Magic-Mirror/llvm-capstone
1
0
Fork 0
mirror of https://github.com/capstone-engine/llvm-capstone.git synced 2024-11-27 07:31:28 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[libc++] Speed up classic locale (#72112)

Browse Source 
Locale objects use atomic reference counting, which may be very
expensive in parallel applications. The classic locale is used by
default by all streams and can be very contended. But it's never
destroyed, so the reference counting is also completely pointless on the
classic locale. Currently ~70% of time in the parallel stringstream
benchmarks is spent in locale ctor/dtor. And the execution radically
slows down with more threads.

Avoid reference counting on the classic locale. With this change
parallel benchmarks start to scale with threads.

Co-authored-by: Louis Dionne <ldionne.2@gmail.com>

```
                              │   baseline   │    optimized                            │
                              │    sec/op    │    sec/op      vs base                  │
Istream_numbers/0/threads:1      4.672µ ± 0%   4.419µ ± 0%     -5.42% (p=0.000 n=30+39)
Istream_numbers/0/threads:72   539.817µ ± 0%   9.842µ ± 1%    -98.18% (p=0.000 n=30+40)
Istream_numbers/1/threads:1      4.890µ ± 0%   4.750µ ± 0%     -2.85% (p=0.000 n=30+40)
Istream_numbers/1/threads:72     66.44µ ± 1%   10.14µ ± 1%    -84.74% (p=0.000 n=30+40)
Istream_numbers/2/threads:1      4.888µ ± 0%   4.746µ ± 0%     -2.92% (p=0.000 n=30+40)
Istream_numbers/2/threads:72     494.8µ ± 0%   410.2µ ± 1%    -17.11% (p=0.000 n=30+40)
Istream_numbers/3/threads:1      4.697µ ± 0%   4.695µ ± 5%          ~ (p=0.391 n=30+37)
Istream_numbers/3/threads:72     421.5µ ± 7%   421.9µ ± 9%          ~ (p=0.665 n=30)
Ostream_number/0/threads:1       183.0n ± 0%   141.0n ± 2%    -22.95% (p=0.000 n=30)
Ostream_number/0/threads:72    24196.5n ± 1%   343.5n ± 3%    -98.58% (p=0.000 n=30)
Ostream_number/1/threads:1       250.0n ± 0%   196.0n ± 2%    -21.60% (p=0.000 n=30)
Ostream_number/1/threads:72    16260.5n ± 0%   407.0n ± 2%    -97.50% (p=0.000 n=30)
Ostream_number/2/threads:1       254.0n ± 0%   196.0n ± 1%    -22.83% (p=0.000 n=30)
Ostream_number/2/threads:72      28.49µ ± 1%   18.89µ ± 5%    -33.72% (p=0.000 n=30)
Ostream_number/3/threads:1       185.0n ± 0%   185.0n ± 0%      0.00% (p=0.017 n=30)
Ostream_number/3/threads:72      19.38µ ± 4%   19.33µ ± 5%          ~ (p=0.425 n=30)
```
This commit is contained in:
Dmitry Vyukov 2023-11-27 07:00:21 +01:00 committed by GitHub
parent e89324219a
commit f8afc53d64
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 164 additions and 55 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

63
libcxx/benchmarks/stringstream.bench.cpp
View File

@ -1,11 +1,12 @@
#include "benchmark/benchmark.h"
#include "test_macros.h"
#include <mutex>
#include <sstream>
TEST_NOINLINE double istream_numbers();
double istream_numbers() {
double istream_numbers(std::locale* loc) {
const char* a[] = {"-6 69 -71 2.4882e-02 -100 101 -2.00005 5000000 -50000000",
"-25 71 7 -9.3262e+01 -100 101 -2.00005 5000000 -50000000",
"-14 53 46 -6.7026e-02 -100 101 -2.00005 5000000 -50000000"};
@ -14,17 +15,73 @@ double istream_numbers() {
double f1 = 0.0, f2 = 0.0, q = 0.0;
for (int i = 0; i < 3; i++) {
std::istringstream s(a[i]);
if (loc)
s.imbue(*loc);
s >> a1 >> a2 >> a3 >> f1 >> a4 >> a5 >> f2 >> a6 >> a7;
q += (a1 + a2 + a3 + a4 + a5 + a6 + a7 + f1 + f2) / 1000000;
}
return q;
}
struct LocaleSelector {
std::locale* imbue;
std::locale old;
static std::mutex mutex;
LocaleSelector(benchmark::State& state) {
std::lock_guard guard(mutex);
switch (state.range(0)) {
case 0: {
old = std::locale::global(std::locale::classic());
imbue = nullptr;
break;
}
case 1: {
old = std::locale::global(std::locale::classic());
thread_local std::locale loc("en_US.UTF-8");
imbue = &loc;
break;
}
case 2: {
old = std::locale::global(std::locale::classic());
static std::locale loc("en_US.UTF-8");
imbue = &loc;
break;
}
case 3: {
old = std::locale::global(std::locale("en_US.UTF-8"));
imbue = nullptr;
break;
}
}
}
~LocaleSelector() {
std::lock_guard guard(mutex);
std::locale::global(old);
}
};
std::mutex LocaleSelector::mutex;
static void BM_Istream_numbers(benchmark::State& state) {
LocaleSelector sel(state);
double i = 0;
while (state.KeepRunning())
benchmark::DoNotOptimize(i += istream_numbers());
benchmark::DoNotOptimize(i += istream_numbers(sel.imbue));
}
BENCHMARK(BM_Istream_numbers)->DenseRange(0, 3)->UseRealTime()->Threads(1)->ThreadPerCpu();
static void BM_Ostream_number(benchmark::State& state) {
LocaleSelector sel(state);
while (state.KeepRunning()) {
std::ostringstream ss;
if (sel.imbue)
ss.imbue(*sel.imbue);
ss << 0;
benchmark::DoNotOptimize(ss.str().c_str());
}
}
BENCHMARK(BM_Ostream_number)->DenseRange(0, 3)->UseRealTime()->Threads(1)->ThreadPerCpu();
BENCHMARK(BM_Istream_numbers)->RangeMultiplier(2)->Range(1024, 4096);
BENCHMARK_MAIN();

