| Age | Commit message (Collapse) | Author | Lines |
|---|
|
Permit MIR inlining without #[inline]
I noticed that there are at least a handful of portable-simd functions that have no `#[inline]` but compile to an assign + return.
I locally benchmarked inlining thresholds between 0 and 50 in increments of 5, and 50 seems to be the best. Interesting. That didn't include check builds though, ~maybe perf will have something to say about that~.
Perf has little useful to say about this. We generally regress all the check builds, as best as I can tell, due to a number of small codegen changes in a particular hot function in the compiler. Probably this is because we've nudged the inlining outcomes all over, and uses of `#[inline(always)]`/`#[inline(never)]` might need to be adjusted.
|
|
|
|
|
|
r=wesleywiser
Preserve argument indexes when inlining MIR
We store argument indexes on VarDebugInfo. Unlike the previous method of relying on the variable index to know whether a variable is an argument, this survives MIR inlining.
We also no longer check if var.source_info.scope is the outermost scope. When a function gets inlined, the arguments to the inner function will no longer be in the outermost scope. What we care about though is whether they were in the outermost scope prior to inlining, which we know by whether we assigned an argument index.
Fixes #83217
I considered using `Option<NonZeroU16>` instead of `Option<u16>` to store the index. I didn't because `TypeFoldable` isn't implemented for `NonZeroU16` and because it looks like due to padding, it currently wouldn't make any difference. But I indexed from 1 anyway because (a) it'll make it easier if later it becomes worthwhile to use a `NonZeroU16` and because the arguments were previously indexed from 1, so it made for a smaller change.
This is my first PR on rust-lang/rust, so apologies if I've gotten anything not quite right.
|
|
Add codegen tests for issues fixed by LLVM 16
Fixes #75978.
Fixes #99960.
Fixes #101048.
Fixes #101082.
Fixes #101814.
Fixes #103132.
Fixes #103327.
|
|
|
|
These don't optimize with debug assertions. For one of them, this
is due to the new alignment checks, for the other I'm not sure
what specifically blocks it.
|
|
We store argument indexes on VarDebugInfo. Unlike the previous method of
relying on the variable index to know whether a variable is an argument,
this survives MIR inlining.
We also no longer check if var.source_info.scope is the outermost scope.
When a function gets inlined, the arguments to the inner function will
no longer be in the outermost scope. What we care about though is
whether they were in the outermost scope prior to inlining, which we
know by whether we assigned an argument index.
|
|
|
|
|
|
|
|
Fixes #110005
|
|
Use SipHash-1-3 instead of SipHash-2-4 for StableHasher
Noticed this, and it seems easy and likely a perf win. IIUC we don't need DDOS resistance (just collision) so we ideally would have an even faster hash, but it's hard to beat this SipHash impl here, since it's been so highly tuned for the interface.
It wouldn't surprise me if there's some subtle reason changing this sucks, as it's so obvious it seems likely to have been done. Still, SipHash-1-3 seems to still have the guarantees StableHasher should need (and seemingly more), and is clearly less work. So it's worth a shot.
Not fully tested locally.
|
|
Validate `ignore` and `only` compiletest directive, and add human-readable ignore reasons
This PR adds strict validation for the `ignore` and `only` compiletest directives, failing if an unknown value is provided to them. Doing so uncovered 79 tests in `tests/ui` that had invalid directives, so this PR also fixes them.
Finally, this PR adds human-readable ignore reasons when tests are ignored due to `ignore` or `only` directives, like *"only executed when the architecture is aarch64"* or *"ignored when the operative system is windows"*. This was the original reason why I started working on this PR and #108659, as we need both of them for Ferrocene.
The PR is a draft because the code is extremely inefficient: it calls `rustc --print=cfg --target $target` for every rustc target (to gather the list of allowed ignore values), which on my system takes between 4s and 5s, and performs a lot of allocations of constant values. I'll fix both of them in the coming days.
r? `@ehuss`
|
|
|
|
|
|
Allow `transmute`s to produce `OperandValue`s instead of needing `alloca`s
LLVM can usually optimize these away, but especially for things like transmutes of newtypes it's silly to generate the `alloc`+`store`+`load` at all when it's actually a nop at LLVM level.
|
|
LLVM can usually optimize these away, but especially for things like transmutes of newtypes it's silly to generate the `alloc`+`store`+`load` at all when it's actually a nop at LLVM level.
|
|
Extend -Cdebuginfo with new options and named aliases
This is a rebase of https://github.com/rust-lang/rust/pull/83947, along with my best guess at what the new options mean. I tried to follow the LLVM source code to get a better idea but ran into quite a lot of trouble (https://rust-lang.zulipchat.com/#narrow/stream/187780-t-compiler.2Fwg-llvm/topic/go-to-definition.20in.20src.2Fllvm-project.3F). The description for the original PR follows below.
