Auto merge of #52426 - ljedrz:#28273_cleanup, r=nikomatsakis
Enable default inlining in platform intrinsics
Since [#28273](https://github.com/rust-lang/rust/issues/28273) has been fixed for quite some time, it might be a good idea to return to default inlining in platform intrinsics.
Auto merge of #52353 - alexcrichton:wasm-custom-section, r=eddyb
rustc: Use link_section, not wasm_custom_section
This commit transitions definitions of custom sections on the wasm target from
the unstable `#[wasm_custom_section]` attribute to the
already-stable-for-other-targets `#[link_section]` attribute. Mostly the same
restrictions apply as before, except that this now applies only to statics.
Auto merge of #52342 - nnethercote:CanonicalVar, r=nikomatsakis
Avoid most allocations in `Canonicalizer`.
Extra allocations are a significant cost of NLL, and the most common
ones come from within `Canonicalizer`. In particular, `canonical_var()`
contains this code:
indices
.entry(kind)
.or_insert_with(|| {
let cvar1 = variables.push(info);
let cvar2 = var_values.push(kind);
assert_eq!(cvar1, cvar2);
cvar1
})
.clone()
`variables` and `var_values` are `Vec`s. `indices` is a `HashMap` used
to track what elements have been inserted into `var_values`. If `kind`
hasn't been seen before, `indices`, `variables` and `var_values` all get
a new element. (The number of elements in each container is always the
same.) This results in lots of allocations.
In practice, most of the time these containers only end up holding a few
elements. This PR changes them to avoid heap allocations in the common
case, by changing the `Vec`s to `SmallVec`s and only using `indices`
once enough elements are present. (When the number of elements is small,
a direct linear search of `var_values` is as good or better than a
hashmap lookup.)
The changes to `variables` are straightforward and contained within
`Canonicalizer`. The changes to `indices` are more complex but also
contained within `Canonicalizer`. The changes to `var_values` are more
intrusive because they require defining a new type
`SmallCanonicalVarValues` -- which is to `CanonicalVarValues` as
`SmallVec` is to `Vec -- and passing stack-allocated values of that type
in from outside.
All this speeds up a number of NLL "check" builds, the best by 2%.
fn main() {
stress!{
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
// ... 65536 copies of "a" total ...
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
}
}
```
which takes 50 seconds to compile prior to the fix and <1s after.
I hope this has a visible impact on the compile times for real code. (I think it is most likely to affect incremental TT munchers that deal with large inputs, though it depends on how they are written)
For a fuller description of the performance issue: https://github.com/rust-lang/rust/issues/51754#issuecomment-403242159
---
There is no test (yet) because I'm not sure how easily to measure this for regressions.
Rollup merge of #52434 - Mark-Simulacrum:incremental-keep-stage, r=alexcrichton
Enable incremental independent of stage
Previously we'd only do so for stage 0 but with keep-stage
improvements it seems likely that we'll see more developers working in
the stage 1, so we should allow enabling incremental for them.
Auto merge of #52335 - nnethercote:BitSlice-fixes, r=nikomatsakis
`BitSlice` fixes
`propagate_bits_into_entry_set_for` and `BitSlice::bitwise` are hot for some benchmarks under NLL. I tried and failed to speed them up. (Increasing the size of `bit_slice::Word` from `usize` to `u128` caused a slowdown, even though decreasing the size of `bitvec::Word` from `u128` to `u64` also caused a slowdown. Weird.)
Anyway, along the way I fixed up several problems in and around the `BitSlice` code.
Extra allocations are a significant cost of NLL, and the most common
ones come from within `Canonicalizer`. In particular, `canonical_var()`
contains this code:
indices
.entry(kind)
.or_insert_with(|| {
let cvar1 = variables.push(info);
let cvar2 = var_values.push(kind);
assert_eq!(cvar1, cvar2);
cvar1
})
.clone()
`variables` and `var_values` are `Vec`s. `indices` is a `HashMap` used
to track what elements have been inserted into `var_values`. If `kind`
hasn't been seen before, `indices`, `variables` and `var_values` all get
a new element. (The number of elements in each container is always the
same.) This results in lots of allocations.
