Matthias Krüger [Wed, 17 Nov 2021 14:58:04 +0000 (15:58 +0100)]
Rollup merge of #90884 - Nilstrieb:fix-span-trivial-trait-bound, r=estebank
Fix span for non-satisfied trivial trait bounds
The spans for "trait bound not satisfied" errors in trivial trait bounds referenced the entire item (fn, impl, struct) before.
Now they only reference the obligation itself (`String: Copy`)
Matthias Krüger [Wed, 17 Nov 2021 14:58:02 +0000 (15:58 +0100)]
Rollup merge of #90861 - 5225225:nonprinting-char, r=davidtwco
Print escaped string if char literal has multiple characters, but only one printable character
Fixes #90857
I'm not sure about the error message here, it could get rather long and *maybe* using the names of characters would be better? That wouldn't help the length any, though.
Matthias Krüger [Wed, 17 Nov 2021 14:58:01 +0000 (15:58 +0100)]
Rollup merge of #90772 - GuillaumeGomez:vec-retain-mut, r=joshtriplett
Add Vec::retain_mut
This is to continue the discussion started in #83218.
Original comment was:
> Take 2 of #34265, since I needed this today.
The reason I think why we should add `retain_mut` is for coherency and for discoverability. For example we have `chunks` and `chunks_mut` or `get` and `get_mut` or `iter` and `iter_mut`, etc. When looking for mutable `retain`, I would expect `retain_mut` to exist. It took me a while to find out about `drain_filter`. So even if it provides an API close to `drain_filter`, just for the discoverability, I think it's worth it.
cc ``````@m-ou-se`````` ``````@jonas-schievink`````` ``````@Mark-Simulacrum``````
Matthias Krüger [Wed, 17 Nov 2021 14:58:00 +0000 (15:58 +0100)]
Rollup merge of #90687 - jhpratt:const_panic, r=oli-obk
Permit const panics in stable const contexts in stdlib
Without this change, it is not possible to use `panic!` and similar (including `assert!`) in stable const contexts inside of stdlib. See #89542 for a real-world case that currently fails for this reason. This does _not_ affect any user code.
For example, this snippet currently fails to compile:
With the addition of `#[rustc_const_unstable]` to `core::panicking::panic`, the error no longer occurs. This snippet has been added verbatim in this PR as a UI test.
To avoid needing to add `#![feature(core_panic)]` to libcore, the two instances of direct calls to `core::panicking::panic` have been switched to use the `panic!` macro.
I am requesting prioritization because this is holding up other stabilizations such as #89542 (which is otherwise ready to merge and succeeds with this change)
Matthias Krüger [Wed, 17 Nov 2021 14:57:57 +0000 (15:57 +0100)]
Rollup merge of #90667 - rukai:improve_static_lifetime_diagnostics, r=estebank
Improve diagnostics when a static lifetime is expected
Makes progress towards https://github.com/rust-lang/rust/issues/90600
The diagnostics here were previously entirely removed due to giving a misleading suggestion but if we instead provide an informative label in that same location it should better help the user understand the situation.
I included the example from the issue as it demonstrates an area where the diagnostics are still lacking.
Happy to remove that if its just adding noise atm.
Matthias Krüger [Tue, 16 Nov 2021 22:58:24 +0000 (23:58 +0100)]
Rollup merge of #90933 - compiler-errors:master, r=estebank
Fix await suggestion on non-future type
Remove a match block that would suggest to add `.await` in the case where the expected type's `Future::Output` equals the found type. We only want to suggest `.await`ing in the opposite case (the found type's `Future::Output` equals the expected type).
