This patch documents the underlying API for implementing atomics on a
platform.
This doesn't change the operations that std::atomic is based on, but it
reorganizes the C11 / GCC implementation split to make it clearer what's
the base support layer and what's not.
Put _LIBCPP_NODEBUG on the new allocator trait aliases introduced in
https://github.com/llvm/llvm-project/pull/115654. This prevents a large
increase in the gdb_index size that was introduced by that PR.
While reference comparators are a terrible idea and it's not entirely
clear whether they are supported, fixing the unintended ABI break is
straightforward so we should do it as a first step.
Fixes#118559
Since we changed the implementation of `reserve(size_type)` to only ever
extend,
it doesn't make a ton of sense anymore to have `__shrink_or_extend`,
since the code
paths of `reserve` and `shrink_to_fit` are now almost completely
separate.
This patch splits up `__shrink_or_extend` so that the individual parts
are in `reserve`
and `shrink_to_fit` depending on where they are needed.
The commit where I switched from using the old locale base API to the
new functions defined inside the __locale namespace forgot to update
references to the strtoNUM functions. This patch fixes that.
This PR optimizes the input iterator overload of `assign(_InputIterator,
_InputIterator)` in `std::vector<_Tp, _Allocator>` by directly assigning
to already initialized memory, rather than first destroying existing
elements and then constructing new ones. By eliminating unnecessary
destruction and construction, the proposed algorithm enhances the
performance by up to 2x for trivial element types (e.g.,
`std::vector<int>`), up to 2.6x for non-trivial element types like
`std::vector<std::string>`, and up to 3.4x for more complex non-trivial
types (e.g., `std::vector<std::vector<int>>`).
### Google Benchmarks
Benchmark tests (`libcxx/test/benchmarks/vector_operations.bench.cpp`)
were conducted for the `assign()` implementations before and after this
patch. The tests focused on trivial element types like
`std::vector<int>`, and non-trivial element types such as
`std::vector<std::string>` and `std::vector<std::vector<int>>`.
#### Before
```
-------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations
-------------------------------------------------------------------------------------------------
BM_AssignInputIterIter/vector_int/1024/1024 1157 ns 1169 ns 608188
BM_AssignInputIterIter<32>/vector_string/1024/1024 14559 ns 14710 ns 47277
BM_AssignInputIterIter<32>/vector_vector_int/1024/1024 26846 ns 27129 ns 25925
```
#### After
```
-------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations
-------------------------------------------------------------------------------------------------
BM_AssignInputIterIter/vector_int/1024/1024 561 ns 566 ns 1242251
BM_AssignInputIterIter<32>/vector_string/1024/1024 5604 ns 5664 ns 128365
BM_AssignInputIterIter<32>/vector_vector_int/1024/1024 7927 ns 8012 ns 88579
```
Now that we've dropped support for older C++ dialects in the
synchronization library, we can use lambdas to clarify some of the code
used to implement atomic_wait.
This simplifies the implementation a bit, since we don't need a lot of
the `__has_x` classes anymore. We just need two template aliases to
implement the `allocator_traits` aliases now.
This PR fixes the issue where `__split_buffer::shrink_to_fit` may
unexpectedly increase the capacity, similar to the issue for
`std::vector` in #97895. The fix follows the same approach
used in #97895 for `std::vector`.
This PR simplifies the implementation of std::vector's move constructor
with an alternative allocator by invoking __init_with_size() instead of
calling assign(), which ultimately calls __assign_with_size(). The
advantage of using __init_with_size() lies in its internal use of
an exception guard, which simplifies the code. Furthermore, from a
semantic standpoint, it is more intuitive for a constructor to call
an initialization function than an assignment function.
The __atomic_base base class is only useful to conditionalize the
operations we provide inside std::atomic. It shouldn't be used directly
from other places in the library which can use std::atomic directly
instead.
Since we've granularized our includes, using std::atomic directly should
not make much of a difference compile-time wise.
This patch starts using std::atomic directly from other classes like
std::barrier and std::latch. Changing this shouldn't be an ABI break
since both classes have the same size and layout.
The benefits of this patch are isolating other parts of the code base
from implementation details of std::atomic and simplifying the mental
model for std::atomic's layers of implementation by making it clear that
__atomic_base is only an implementation detail of std::atomic.
Now that we don't use __compressed_pair anymore inside std::vector, we
can remove some unnecessary accessors. This is a mechanical replacement
of the __alloc() and __end_cap() accessors, and similar for
std::vector<bool>.
Note that I consistently used this->__alloc_ instead of just __alloc_
since most of the code in <vector> uses that pattern to access members.
I don't think this is necessary anymore (and I'm even less certain I
like this), but I went for consistency with surrounding code. If we want
to change that, we can do a follow-up mechanical change.
This generalizes the algorithm a bit. Unfortunately, we can't make
the call sites cleaner inside std::vector because the arguments being
passed can all be fancy pointers, which may not be contiguous iterators.
`__shared_count` is used in a few places where `shared_ptr` isn't. This
avoids a bunch of transitive includes needed for the implementation of
`shared_ptr` in these places.
The variables are all `constexpr`, which implies `inline`. Since they
aren't `constexpr` in C++03 they're also not `inline` there. Because of
that we define them out-of-line currently. Instead we can use the C++17
extension of `inline` variables, which results in the same weak
definitions of the variables but without having all the boilerplate.
Related to PR #114423, this PR proposes to unify the naming of the
internal pointer members in `std::vector` and `std::__split_buffer` for
consistency and clarity.
Both `std::vector` and `std::__split_buffer` originally used a
`__compressed_pair<pointer, allocator_type>` member named `__end_cap_`
to store an internal capacity pointer and an allocator. However,
inconsistent naming changes have been made in both classes:
- `std::vector` now uses `__cap_` and `__alloc_` for its internal
pointer and allocator members.
- In contrast, `std::__split_buffer` retains the name `__end_cap_` for
the capacity pointer, along with `__alloc_`.
This inconsistency between the names `__cap_` and `__end_cap_` has
caused confusions (especially to myself when I was working on both
classes). I suggest unifying these names by renaming `__end_cap_` to
`__cap_` in `std::__split_buffer`.
`std::copy` doesn't use the `_AlgPolicy` for anything other than calling
itself with it, so we can just remove the argument. This also removes
the need in a few other algorithms which had an `_AlgPolicy` argument
only to call `copy`.
This introduces a new `__scope_guard` without any fancy features. The
scope guard is used in `<string>` to simplify some of the ASan
annotations (especially by making it harder to forget them where
exceptions are thrown).