1
libcxx/include/CMakeLists.txt
View File

@ -849,6 +849,7 @@ set(files
__utility/integer_sequence.h
__utility/is_pointer_in_range.h
__utility/move.h
__utility/no_destroy.h
__utility/pair.h
__utility/piecewise_construct.h
__utility/priority_tag.h

1
libcxx/include/__locale
View File

@ -15,6 +15,7 @@
#include <__memory/shared_ptr.h> // __shared_count
#include <__mutex/once_flag.h>
#include <__type_traits/make_unsigned.h>
#include <__utility/no_destroy.h>
#include <cctype>
#include <clocale>
#include <cstdint>

56
libcxx/include/__utility/no_destroy.h Normal file
View File

@ -0,0 +1,56 @@
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef _LIBCPP___UTILITY_NO_DESTROY_H
#define _LIBCPP___UTILITY_NO_DESTROY_H
#include <__config>
#include <__utility/forward.h>
#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
# pragma GCC system_header
#endif
_LIBCPP_BEGIN_NAMESPACE_STD
struct __uninitialized_tag {};
// This class stores an object of type _Tp but never destroys it.
//
// This is akin to using __attribute__((no_destroy)), except that it is possible
// to control the lifetime of the object with more flexibility by deciding e.g.
// whether to initialize the object at construction or to defer to a later
// initialization using __emplace.
template <class _Tp>
struct __no_destroy {
_LIBCPP_HIDE_FROM_ABI explicit __no_destroy(__uninitialized_tag) {}
_LIBCPP_HIDE_FROM_ABI ~__no_destroy() {
// nothing
}
template <class... _Args>
_LIBCPP_HIDE_FROM_ABI explicit __no_destroy(_Args&&... __args) : __obj_(std::forward<_Args>(__args)...) {}
template <class... _Args>
_LIBCPP_HIDE_FROM_ABI _Tp& __emplace(_Args&&... __args) {
new (&__obj_) _Tp(std::forward<_Args>(__args)...);
return __obj_;
}
_LIBCPP_HIDE_FROM_ABI _Tp& __get() { return __obj_; }
_LIBCPP_HIDE_FROM_ABI _Tp const& __get() const { return __obj_; }
private:
union {
_Tp __obj_;
};
};
_LIBCPP_END_NAMESPACE_STD
#endif // _LIBCPP___UTILITY_NO_DESTROY_H

1
libcxx/include/module.modulemap.in
View File

@ -2087,6 +2087,7 @@ module std_private_utility_move [system] {
export std_private_type_traits_is_nothrow_move_constructible
export std_private_type_traits_remove_reference
}
module std_private_utility_no_destroy [system] { header "__utility/no_destroy.h" }
module std_private_utility_pair [system] {
header "__utility/pair.h"
export std_private_ranges_subrange_fwd