The code sample is here: https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=6ba6b83d4dddda263553b79dca9f6bcb
Before:
```
➜ ~ rustc --edition=2021 --crate-type=lib test.rs
error[E0308]: `match` arms have incompatible types
--> test.rs:4:14
|
2 | let x = match 1 {
| _____________-
3 | | 1 => other(),
| | ------- this is found to be of type `impl Future`
4 | | 2 => other().await,
| | ^^^^^^^^^^^^^ expected opaque type, found enum `Result`
5 | | };
| |_____- `match` arms have incompatible types
|
= note: expected type `impl Future`
found enum `Result<(), ()>`
help: consider `await`ing on the `Future`
|
4 | 2 => other().await.await,
| ++++++
error: aborting due to previous error
For more information about this error, try `rustc --explain E0308`.
```
After:
```
➜ ~ rustc +stage1 --edition=2021 --crate-type=lib test.rs
error[E0308]: `match` arms have incompatible types
--> test.rs:4:14
|
2 | let x = match 1 {
| _____________-
3 | | 1 => other(),
| | ------- this is found to be of type `impl Future`
4 | | 2 => other().await,
| | ^^^^^^^^^^^^^ expected opaque type, found enum `Result`
5 | | };
| |_____- `match` arms have incompatible types
|
= note: expected type `impl Future`
found enum `Result<(), ()>`
error: aborting due to previous error
For more information about this error, try `rustc --explain E0308`.
```
Matthias Krüger [Tue, 16 Nov 2021 22:58:21 +0000 (23:58 +0100)]
Rollup merge of #90733 - wesleywiser:musl_debuginfo, r=Mark-Simulacrum
Build musl dist artifacts with debuginfo enabled
Since our musl targets link to a version of musl we build and bundle
with the targets, if users need to debug into musl or generate
backtraces which contain parts of the musl library, they will be unable
to do so unless we enable and ship the debug info.
This patch changes our dist builds so they enabled debug info when
building musl. This patch also includes a fix for CFI detection in
musl's `configure` script which has been [posted upstream](https://www.openwall.com/lists/musl/2021/10/21/2).
The net effect of this is that we now ship debug info for musl in those
targets. This adds ~90kb to those artifacts but running `strip` on
binaries produced removes all of that. For a "hello world" Rust binary
on x86_64, the numbers are:
bors [Tue, 16 Nov 2021 11:48:37 +0000 (11:48 +0000)]
Auto merge of #90919 - nnethercote:rm-DropArena, r=Mark-Simulacrum
Remove `DropArena`.
Most arena-allocate types that impl `Drop` get their own `TypedArena`, but a
few infrequently used ones share a `DropArena`. This sharing adds complexity
but doesn't help performance or memory usage. Perhaps it was more effective in
the past prior to some other improvements to arenas.
This commit removes `DropArena` and the sharing of arenas via the `few`
attribute of the `arena_types` macro. This change removes over 100 lines of
code and nine uses of `unsafe` (one of which affects the parallel compiler) and
makes the remaining code easier to read.
bors [Tue, 16 Nov 2021 08:22:55 +0000 (08:22 +0000)]
Auto merge of #90945 - JohnTitor:rollup-wc35xss, r=JohnTitor
Rollup of 8 pull requests
Successful merges:
- #86455 (check where-clause for explicit `Sized` before suggesting `?Sized`)
- #90801 (Normalize both arguments of `equate_normalized_input_or_output`)
- #90803 (Suggest `&str.chars()` on attempt to `&str.iter()`)
- #90819 (Fixes incorrect handling of TraitRefs when emitting suggestions.)
- #90910 (fix getting the discriminant of a zero-variant enum)
- #90925 (rustc_mir_build: reorder bindings)
- #90928 (Use a different server for checking clock drift)
- #90936 (Add a regression test for #80772)
Yuki Okushi [Tue, 16 Nov 2021 06:59:43 +0000 (15:59 +0900)]
Rollup merge of #90928 - Mark-Simulacrum:fix-date-logging, r=pietroalbini
Use a different server for checking clock drift
The detectportal.firefox.com server seems to return a random-ish date; for
example I see the following across 5 curl's done consecutively locally, where
the real date is approximately 15 Nov 2021 06:36 UTC.