Note that the changes in this PR have already been through FCP: https://github.com/rust-lang/rust/pull/83947#issuecomment-878384979
Closes https://github.com/rust-lang/rust/pull/109311. Helps with https://github.com/rust-lang/rust/pull/104968.
r? `@michaelwoerister` cc `@cuviper`
---
The -Cdebuginfo=1 option was never line tables only and can't be due to backwards compatibility issues. This was clarified and an option for emitting line tables only was added. Additionally an option for emitting line info directives only was added, which is needed for some targets, i.e. nvptx. The debug info options should now behave similarly to clang's debug info options.
Fix https://github.com/rust-lang/rust/issues/60020
Fix https://github.com/rust-lang/rust/issues/64405
|
|
Fixes #75978.
Fixes #99960.
Fixes #101048.
Fixes #101082.
Fixes #101814.
Fixes #103132.
Fixes #103327.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
`-Cdebuginfo=1` was never line tables only and
can't be due to backwards compatibility issues.
This was clarified and an option for line tables only
was added. Additionally an option for line info
directives only was added, which is well needed for
some targets. The debug info options should now
behave the same as clang's debug info options.
|
|
Insert alignment checks for pointer dereferences when debug assertions are enabled
Closes https://github.com/rust-lang/rust/issues/54915
- [x] Jake tells me this sounds like a place to use `MirPatch`, but I can't figure out how to insert a new basic block with a new terminator in the middle of an existing basic block, using `MirPatch`. (if nobody else backs up this point I'm checking this as "not actually a good idea" because the code looks pretty clean to me after rearranging it a bit)
- [x] Using `CastKind::PointerExposeAddress` is definitely wrong, we don't want to expose. Calling a function to get the pointer address seems quite excessive. ~I'll see if I can add a new `CastKind`.~ `CastKind::Transmute` to the rescue!
- [x] Implement a more helpful panic message like slice bounds checking.
r? `@oli-obk`
|
|
Changing the layout of the InitMask changed the const
allocations' hashes.
|
|
Upgrade to LLVM 16, again
Relative to the previous attempt in https://github.com/rust-lang/rust/pull/107224:
* Update to GCC 8.5 on dist-x86_64-linux, to avoid std::optional ABI-incompatibility between libstdc++ 7 and 8.
* Cherry-pick https://github.com/llvm/llvm-project/commit/96df79af029b85616ab90e73143e0e8ae89c7b46.
* Cherry-pick https://github.com/llvm/llvm-project/commit/6fc670e5e370e90759b5adecd9980b63609dc2f8.
r? `@cuviper`
|
|
Use poison instead of undef
In cases where it is legal, we should prefer poison values over undef values.
This replaces undef with poison for aggregate construction and for uninhabited types. There are more places where we can likely use poison, but I wanted to stay conservative to start with.
In particular the aggregate case is important for newer LLVM versions, which are not able to handle an undef base value during early optimization due to poison-propagation concerns.
r? `@cuviper`
|
|
|
|
Add `CastKind::Transmute` to MIR
~~Nothing actually produces it in this commit, so I don't know how to test it, but it also means it shouldn't be possible for it to break anything.~~
Includes lowering `transmute` calls to it, so it's used.
Zulip Conversation: <https://rust-lang.zulipchat.com/#narrow/stream/189540-t-compiler.2Fwg-mir-opt/topic/Good.20first.20isssue/near/321849610>
|
|
mv tests/codegen/issue-* tests/codegen/issues/
No changes to the contents; just a move.
Like how there's a <https://github.com/rust-lang/rust/tree/master/tests/ui/issues> folder.
|
|
Updates `interpret`, `codegen_ssa`, and `codegen_cranelift` to consume the new cast instead of the intrinsic.
