Add hotness data to LLVM remarks
Slight improvement of https://github.com/rust-lang/rust/pull/113040. This makes sure that if PGO is used, remarks generated using `-Zremark-dir` will include the `Hotness` attribute.
r? `@tmiasko`
Add a new `compare_bytes` intrinsic instead of calling `memcmp` directly
As discussed in #113435, this lets the backends be the place that can have the "don't call the function if n == 0" logic, if it's needed for the target. (I didn't actually *add* those checks, though, since as I understood it we didn't actually need them on known targets?)
Doing this also let me make it `const` (unstable), which I don't think `extern "C" fn memcmp` can be.
cc `@RalfJung` `@Amanieu`
cg_llvm: stop identifying ADTs in LLVM IR
This is an extension of https://github.com/rust-lang/rust/pull/94107. It may be a minor perf win.
Fixes#96242.
Now that we use opaque pointers, ADTs can no longer be recursive, so we
do not need to name them. Previously, this would be necessary if you had
a struct like
```rs
struct Foo(Box<Foo>, u64, u64);
```
which would be represented with something like
```ll
%Foo = type { %Foo*, i64, i64 }
```
which is now just
```ll
{ ptr, i64, i64 }
```
r? `@tmiasko`
Coverage FFI types were historically split across two modules, because some of
them were needed by code in `rustc_codegen_ssa`.
Now that all of the coverage codegen code has been moved into
`rustc_codegen_llvm` (#113355), it's possible to move all of the FFI types into
a single module, making it easier to see all of them at once.
Filter out short-lived LLVM diagnostics before they reach the rustc handler
During profiling I saw remark passes being unconditionally enabled: for example `Machine Optimization Remark Emitter`.
The diagnostic remarks enabled by default are [from missed optimizations and opt analyses](https://github.com/rust-lang/rust/pull/113339#discussion_r1259480303). They are created by LLVM, passed to the diagnostic handler on the C++ side, emitted to rust, where they are unpacked, C++ strings are converted to rust, etc.
Then they are discarded in the vast majority of the time (i.e. unless some kind of `-Cremark` has enabled some of these passes' output to be printed).
These unneeded allocations are very short-lived, basically only lasting between the LLVM pass emitting them and the rust handler where they are discarded. So it doesn't hugely impact max-rss, and is only a slight reduction in instruction count (cachegrind reports a reduction between 0.3% and 0.5%) _on linux_. It's possible that targets without `jemalloc` or with a worse allocator, may optimize these less.
It is however significant in the aggregate, looking at the total number of allocated bytes:
- it's the biggest source of allocations according to dhat, on the benchmarks I've tried e.g. `syn` or `cargo`
- allocations on `syn` are reduced by 440MB, 17% (from 2440722647 bytes total, to 2030461328 bytes)
- allocations on `cargo` are reduced by 6.6GB, 19% (from 35371886402 bytes total, to 28723987743 bytes)
Some of these diagnostics objects [are allocated in LLVM](https://github.com/rust-lang/rust/pull/113339#discussion_r1252387484) *before* they're emitted to our diagnostic handler, where they'll be filtered out. So we could remove those in the future, but that will require changing a few LLVM call-sites upstream, so I left a FIXME.
now that remarks are filtered before cg_llvm's diagnostic handler callback
is called, we don't need to do the filtering post c++-to-rust conversion
of the diagnostic.
cleanup: remove pointee types
This can't be merged until the oldest LLVM version we support uses opaque pointers, which will be the case after #114148. (Also note `-Cllvm-args="-opaque-pointers=0"` can technically be used in LLVM 15, though I don't think we should support that configuration.)
I initially hoped this would provide some minor perf win, but in https://github.com/rust-lang/rust/pull/105412#issuecomment-1341224450 it had very little impact, so this is only valuable as a cleanup.
As a followup, this will enable #96242 to be resolved.
r? `@ghost`
`@rustbot` label S-blocked
Operand types are now tracked explicitly, so there is no need to reserve ID 0
for the special always-zero counter.