Date: Mon, 15 Nov 2021 13:34:53 GMT
Date: Mon, 15 Nov 2021 12:20:21 GMT
Date: Mon, 15 Nov 2021 00:06:47 GMT
Date: Mon, 15 Nov 2021 17:14:33 GMT
Date: Mon, 15 Nov 2021 13:33:21 GMT
Yuki Okushi [Tue, 16 Nov 2021 06:59:42 +0000 (15:59 +0900)]
Rollup merge of #90925 - krasimirgg:rustc_mir_build_fix, r=petrochenkov
rustc_mir_build: reorder bindings
No functional changes intended.
I'm playing around with building compiler components using nightly rust
(2021-11-02) in a non-standard way. I encountered the following error while
trying to build rustc_mir_build:
```
error[E0597]: `wildcard` does not live long enough
--> rust/src/nightly/compiler/rustc_mir_build/src/build/matches/mod.rs:1767:82
|
1767 | let mut otherwise_candidate = Candidate::new(expr_place_builder.clone(), &wildcard, false);
| ^^^^^^^^^ borrowed value does not live long enough
...
1799 | }
| -
| |
| `wildcard` dropped here while still borrowed
| borrow might be used here, when `guard_candidate` is dropped and runs the destructor for type `Candidate<'_, '_>`
|
= note: values in a scope are dropped in the opposite order they are defined
```
I believe this flags an issue that may become an error in the future.
Swapping the order of `wildcard` and `guard_candidate` resolves it.
Yuki Okushi [Tue, 16 Nov 2021 06:59:40 +0000 (15:59 +0900)]
Rollup merge of #90819 - JakobDegen:issue-90804, r=petrochenkov
Fixes incorrect handling of TraitRefs when emitting suggestions.
Closes #90804 , although there were more issues here that were hidden by the thing that caused this ICE.
Underlying problem was that substitutions were being thrown out, which not only leads to an ICE but also incorrect diagnostics. On top of that, in some cases the self types from the root obligations were being mixed in with those from derived obligations.
This makes a couple diagnostics arguable worse ("`B<C>` does not implement `Copy`" instead of "`C` does not implement `Copy`") but the worse diagnostics are at least still correct and that downside is in my opinion clearly outweighed by the benefits of fixing the ICE and unambiguously wrong diagnostics.
bors [Tue, 16 Nov 2021 05:18:57 +0000 (05:18 +0000)]
Auto merge of #90845 - JakobDegen:adt-drop-perf, r=Mark-Simulacrum
Address performance regression introduced by #90218
As part of the changes in #90218 , the `adt_drop_tys` and friends code stopped recursing through the query system, meaning that intermediate computations did not get cached. This change adds the recursions back in without re-introducing any of the old issues.
On local benchmarks this fixes the 5% regressions in #90504 ; the wg-grammar regressions didn't seem to move too much. I may take some time later to look into those.
Not sure who to request for review here, so will leave it up to whoever gets it.
bors [Tue, 16 Nov 2021 02:23:42 +0000 (02:23 +0000)]
Auto merge of #90934 - JohnTitor:rollup-5soqo0j, r=JohnTitor
Rollup of 10 pull requests
Successful merges:
- #85766 (Stabilize File::options())
- #88601 (Implement `Termination` for `Result<Infallible, E>`)
- #90058 (Stabilize -Z strip as -C strip)
- #90790 (Fix standard library test with read_link)
- #90834 (Android is not GNU)
- #90835 (Rename WASI's `is_character_device` to `is_char_device`.)