In practice, most of the time these containers only end up holding a few
elements. This PR changes them to avoid heap allocations in the common
case, by changing the `Vec`s to `SmallVec`s and only using `indices`
once enough elements are present. (When the number of elements is small,
a direct linear search of `var_values` is as good or better than a
hashmap lookup.)
The changes to `variables` are straightforward and contained within
`Canonicalizer`. The changes to `indices` are more complex but also
contained within `Canonicalizer`. The changes to `var_values` are more
intrusive because they require defining a new type
`SmallCanonicalVarValues` -- which is to `CanonicalVarValues` as
`SmallVec` is to `Vec -- and passing stack-allocated values of that type
in from outside.
All this speeds up a number of NLL "check" builds, the best by 2%.
Auto merge of #52081 - alexcrichton:proc-macro-stable, r=petrochenkov
rustc: Stabilize the `proc_macro` feature
This commit stabilizes some of the `proc_macro` language feature as well as a
number of APIs in the `proc_macro` crate as [previously discussed][1]. This
means that on stable Rust you can now define custom procedural macros which
operate as attributes attached to items or `macro_rules!`-like bang-style
invocations. This extends the suite of currently stable procedural macros,
custom derives, with custom attributes and custom bang macros.
Note though that despite the stabilization in this commit procedural macros are
still not usable on stable Rust. To stabilize that we'll need to stabilize at
least part of the `use_extern_macros` feature. Currently you can define a
procedural macro attribute but you can't import it to call it!
A summary of the changes made in this PR (as well as the various consequences)
is:
* The `proc_macro` language and library features are now stable.
* Other APIs not stabilized in the `proc_macro` crate are now named under a
different feature, such as `proc_macro_diagnostic` or `proc_macro_span`.
* A few checks in resolution for `proc_macro` being enabled have switched over
to `use_extern_macros` being enabled. This means that code using
`#![feature(proc_macro)]` today will likely need to move to
`#![feature(use_extern_macros)]`.
It's intended that this PR, once landed, will be followed up with an attempt to
stabilize a small slice of `use_extern_macros` just for procedural macros to
make this feature 100% usable on stable.
Alex Crichton [Fri, 13 Jul 2018 18:30:47 +0000 (11:30 -0700)]
rustc: Use link_section, not wasm_custom_section
This commit transitions definitions of custom sections on the wasm target from
the unstable `#[wasm_custom_section]` attribute to the
already-stable-for-other-targets `#[link_section]` attribute. Mostly the same
restrictions apply as before, except that this now applies only to statics.
Mark Rousskov [Mon, 16 Jul 2018 15:22:15 +0000 (09:22 -0600)]
Enable incremental independent of stage
Previously we'd only do so for stage 0 but with keep-stage
improvements it seems likely that we'll see more developers working in
the stage 1, so we should allow enabling incremental for them.
Ideally, the check we probably want is to only enable incremental for
the last compiler build scheduled, but there's no good way to do so
today. Just enabling incremental in all stages should be sufficient;
we may be doing extra work that's needles -- compiling incrementally
something that will never be recompiled in-place -- but that should be
sufficiently unlikely (i.e., users either don't care or won't be
compiling the compiler twice).
Alex Crichton [Tue, 3 Jul 2018 22:36:31 +0000 (15:36 -0700)]
rustc: Stabilize much of the `proc_macro` feature
This commit stabilizes some of the `proc_macro` language feature as well as a
number of APIs in the `proc_macro` crate as [previously discussed][1]. This
means that on stable Rust you can now define custom procedural macros which
operate as attributes attached to items or `macro_rules!`-like bang-style
invocations. This extends the suite of currently stable procedural macros,
custom derives, with custom attributes and custom bang macros.
Note though that despite the stabilization in this commit procedural macros are
still not usable on stable Rust. To stabilize that we'll need to stabilize at
least part of the `use_extern_macros` feature. Currently you can define a
procedural macro attribute but you can't import it to call it!
A summary of the changes made in this PR (as well as the various consequences)
is:
* The `proc_macro` language and library features are now stable.
* Other APIs not stabilized in the `proc_macro` crate are now named under a
different feature, such as `proc_macro_diagnostic` or `proc_macro_span`.
* A few checks in resolution for `proc_macro` being enabled have switched over
to `use_extern_macros` being enabled. This means that code using
`#![feature(proc_macro)]` today will likely need to move to
`#![feature(use_extern_macros)]`.
