
Add basic PGO support.
This PR adds two mutually exclusive options for profile usage and generation using LLVM's instruction profile generation (the same as clang uses), `-C pgo-use` and `-C pgo-gen`.
See each commit for details.
This commit updates our Fat LTO logic to tweak our custom wrapper around LLVM's
"link modules" functionality. Previously whenever the
`LLVMRustLinkInExternalBitcode` function was called it would call LLVM's
`Linker::linkModules` wrapper. Internally this would crate an instance of a
`Linker` which internally creates an instance of an `IRMover`. Unfortunately for
us the creation of `IRMover` is somewhat O(n) with the input module. This means
that every time we linked a module it was O(n) with respect to the entire module
we had built up!
Now the modules we build up during LTO are quite large, so this quickly started
creating an O(n^2) problem for us! Discovered in #48025 it turns out this has
always been a problem and we just haven't noticed it. It became particularly
worse recently though due to most libraries having 16x more object files than
they previously did (1 -> 16).
This commit fixes this performance issue by preserving the `Linker` instance
across all links into the main LLVM module. This means we only create one
`IRMover` and allows LTO to progress much speedier.
From the `cargo-cache` project in #48025 a **full build** locally when from
5m15s to 2m24s. Looking at the timing logs each object file was linked in in
single-digit millisecond rather than hundreds, clearly being a nice improvement!
Closes#48025
The following submodules have been updated for a new version of LLVM:
- `src/llvm`
- `src/libcompiler_builtins` - transitively contains compiler-rt
- `src/dlmalloc`
This also updates the docker container for dist-i686-freebsd as the old 16.04
container is no longer capable of building LLVM. The
compiler-rt/compiler-builtins and dlmalloc updates are pretty routine without
much interesting happening, but the LLVM update here is of particular note.
Unlike previous updates I haven't cherry-picked all existing patches we had on
top of our LLVM branch as we have a [huge amount][patches4] and have at this
point forgotten what most of them are for. Instead I started from the current
`release_60` branch in LLVM and only applied patches that were necessary to get
our tests working and building.
The current set of custom rustc-specific patches included in this LLVM update are:
* rust-lang/llvm@1187443 - this is how we actually implement
`cfg(target_feature)` for now and continues to not be upstreamed. While a
hazard for SIMD stabilization this commit is otherwise keeping the status
quo of a small rustc-specific feature.
* rust-lang/llvm@013f2ec - this is a rustc-specific optimization that we haven't
upstreamed, notably teaching LLVM about our allocation-related routines (which
aren't malloc/free). Once we stabilize the global allocator routines we will
likely want to upstream this patch, but for now it seems reasonable to keep it
on our fork.
* rust-lang/llvm@a65bbfd - I found this necessary to fix compilation of LLVM in
our 32-bit linux container. I'm not really sure why it's necessary but my
guess is that it's because of the absolutely ancient glibc that we're using.
In any case it's only updating pieces we're not actually using in LLVM so I'm
hoping it'll turn out alright. This doesn't seem like something we'll want to
upstream.c
* rust-lang/llvm@77ab1f0 - this is what's actually enabling LLVM to build in our
i686-freebsd container, I'm not really sure what's going on but we for sure
probably don't want to upstream this and otherwise it seems not too bad for
now at least.
* rust-lang/llvm@9eb9267 - we currently suffer on MSVC from an [upstream bug]
which although diagnosed to a particular revision isn't currently fixed
upstream (and the bug itself doesn't seem too active). This commit is a
partial revert of the suspected cause of this regression (found via a
bisection). I'm sort of hoping that this eventually gets fixed upstream with a
similar fix (which we can replace in our branch), but for now I'm also hoping
it's a relatively harmless change to have.
After applying these patches (plus one [backport] which should be [backported
upstream][llvm-back]) I believe we should have all tests working on all
platforms in our current test suite. I'm like 99% sure that we'll need some more
backports as issues are reported for LLVM 6 when this propagates through
nightlies, but that's sort of just par for the course nowadays!
In any case though some extra scrutiny of the patches here would definitely be
welcome, along with scrutiny of the "missing patches" like a [change to pass
manager order](rust-lang/llvm@2717444753), [another change to pass manager
order](rust-lang/llvm@c782febb7b), some [compile fixes for
sparc](rust-lang/llvm@1a83de63c4), and some [fixes for
solaris](rust-lang/llvm@c2bfe0abb).