- #90837 (Move some tests to more reasonable directories - 9)
- #90848 (Remove bigint_helper_methods for *signed* types)
- #90892 (fix ICE on Miri/CTFE copy of half a pointer)
- #90909 (disable portable SIMD tests in Miri)
Failed merges:
- #90128 (Stabilize -Z symbol-mangling-version=v0 as -C symbol-mangling-version=v0)
Yuki Okushi [Tue, 16 Nov 2021 00:14:21 +0000 (09:14 +0900)]
Rollup merge of #90848 - scottmcm:remove-signed-bigint-helpers, r=joshtriplett
Remove bigint_helper_methods for *signed* types
This PR inspired by `@cuviper's` comment @ https://github.com/rust-lang/rust/issues/90541#issuecomment-967309808
These are working well for *unsigned* types, so keep those, but for the the *signed* ones there are a bunch of questions about what the semantics and API should be. For the main "helpers for big integer implementations" use, there's no need for the signed versions anyway. There are plenty of other methods which exist for unsigned types but not signed ones, like `next_power_of_two`, so this isn't unusual.
Yuki Okushi [Tue, 16 Nov 2021 00:14:18 +0000 (09:14 +0900)]
Rollup merge of #90834 - cuviper:android-gnu, r=petrochenkov
Android is not GNU
For a long time, the Android targets had `target_env=""`, but this changed to `"gnu"` in Rust 1.49.0. I tracked this down to #77729 which started setting `"gnu"` in the `linux_base` target options, and this was inherited by `android_base`. Then #78929 split the env into `linux_gnu_base`, but `android_base` was also changed to follow that. Android was not specifically mentioned in either pull request, so I believe this was an accident. Moving it back to `linux_base` will use an empty `env` again.
r? ````@Mark-Simulacrum````
cc ````@petrochenkov````
bors [Mon, 15 Nov 2021 23:27:59 +0000 (23:27 +0000)]
Auto merge of #90827 - matthewjasper:assoc-item-cleanup-2, r=cjgillot
Assoc item cleanup Part 2
- Remove `AssocItem` from `RegionVariableOrigin::AutoRef`
- Use the `associated_item_def_ids` query instead of the `associated_items` query when possible
The change to `ObligationCauseCode` from #90639 is omitted because it caused a perf regression.
bors [Mon, 15 Nov 2021 20:19:23 +0000 (20:19 +0000)]
Auto merge of #90821 - scottmcm:new-slice-reverse, r=Mark-Simulacrum
MIRI says `reverse` is UB, so replace it with something LLVM can vectorize
For small types with padding, the current implementation is UB because it does integer operations on uninit values.
```
error: Undefined Behavior: using uninitialized data, but this operation requires initialized memory
--> /playground/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/num/mod.rs:836:5
|
836 | / uint_impl! { u32, u32, i32, 32, 4294967295, 8, "0x10000b3", "0xb301", "0x12345678",
837 | | "0x78563412", "0x1e6a2c48", "[0x78, 0x56, 0x34, 0x12]", "[0x12, 0x34, 0x56, 0x78]", "", "" }
| |________________________________________________________________________________________________^ using uninitialized data, but this operation requires initialized memory
|
= help: this indicates a bug in the program: it performed an invalid operation, and caused Undefined Behavior
= help: see https://doc.rust-lang.org/nightly/reference/behavior-considered-undefined.html for further information
= note: inside `core::num::<impl u32>::rotate_left` at /playground/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/num/uint_macros.rs:211:13
= note: inside `core::slice::<impl [Foo]>::reverse` at /playground/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/slice/mod.rs:701:58
```
<https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=340739f22ca5b457e1da6f361768edc6>
But LLVM has gotten smarter since I wrote the previous implementation in 2017, so this PR removes all the manual magic and just writes it in such a way that LLVM will vectorize. This code is much simpler and has very little `unsafe`, and is actually faster to boot!
If you're curious to see the codegen: <https://rust.godbolt.org/z/Pcn13Y9E3>
Mark Rousskov [Mon, 15 Nov 2021 18:35:06 +0000 (13:35 -0500)]
Use a different server for checking clock drift
The detectportal.firefox.com server seems to return a random-ish date; for
example I see the following across 5 curl's done consecutively locally, where
the real date is approximaly 15 Nov 2021 06:36 UTC.