It's intended that this PR, once landed, will be followed up with an attempt to
stabilize a small slice of `use_extern_macros` just for procedural macros to
make this feature 100% usable on stable.
Auto merge of #52401 - semarie:tidy-extdeps, r=alexcrichton
tidy: add a new test for external dependencies
ensure all packages in Cargo.lock will be vendored, and fail if the
source packages isn't whitelisted.
the purpose is to avoid such kind of issues:
- #52029 Rustfmt isn't vendored correctly
- #42719 building beta with vendor=true fail due to network dependencies
as Rust comes with several external dependencies (clippy, miri, rustfmt, rls), it is important to have a way to catch some errors in the update of this submodules.
The new check in tidy quickly reads `Cargo.lock` to search for the `source` of all packages. This attribute is present when the package comes from external source (like `crates.io-index` or some `git` repository). Some sources are whitelisted (like `crates.io-index`) as the crates are vendored.
CTFE: use binary_op to compare integer with match disriminant
This is needed to unblock https://github.com/solson/miri/pull/401: There is code in the Windows initialization functions that uses `match` to test whether a pointer is NULL.
I will add a testcase in miri; I was not sure where to add a testcase in Rust itself.
Auto merge of #52383 - petrochenkov:pmns, r=alexcrichton
resolve: Functions introducing procedural macros reserve a slot in the macro namespace as well
Similarly to https://github.com/rust-lang/rust/pull/52234, this gives us symmetry between internal and external views of a crate, but in this case it's always an error to call a procedural macro in the same crate in which it's defined.
Auto merge of #52372 - KarolinePlum:Duration-rounding-documentation, r=joshtriplett
Document rounding down in std::time::Durations's subsec_millis etc.
Now also the documentations of `subsec_millis`, `subsec_micros`, `as_millis` and `as_micros` make clear that the fractional nanosecond component is rounded down to whole units.
Auto merge of #52361 - QuietMisdreavus:proc-macro-doc, r=ollie27
rustdoc: don't panic when the cross-re-export handler sees a proc-macro
When i moved the macro cross-re-export inlining code into `clean::inline`, i thought that if a macro had a `Def` that said it was a bang macro, it wouldn't be a proc macro. I thought wrong. Turns out, the `quote!()` in `libproc_macro` is actually a proc-macro, and when the `quote!()` macro is re-exported, this proc-macro is accessed in its place. This causes any `proc_macro::*` glob re-export to pull in this proc-macro, causing the assertion i added to fire, leading to an ICE. This replaces that with an Option that ignores proc-macros for the time being.
Auto merge of #52360 - Mark-Simulacrum:fix-keep-stage-for-cg-backends, r=alexcrichton
Do not attempt to recompile codegen backend(s) with --keep-stage
Previously we'd attempt to recompile them and that would fail since
we've essentially not built the entire compiler yet, or we're faking
that fact. This commit should make us ignore the codegen backend build
as well.
Unlike the other compile steps, there is no CodegenBackendLink step that
we run here, because that is done later as a part of assembling the
final compiler and as an explicit function call.
r? @alexcrichton
I think this may fix or at least assist with #52174.
cc @RalfJung @tinco -- if you can test this patch locally that'd be
amazing; I don't want to recompile for the next couple hours to test it
locally. I don't think it can make the situation worse, and in fact, if
I've interpreted the cause of the failure correctly then this will fix
your problem.
Auto merge of #52326 - alexcrichton:tweak-proc-macro-expand, r=petrochenkov
rustc: Tweak expansion of #[proc_macro] for 2018
The syntactical expansion of `#[proc_macro]` and related attributes currently
contains absolute paths which conflicts with a lint for the 2018 edition,
causing issues like #52214. This commit puts a band-aid on the issue by ensuring
that procedural macros can also migrate to the 2018 edition for now by tweaking
the expansion based on what features are activated. A more long-term solution
would probably tweak the edition hygiene of spans, but this should do the trick
for now.
Make rounding down clear in duration documentation
Now also the documentations of `subsec_millis`, `subsec_micros`, `as_millis` and `as_micros` make clear that the fractional nanosecond component is rounded down to whole units.