[patches4]: https://github.com/rust-lang/llvm/compare/5401fdf23...rust-llvm-release-4-0-1
[backport]: https://github.com/rust-lang/llvm/commit/5c54c252db
[llvm-back]: https://bugs.llvm.org/show_bug.cgi?id=36114
[upstream bug]: https://bugs.llvm.org/show_bug.cgi?id=36096
---
The update to LLVM 6 is desirable for a number of reasons, notably:
* This'll allow us to keep up with the upstream wasm backend, picking up new
features as they start landing.
* Upstream LLVM has fixed a number of SIMD-related compilation errors,
especially around AVX-512 and such.
* There's a few assorted known bugs which are fixed in LLVM 5 and aren't fixed
in the LLVM 4 branch we're using.
* Overall it's not a great idea to stagnate with our codegen backend!
This update is mostly powered by #47730 which is allowing us to update LLVM
*independent* of the version of LLVM that Emscripten is locked to. This means
that when compiling code for Emscripten we'll still be using the old LLVM 4
backend, but when compiling code for any other target we'll be using the new
LLVM 6 target. Once Emscripten updates we may no longer need this distinction,
but we're not sure when that will happen!
Closes#43370Closes#43418Closes#47015Closes#47683Closesrust-lang-nursery/stdsimd#157Closesrust-lang-nursery/rust-wasm#3
First round of LLVM 6.0.0 compatibility
This includes a number of commits for the first round of upgrading to LLVM 6. There are still [lingering bugs](https://github.com/rust-lang/rust/issues/47683) but I believe all of this will nonetheless be necessary!
Teach rustc about DW_AT_noreturn and a few more DIFlags
We achieve two small things with this PR:
1. We provide definitions for a few additional llvm debuginfo flags
1. We _use_ one of these new flags, `FlagNoReturn`, and add it to debuginfo for functions with the never return type (`!`).
It looks like LLVM also removed it in llvm-mirror/llvm@f45adc29d in favor of the
name "GNU64". This was added in the thought that we'd need such a variant when
adding mips64 support but we ended up not needing it! For now let's just
removing the various support on the Rust side of things.
Looks like they did some refactoring of flags in the backend and this should
catch us up! The "unsafe algebra" boolean has been split into a number of
boolean flags for various operations, and this updates to use the `setFast`
function which should hopefully have the same behavior as before.
This was updated in llvm-mirror/llvm@00e900afd
LLVM has since removed the `CodeModel::Default` enum value in favor of an
`Optional` implementationg throughout LLVM. Let's mirror the same change in Rust
and update the various bindings we call accordingly.
Removed in llvm-mirror/llvm@9aafb854c
LLVM <= 4.0 used a non-standard interpretation of `DW_OP_plus`. In the
DWARF standard, this adds two items on the expressions stack. LLVM's
behavior was more like DWARF's `DW_OP_plus_uconst` -- adding a constant
that follows the op. The patch series starting with [D33892] switched
to the standard DWARF interpretation, so we need to follow.
[D33892]: https://reviews.llvm.org/D33892
Use name-discarding LLVM context
This is only applicable when neither of --emit=llvm-ir or --emit=llvm-bc are not
requested.
In case either of these outputs are wanted, but the benefits of such context are
desired as well, -Zfewer_names option provides the same functionality regardless
of the outputs requested.
Should be a viable fix for https://github.com/rust-lang/rust/issues/46449
This is only applicable when neither of --emit=llvm-ir or --emit=llvm-bc are not
requested.
In case either of these outputs are wanted, but the benefits of such context are
desired as well, -Zfewer_names option provides the same functionality regardless
of the outputs requested.
This commit is the next attempt to enable multiple codegen units by default in
release mode, getting some of those sweet, sweet parallelism wins by running
codegen in parallel. Performance should not be lost due to ThinLTO being on by
default as well.
Closes#45320
This commit implements a workaround for #46346 which basically just
avoids triggering the situation that LLVM's bug
https://bugs.llvm.org/show_bug.cgi?id=35562 arises. More details can be
found in the code itself but this commit is also intended to ...
Closes#46346