Date: Mon, 15 Nov 2021 13:34:53 GMT
Date: Mon, 15 Nov 2021 12:20:21 GMT
Date: Mon, 15 Nov 2021 00:06:47 GMT
Date: Mon, 15 Nov 2021 17:14:33 GMT
Date: Mon, 15 Nov 2021 13:33:21 GMT
- We don't support capturing arguments from a non-literal format string like `format_args!(concat!(...))`. We could add that in a future enhancement, or we can decide that it isn't supported (as suggested in https://github.com/rust-lang/rust/issues/67984#issuecomment-801394736 ).
- I've updated the documentation.
- `panic!` now supports capture as well.
- There are potentially opportunities to further improve diagnostics for invalid usage, such as if it looks like the user tried to use an expression rather than a variable. However, such cases are all already caught and provide reasonable syntax errors now, and we can always provided even friendlier diagnostics in the future.
I'm playing around with building compiler components using nightly rust
(2021-11-02) in a non-standard way. I encountered the following error while
trying to build rustc_mir_build:
```
error[E0597]: `wildcard` does not live long enough
--> rust/src/nightly/compiler/rustc_mir_build/src/build/matches/mod.rs:1767:82
|
1767 | let mut otherwise_candidate = Candidate::new(expr_place_builder.clone(), &wildcard, false);
| ^^^^^^^^^ borrowed value does not live long enough
...
1799 | }
| -
| |
| `wildcard` dropped here while still borrowed
| borrow might be used here, when `guard_candidate` is dropped and runs the destructor for type `Candidate<'_, '_>`
|
= note: values in a scope are dropped in the opposite order they are defined
```
I believe this flags an issue that may become an error in the future.
Swapping the order of `wildcard` and `guard_candidate` resolves it.
bors [Mon, 15 Nov 2021 11:18:44 +0000 (11:18 +0000)]
Auto merge of #90717 - kit-981:fix-ld64-flags, r=petrochenkov
Fix ld64 flags
- The `-exported_symbols_list` argument appears to be malformed for `ld64` (if you are not going through `clang`).
- The `-dynamiclib` argument isn't support for `ld64`. It should be guarded behind a compiler flag.
These problems are fixed by these changes. I have also refactored the way linker arguments are generated to be ld/compiler agnostic and therefore less error prone.
These changes are necessary to support cross-compilation to darwin targets.
Josh Triplett [Thu, 21 Oct 2021 11:19:46 +0000 (13:19 +0200)]
Stabilize -Z strip as -C strip
Leave -Z strip available temporarily as an alias, to avoid breaking
cargo until cargo transitions to using -C strip. (If the user passes
both, the -C version wins.)
Most arena-allocate types that impl `Drop` get their own `TypedArena`, but a
few infrequently used ones share a `DropArena`. This sharing adds complexity
but doesn't help performance or memory usage. Perhaps it was more effective in
the past prior to some other improvements to arenas.
This commit removes `DropArena` and the sharing of arenas via the `few`
attribute of the `arena_types` macro. This change removes over 100 lines of
code and nine uses of `unsafe` (one of which affects the parallel compiler) and
makes the remaining code easier to read.
bors [Mon, 15 Nov 2021 03:19:01 +0000 (03:19 +0000)]
Auto merge of #90684 - jyn514:dist-aliases, r=Mark-Simulacrum
Change paths for `dist` command to match the components they generate
Before, you could have the confusing situation where the command to
generate a component had no relation to the name of that component (e.g.
the `rustc` component was generated with `src/librustc`). This changes
the name to make them match up.
bors [Sun, 14 Nov 2021 18:47:42 +0000 (18:47 +0000)]
Auto merge of #88282 - Neutron3529:patch-4, r=Mark-Simulacrum
Optimize BinaryHeap::extend from Vec
This improves the performance of extending `BinaryHeap`s from vectors directly. Future work may involve extending this optimization to other, similar, cases where the length of the added elements is well-known, but this is not yet done in this PR.