97
libcxx/src/locale.cpp
View File

@ -6,6 +6,7 @@
//
//===----------------------------------------------------------------------===//
#include <__utility/no_destroy.h>
#include <algorithm>
#include <clocale>
#include <codecvt>
@ -81,9 +82,8 @@ locale_t __cloc() {
namespace {
struct release
{
void operator()(locale::facet* p) {p->__release_shared();}
struct releaser {
void operator()(locale::facet* p) { p->__release_shared(); }
};
template <class T, class ...Args>
@ -155,7 +155,11 @@ public:
{return static_cast<size_t>(id) < facets_.size() && facets_[static_cast<size_t>(id)];}
const locale::facet* use_facet(long id) const;
private:
void acquire();
void release();
static __no_destroy<__imp> classic_locale_imp_;
private:
void install(facet* f, long id);
template <class F> void install(F* f) {install(f, f->id.__get());}
template <class F> void install_from(const __imp& other);
@ -500,7 +504,7 @@ locale::__imp::__imp(const __imp& other, facet* f, long id)
name_("*")
{
f->__add_shared();
unique_ptr<facet, release> hold(f);
unique_ptr<facet, releaser> hold(f);
facets_ = other.facets_;
for (unsigned i = 0; i < other.facets_.size(); ++i)
if (facets_[i])
@ -519,7 +523,7 @@ void
locale::__imp::install(facet* f, long id)
{
f->__add_shared();
unique_ptr<facet, release> hold(f);
unique_ptr<facet, releaser> hold(f);
if (static_cast<size_t>(id) >= facets_.size())
facets_.resize(static_cast<size_t>(id+1));
if (facets_[static_cast<size_t>(id)])
@ -537,89 +541,78 @@ locale::__imp::use_facet(long id) const
// locale
// This class basically implements __attribute__((no_destroy)), which isn't supported
// by GCC as of writing this.
template <class T>
struct __no_destroy {
template <class... Args>
explicit __no_destroy(Args&&... args) {
T* obj = reinterpret_cast<T*>(&buf);
new (obj) T(std::forward<Args>(args)...);
}
T& get() { return *reinterpret_cast<T*>(&buf); }
T const& get() const { return *reinterpret_cast<T const*>(&buf); }
private:
alignas(T) byte buf[sizeof(T)];
};
// We don't do reference counting on the classic locale.
// It's never destroyed anyway, but atomic reference counting may be very
// expensive in parallel applications. The classic locale is used by default
// in all streams. Note: if a new global locale is installed, then we lose
// the benefit of no reference counting.
__no_destroy<locale::__imp> locale::__imp::classic_locale_imp_(__uninitialized_tag{}); // initialized below in classic()
const locale& locale::classic() {
static const __no_destroy<locale> c(__private_tag{}, &make<__imp>(1u));
return c.get();
static const __no_destroy<locale> classic_locale(__private_tag{}, [] {
// executed exactly once on first initialization of `classic_locale`
locale::__imp::classic_locale_imp_.__emplace(1u);
return &locale::__imp::classic_locale_imp_.__get();
}());
return classic_locale.__get();
}
locale& locale::__global() {
static __no_destroy<locale> g(locale::classic());
return g.get();
static __no_destroy<locale> g(locale::classic());
return g.__get();
}
locale::locale() noexcept
: __locale_(__global().__locale_)
{
__locale_->__add_shared();
void locale::__imp::acquire() {
if (this != &locale::__imp::classic_locale_imp_.__get())
__add_shared();
}
locale::locale(const locale& l) noexcept
: __locale_(l.__locale_)
{
__locale_->__add_shared();
void locale::__imp::release() {
if (this != &locale::__imp::classic_locale_imp_.__get())
__release_shared();
}
locale::~locale()
{
__locale_->__release_shared();
}
locale::locale() noexcept : __locale_(__global().__locale_) { __locale_->acquire(); }
locale::locale(const locale& l) noexcept : __locale_(l.__locale_) { __locale_->acquire(); }
locale::~locale() { __locale_->release(); }
const locale&
locale::operator=(const locale& other) noexcept
{
other.__locale_->__add_shared();
__locale_->__release_shared();
__locale_ = other.__locale_;
return *this;
other.__locale_->acquire();
__locale_->release();
__locale_ = other.__locale_;
return *this;
}
locale::locale(const char* name)
: __locale_(name ? new __imp(name)
: (__throw_runtime_error("locale constructed with null"), nullptr))
{
__locale_->__add_shared();
__locale_->acquire();
}
locale::locale(const string& name)
: __locale_(new __imp(name))
{
__locale_->__add_shared();
}
locale::locale(const string& name) : __locale_(new __imp(name)) { __locale_->acquire(); }
locale::locale(const locale& other, const char* name, category c)
: __locale_(name ? new __imp(*other.__locale_, name, c)
: (__throw_runtime_error("locale constructed with null"), nullptr))
{
__locale_->__add_shared();
__locale_->acquire();
}
locale::locale(const locale& other, const string& name, category c)
: __locale_(new __imp(*other.__locale_, name, c))
{
__locale_->__add_shared();
__locale_->acquire();
}
locale::locale(const locale& other, const locale& one, category c)
: __locale_(new __imp(*other.__locale_, *one.__locale_, c))
{
__locale_->__add_shared();
__locale_->acquire();
}
string
@ -635,7 +628,7 @@ locale::__install_ctor(const locale& other, facet* f, long id)
__locale_ = new __imp(*other.__locale_, f, id);
else
__locale_ = other.__locale_;
__locale_->__add_shared();
__locale_->acquire();
}
locale