Includes `CastTransmute` for custom MIR building, to be able to test the extra UB.
|
|
adapt tests/codegen/vec-shrink-panik for LLVM 17
After https://github.com/llvm/llvm-project/commit/0d4a709bb876824a0afa5f86e138e8ffdcaf7661 LLVM now doesn't generate references to panic_cannot_unwind:
https://buildkite.com/llvm-project/rust-llvm-integrate-prototype/builds/17978#0186ff55-ca6f-4bc5-b1ec-2622c77d0ed5/744-746
Adapted as suggested by ````@nikic```` on Zulip:
https://rust-lang.zulipchat.com/#narrow/stream/187780-t-compiler.2Fwg-llvm/topic/a.20couple.20codegen.20test.20failures.20after.20llvm.200d4a709bb876824a/near/342664944
>Okay, so LLVM now realizes that double panic is not possible, so that's fine.
|
|
Make sure that the loop is not fully unrolled (which allows
eliminating the allocas) in LLVM 16 either.
|
|
Remove the assume(!is_null) from Vec::as_ptr
At a guess, this code is leftover from LLVM was worse at keeping track of the niche information here. In any case, we don't need this anymore: Removing this `assume` doesn't get rid of the `nonnull` attribute on the return type.
|
|
After https://github.com/llvm/llvm-project/commit/0d4a709bb876824a0afa5f86e138e8ffdcaf7661
LLVM now doesn't generate references to panic_cannot_unwind:
@nikic:
https://rust-lang.zulipchat.com/#narrow/stream/187780-t-compiler.2Fwg-llvm/topic/a.20couple.20codegen.20test.20failures.20after.20llvm.200d4a709bb876824a/near/342664944
>Okay, so LLVM now realizes that double panic is not possible, so that's fine.
|
|
|
|
|
|
This reverts commit 4a04d086cac54a41517d5657b59d5fe2caca2d71, reversing
changes made to 2d0a7def330ed90f416a3429dbb44c5ee2a914e5.
|
|
Upgrade to LLVM 16
This updates Rust to LLVM 16. It also updates our host compiler for dist-x86_64-linux to LLVM 16. The reason for that is that Bolt from LLVM 15 is not capable of compiling LLVM 16 (https://github.com/llvm/llvm-project/issues/61114).
LLVM 16.0.0 has been [released](https://discourse.llvm.org/t/llvm-16-0-0-release/69326) on March 18, while Rust 1.70 will become stable on June 1.
Tested images: `dist-x86_64-linux`, `dist-riscv64-linux` (alt), `dist-x86_64-illumos`, `dist-various-1`, `dist-various-2`, `dist-powerpc-linux`, `wasm32`, `armhf-gnu`
Tested images until the usual IPv6 failures: `test-various`
|
|
Make sure that the loop is not fully unrolled (which allows
eliminating the allocas) in LLVM 16 either.
|
|
inherit_overflow: adapt pattern to also work with v0 mangling
This test was failing under new-symbol-mangling = true. Adapt pattern to work in both cases.
Related to #106002 from December.
|
|
Wrap the whole LocalInfo in ClearCrossCrate.
MIR contains a lot of information about locals. The primary purpose of this information is the quality of borrowck diagnostics.
This PR aims to drop this information after MIR analyses are finished, ie. starting from post-cleanup runtime MIR.
|
|
In cases where it is legal, we should prefer poison values over
undef values.
This replaces undef with poison for aggregate construction and
for uninhabited types. There are more places where we can likely
use poison, but I wanted to stay conservative to start with.
In particular the aggregate case is important for newer LLVM
versions, which are not able to handle an undef base value during
early optimization due to poison-propagation concerns.
|
|
This test was failing under new-symbol-mangling = true. Adapt pattern to
work in both cases.
Related to #106002 from December.
|
|
r=WaffleLapkin,JakobDegen
Ensure `ptr::read` gets all the same LLVM `load` metadata that dereferencing does
I was looking into `array::IntoIter` optimization, and noticed that it wasn't annotating the loads with `noundef` for simple things like `array::IntoIter<i32, N>`. Trying to narrow it down, it seems that was because `MaybeUninit::assume_init_read` isn't marking the load as initialized (<https://rust.godbolt.org/z/Mxd8TPTnv>), which is unfortunate since that's basically its reason to exist.
The root cause is that `ptr::read` is currently implemented via the *untyped* `copy_nonoverlapping`, and thus the `load` doesn't get any type-aware metadata: no `noundef`, no `!range`. This PR solves that by lowering `ptr::read(p)` to `copy *p` in MIR, for which the backends already do the right thing.
Fortuitiously, this also improves the IR we give to LLVM for things like `mem::replace`, and fixes a couple of long-standing bugs where `ptr::read` on `Copy` types was worse than `*`ing them.
Zulip conversation: <https://rust-lang.zulipchat.com/#narrow/stream/219381-t-libs/topic/Move.20array.3A.3AIntoIter.20to.20ManuallyDrop/near/341189936>
cc `@erikdesjardins` `@JakobDegen` `@workingjubilee` `@the8472`
Fixes #106369
Fixes #73258
|
|
Apparently in CI it's getting generated in the opposite order, one function per file will make the test pass either way.
|