Nilstrieb [Sat, 13 Nov 2021 22:56:22 +0000 (23:56 +0100)]
Fix span for non-satisfied trivial trait bounds
The spans for "trait bound not satisfied" errors in trivial trait bounds referenced the entire item (fn, impl, struct) before.
Now they only reference the obligation itself (`String: Copy`)
kit [Mon, 8 Nov 2021 23:36:47 +0000 (10:36 +1100)]
Refactor linker argument generation
This commit refactors linker argument generation to leverage a helper
function that abstracts away details governing how these arguments are
transformed and provided to the linker.
This fixes the misuse of the `-exported_symbols_list` when an ld-like
linker is used rather than a compiler. A compiler would expect
`-Wl,-exported_symbols_list,path` but ld would expect
`-exported_symbols_list` and `path` as two seperate arguments. Prior
to this change, an ld-like linker was given
`-exported_symbols_list,path`.
kit [Mon, 8 Nov 2021 09:45:04 +0000 (20:45 +1100)]
Add a helper method for linker arguments
Linker arguments must transformed when Rust is interacting with the
linker through a compiler. This commit introduces a helper function
that abstracts away details of this transformation.
bors [Sun, 14 Nov 2021 01:06:05 +0000 (01:06 +0000)]
Auto merge of #90883 - matthiaskrgr:rollup-iu9k5pe, r=matthiaskrgr
Rollup of 3 pull requests
Successful merges:
- #90771 (Fix trait object error code)
- #90840 (relate lifetime in `TypeOutlives` bounds on drop impls)
- #90853 (rustdoc: Use an empty Vec instead of Option<Vec>)
Jakob Degen [Thu, 11 Nov 2021 23:28:19 +0000 (18:28 -0500)]
Fix handling of substitutions and binders when deciding whether to suggest references
When suggesting references, substitutions were being forgotten and some types were misused. This led to at
least one ICE and other incorrectly emitted diagnostics. This has been fixed; in some cases this leads to
diagnostics changing, and tests have been adjusted.
bors [Sat, 13 Nov 2021 17:10:15 +0000 (17:10 +0000)]
Auto merge of #89551 - jhpratt:stabilize-const_raw_ptr_deref, r=oli-obk
Stabilize `const_raw_ptr_deref` for `*const T`
This stabilizes dereferencing immutable raw pointers in const contexts.
It does not stabilize `*mut T` dereferencing. This is behind the
same feature gate as mutable references.
Since it is unstable, doc-hidden and has no associated tracking issue it was never meant for public use. And since
it is no longer used outside alloc itself it can be made private again.
Also remove some functions that are dead due to lack of internal users.
Joshua Nelson [Mon, 8 Nov 2021 04:37:28 +0000 (04:37 +0000)]
Change paths for `dist` command to match the components they generate
Before, you could have the confusing situation where the command to
generate a component had no relation to the name of that component (e.g.
the `rustc` component was generated with `src/librustc`). This changes
the name to make them match up.
bors [Sat, 13 Nov 2021 08:22:52 +0000 (08:22 +0000)]
Auto merge of #87264 - mystor:expand_literal, r=petrochenkov
proc_macro: Add an expand_expr method to TokenStream
This feature is aimed at giving proc macros access to powers similar to those used by builtin macros such as `format_args!` or `concat!`. These macros are able to accept macros in place of string literal parameters, such as the format string, as they perform recursive macro expansion while being expanded.
This can be especially useful in many cases thanks to helper macros like `concat!`, `stringify!` and `include_str!` which are often used to construct string literals at compile-time in user code.
For now, this method only allows expanding macros which produce literals, although more expressions will be supported before the method is stabilized.