As part of the renumbering, this change fixes an off-by-one error in the way
counters were counted by the `coverageinfo` query. As a result, functions
should now have exactly the number of counters they actually need, instead of
always having an extra counter that is never used.
Operand types are now tracked explicitly, so there is no need for expression
IDs to avoid counter IDs by descending from `u32::MAX`. Instead they can just
count up from 0, and can be used directly as indices when necessary.
Because the three kinds of operand are now distinguished explicitly, we no
longer need fiddly code to disambiguate counter IDs and expression IDs based on
the total number of counters/expressions in a function.
This does increase the size of operands from 4 bytes to 8 bytes, but that
shouldn't be a big deal since they are mostly stored inside boxed structures,
and the current coverage code is not particularly size-optimized anyway.
Now that we use opaque pointers, ADTs can no longer be recursive, so we
do not need to name them. Previously, this would be necessary if you had
a struct like
```rs
struct Foo(Box<Foo>, u64, u64);
```
which would be represented with something like
```ll
%Foo = type { %Foo*, i64, i64 }
```
which is now just
```ll
{ ptr, i64, i64 }
```
This section name is always constant for a given target, but obtaining it from
LLVM requires a few intermediate allocations. There's no need to do so
repeatedly from inside a per-function loop.
Both GCC and Clang write by default a `.comment` section with compiler
information:
```txt
$ gcc -c -xc /dev/null && readelf -p '.comment' null.o
String dump of section '.comment':
[ 1] GCC: (GNU) 11.2.0
$ clang -c -xc /dev/null && readelf -p '.comment' null.o
String dump of section '.comment':
[ 1] clang version 14.0.1 (https://github.com/llvm/llvm-project.git c62053979489ccb002efe411c3af059addcb5d7d)
```
They also implement the `-Qn` flag to avoid doing so:
```txt
$ gcc -Qn -c -xc /dev/null && readelf -p '.comment' null.o
readelf: Warning: Section '.comment' was not dumped because it does not exist!
$ clang -Qn -c -xc /dev/null && readelf -p '.comment' null.o
readelf: Warning: Section '.comment' was not dumped because it does not exist!
```
So far, `rustc` only does it for WebAssembly targets and only
when debug info is enabled:
```txt
$ echo 'fn main(){}' | rustc --target=wasm32-unknown-unknown --emit=llvm-ir -Cdebuginfo=2 - && grep llvm.ident rust_out.ll
!llvm.ident = !{!27}
```
In the RFC part of this PR it was decided to always add
the information, which gets us closer to other popular compilers.
An opt-out flag like GCC and Clang may be added later on if deemed
necessary.
Implementation-wise, this covers both `ModuleLlvm::new()` and
`ModuleLlvm::new_metadata()` cases by moving the addition to
`context::create_module` and adds a few test cases.
ThinLTO also sees the `llvm.ident` named metadata duplicated (in
temporary outputs), so this deduplicates it like it is done for
`wasm.custom_sections`. The tests also check this duplication does
not take place.
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>
Prototype: Add unstable `-Z reference-niches` option
MCP: rust-lang/compiler-team#641
Relevant RFC: rust-lang/rfcs#3204
This prototype adds a new `-Z reference-niches` option, controlling the range of valid bit-patterns for reference types (`&T` and `&mut T`), thereby enabling new enum niching opportunities. Like `-Z randomize-layout`, this setting is crate-local; as such, references to built-in types (primitives, tuples, ...) are not affected.
The possible settings are (here, `MAX` denotes the all-1 bit-pattern):
| `-Z reference-niches=` | Valid range |
|:---:|:---:|
| `null` (the default) | `1..=MAX` |
| `size` | `1..=(MAX- size)` |
| `align` | `align..=MAX.align_down_to(align)` |
| `size,align` | `align..=(MAX-size).align_down_to(align)` |
------
This is very WIP, and I'm not sure the approach I've taken here is the best one, but stage 1 tests pass locally; I believe this is in a good enough state to unleash this upon unsuspecting 3rd-party code, and see what breaks.
Support `--print KIND=PATH` command line syntax
As is already done for `--emit KIND=PATH` and `-L KIND=PATH`.
