Optimize allocation paths in RawVec
Since the `Alloc` trait was introduced (https://github.com/rust-lang/rust/pull/42313) and it was integrated everywhere (https://github.com/rust-lang/rust/pull/42727) there's been some slowdowns and regressions that have slipped through. The intention of this PR is to try to tackle at least some of them, but they've been very difficult to quantify up to this point so it probably doesn't solve everything.
This PR primarily targets the `RawVec` type, specifically the `double` function. The codegen for this function is now much closer to what it was before #42313 landed as many runtime checks have been elided.
None of these require a significant amount of code and using `#[inline]` will
allow constructors to get inlined, improving codegen at allocation callsites.
Add precondition to `Layout` that the `align` fit in a u32.
Add precondition to `Layout` that the `align` not exceed 2^31.
This precondition takes the form of a behavorial change in `Layout::from_size_align` (so it returns `None` if the input `align` is too large) and a new requirement for safe usage of `Layout::from_size_align_unchecked`.
Fix#30170.
This precondition takes the form of a behavorial change in
`Layout::from_size_align` (so it returns `None` if the `align` is too
large) and a new requirement for safe usage of
`Layout::from_size_align_unchecked`.
Fix#30170.
This PR is an implementation of [RFC 1974] which specifies a new method of
defining a global allocator for a program. This obsoletes the old
`#![allocator]` attribute and also removes support for it.
[RFC 1974]: https://github.com/rust-lang/rfcs/pull/197
The new `#[global_allocator]` attribute solves many issues encountered with the
`#![allocator]` attribute such as composition and restrictions on the crate
graph itself. The compiler now has much more control over the ABI of the
allocator and how it's implemented, allowing much more freedom in terms of how
this feature is implemented.
cc #27389
Added `unwrap` calls in all the places where I can infer that the
conditions are met to avoid panic (or when the calling method itself
says it will panic in such a case).
Includes `alloc_zeroed` method that `RawVec` has come to depend on.
Exposed private `Layout::from_size_align` ctor to be `pub`, and added
explicit conditions for when it will panic (namely, when `align` is
not power-of-two, or if rounding up `size` to a multiple of `align`
overflows). Normalized all `Layout` construction to go through
`Layout::from_size_align`.
Addressed review feedback regarding `struct Layout` and zero-sized
layouts.
Restrict specification for `dealloc`, adding additional constraint
that the given alignment has to match that used to allocate the block.
(This is a maximally conservative constraint on the alignment. An open
question to resolve (before stabilization) is whether we can return to
a looser constraint such as the one previously specified.)
Split `fn realloc_in_place` into separate `fn grow_in_place` and `fn
shrink_in_place` methods, which have default impls that check against
usable_size for reuse. Make `realloc` default impl try `grow_in_place`
or `shrink_in_place` as appropriate before fallback on
alloc+copy+dealloc.
Drive-by: When reviewing calls to `padding_needed_for`, discovered
what I think was an over-conservative choice for its argument
alignment. Namely, in `fn extend`, we automatically realign the whole
resulting layout to satisfy both old (self) and new alignments. When
the old alignment exceeds the new, this means we would insert
unnecessary padding. So I changed the code to pass in `next.align`
instead of `new_align` to `padding_needed_for`.
Replaced ref to `realloc_in_place` with `grow_in_place`/`shrink_in_place`.
Revised docs replacing my idiosyncratic style of `fn foo` with just
`foo` when referring to the function or method `foo`.
(Alpha-renamed `Allocator` to `Alloc`.)
Post-rebased, added `Debug` derive for `allocator::Excess` to satisfy
`missing_debug_implementations`.