In earlier versions of this PR, this method exclusively returned `Literal`, and spans on returned literals were stripped of expansion context before being returned to be as conservative as possible about permission leakage. The method's naming has been generalized to eventually support arbitrary expressions, and the context stripping has been removed (https://github.com/rust-lang/rust/pull/87264#discussion_r674863279), which should allow for more general APIs like "format_args_implicits" (https://github.com/rust-lang/rust/issues/67984) to be supported as well.
bors [Sat, 13 Nov 2021 05:19:39 +0000 (05:19 +0000)]
Auto merge of #90041 - jfrimmel:rt_copy_checks, r=Mark-Simulacrum
Re-enable `copy[_nonoverlapping]()` debug-checks
This commit re-enables the debug checks for valid usages of the two functions `copy()` and `copy_nonoverlapping()`. Those checks were commented out in #79684 in order to make the functions const. All that's been left was a FIXME, that could not be resolved until there is was way to only do the checks at runtime.
Since #89247 there is such a way: `const_eval_select()`. This commit uses that new intrinsic in order to either do nothing (at compile time) or to do the old checks (at runtime).
The change itself is rather small: in order to make the checks usable with `const_eval_select`, they are moved into a local function (one for `copy` and one for `copy_nonoverlapping` to keep symmetry).
The change does not break referential transparency, as there is nothing you can do at compile time, which you cannot do on runtime without getting undefined behavior. The CTFE-engine won't allow missuses. The other way round is also fine.
I've refactored the code to use `#[cfg(debug_assertions)]` on the new items. If that is not desired, the second commit can be dropped.
I haven't added any checks, as I currently don't know, how to test this properly.
Closes #90012.
cc `@rust-lang/lang,` `@rust-lang/libs` and `@rust-lang/wg-const-eval` (as those teams are linked in the issue above).
bors [Sat, 13 Nov 2021 02:17:20 +0000 (02:17 +0000)]
Auto merge of #89167 - workingjubilee:use-simd, r=MarkSimulacrum
pub use core::simd;
A portable abstraction over SIMD has been a major pursuit in recent years for several programming languages. In Rust, `std::arch` offers explicit SIMD acceleration via compiler intrinsics, but it does so at the cost of having to individually maintain each and every single such API, and is almost completely `unsafe` to use. `core::simd` offers safe abstractions that are resolved to the appropriate SIMD instructions by LLVM during compilation, including scalar instructions if that is all that is available.
`core::simd` is enabled by the `#![portable_simd]` nightly feature tracked in https://github.com/rust-lang/rust/issues/86656 and is introduced here by pulling in the https://github.com/rust-lang/portable-simd repository as a subtree. We built the repository out-of-tree to allow faster compilation and a stochastic test suite backed by the proptest crate to verify that different targets, features, and optimizations produce the same result, so that using this library does not introduce any surprises. As these tests are technically non-deterministic, and thus can introduce overly interesting Heisenbugs if included in the rustc CI, they are visible in the commit history of the subtree but do nothing here. Some tests **are** introduced via the documentation, but these use deterministic asserts.
There are multiple unsolved problems with the library at the current moment, including a want for better documentation, technical issues with LLVM scalarizing and lowering to libm, room for improvement for the APIs, and so far I have not added the necessary plumbing for allowing the more experimental or libm-dependent APIs to be used. However, I thought it would be prudent to open this for review in its current condition, as it is both usable and it is likely I am going to learn something else needs to be fixed when bors tries this out.
The major types are
- `core::simd::Simd<T, N>`
- `core::simd::Mask<T, N>`
There is also the `LaneCount` struct, which, together with the SimdElement and SupportedLaneCount traits, limit the implementation's maximum support to vectors we know will actually compile and provide supporting logic for bitmasks. I'm hoping to simplify at least some of these out of the way as the compiler and library evolve.
projects / rust.git / log