In the discussion of #110785, it was pointed out that `--print KIND=PATH` is nicer than trying to apply the single global `-o` path to `--print`'s output, because in general there can be multiple print requests within a single rustc invocation, and anyway `-o` would already be used for a different meaning in the case of `link-args` and `native-static-libs`.
I am interested in using `--print cfg=PATH` in Buck2. Currently Buck2 works around the lack of support for `--print KIND=PATH` by [indirecting through a Python wrapper script](d43cf3a51a/prelude/rust/tools/get_rustc_cfg.py) to redirect rustc's stdout into the location dictated by the build system.
From skimming Cargo's usages of `--print`, it definitely seems like it would benefit from `--print KIND=PATH` too. Currently it is working around the lack of this by inserting `--crate-name=___ --print=crate-name` so that it can look for a line containing `___` as a delimiter between the 2 other `--print` informations it actually cares about. This is commented as a "HACK" and "abuse". 31eda6f7c3/src/cargo/core/compiler/build_context/target_info.rs (L242) (FYI `@weihanglo` as you dealt with this recently in https://github.com/rust-lang/cargo/pull/11633.)
Mentioning reviewers active in #110785: `@fee1-dead` `@jyn514` `@bjorn3`
Resurrect: rustc_llvm: Add a -Z `print-codegen-stats` option to expose LLVM statistics.
This resurrects PR https://github.com/rust-lang/rust/pull/104000, which has sat idle for a while. And I want to see the effect of stack-move optimizations on LLVM (like https://reviews.llvm.org/D153453) :).
I have applied the changes requested by `@oli-obk` and `@nagisa` https://github.com/rust-lang/rust/pull/104000#discussion_r1014625377 and https://github.com/rust-lang/rust/pull/104000#discussion_r1014642482 in the latest commits.
r? `@oli-obk`
-----
LLVM has a neat [statistics](https://llvm.org/docs/ProgrammersManual.html#the-statistic-class-stats-option) feature that tracks how often optimizations kick in. It's very handy for optimization work. Since we expose the LLVM pass timings, I thought it made sense to expose the LLVM statistics too.
-----
(Edit: fix broken link
(Edit2: fix segmentation fault and use malloc
If `rustc` is built with
```toml
[llvm]
assertions = true
```
Then you can see like
```
rustc +stage1 -Z print-codegen-stats -C opt-level=3 tmp.rs
===-------------------------------------------------------------------------===
... Statistics Collected ...
===-------------------------------------------------------------------------===
3 aa - Number of MayAlias results
193 aa - Number of MustAlias results
531 aa - Number of NoAlias results
...
```
And the current default build emits only
```
$ rustc +stage1 -Z print-codegen-stats -C opt-level=3 tmp.rs
===-------------------------------------------------------------------------===
... Statistics Collected ...
===-------------------------------------------------------------------------===
$
```
This might be better to emit the message to tell assertion flag necessity, but now I can't find how to do that...
Add the `no-builtins` attribute to functions when `no_builtins` is applied at the crate level.
**When `no_builtins` is applied at the crate level, we should add the `no-builtins` attribute to each function to ensure it takes effect in LTO.**
This is also the reason why no_builtins does not take effect in LTO as mentioned in #35540.
Now, `#![no_builtins]` should be similar to `-fno-builtin` in clang/gcc, see https://clang.godbolt.org/z/z4j6Wsod5.
Next, we should make `#![no_builtins]` participate in LTO again. That makes sense, as LTO also takes into consideration function-level instruction optimizations, such as the MachineOutliner. More importantly, when a user writes a large `#![no_builtins]` crate, they would like this crate to participate in LTO as well.
We should also add a function-level no_builtins attribute to allow users to have more control over it. This is similar to Clang's `__attribute__((no_builtin))` feature, see https://clang.godbolt.org/z/Wod6KK6eq. Before implementing this feature, maybe we should discuss whether to support more fine-grained control, such as `__attribute__((no_builtin("memcpy")))`.
Related discussions:
- #109821
- #35540
Next (a separate pull request?):
- [ ] Revert #35637
- [ ] Add a function-level `no_builtin` attribute?
