Abseil Common Libraries (C++) (grcp 依赖) https://abseil.io/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1130 lines
44 KiB

Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// Copyright 2017 The Abseil Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
// -----------------------------------------------------------------------------
// mutex.h
// -----------------------------------------------------------------------------
//
// This header file defines a `Mutex` -- a mutually exclusive lock -- and the
// most common type of synchronization primitive for facilitating locks on
// shared resources. A mutex is used to prevent multiple threads from accessing
// and/or writing to a shared resource concurrently.
//
// Unlike a `std::mutex`, the Abseil `Mutex` provides the following additional
// features:
// * Conditional predicates intrinsic to the `Mutex` object
// * Shared/reader locks, in addition to standard exclusive/writer locks
// * Deadlock detection and debug support.
//
// The following helper classes are also defined within this file:
//
// MutexLock - An RAII wrapper to acquire and release a `Mutex` for exclusive/
// write access within the current scope.
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
//
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// ReaderMutexLock
// - An RAII wrapper to acquire and release a `Mutex` for shared/read
// access within the current scope.
//
// WriterMutexLock
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// - Effectively an alias for `MutexLock` above, designed for use in
// distinguishing reader and writer locks within code.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
// In addition to simple mutex locks, this file also defines ways to perform
// locking under certain conditions.
//
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// Condition - (Preferred) Used to wait for a particular predicate that
// depends on state protected by the `Mutex` to become true.
// CondVar - A lower-level variant of `Condition` that relies on
// application code to explicitly signal the `CondVar` when
// a condition has been met.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
// See below for more information on using `Condition` or `CondVar`.
//
// Mutexes and mutex behavior can be quite complicated. The information within
// this header file is limited, as a result. Please consult the Mutex guide for
// more complete information and examples.
#ifndef ABSL_SYNCHRONIZATION_MUTEX_H_
#define ABSL_SYNCHRONIZATION_MUTEX_H_
#include <atomic>
#include <cstdint>
#include <cstring>
#include <iterator>
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
#include <string>
#include "absl/base/const_init.h"
#include "absl/base/internal/identity.h"
#include "absl/base/internal/low_level_alloc.h"
#include "absl/base/internal/thread_identity.h"
#include "absl/base/internal/tsan_mutex_interface.h"
#include "absl/base/port.h"
#include "absl/base/thread_annotations.h"
#include "absl/synchronization/internal/kernel_timeout.h"
#include "absl/synchronization/internal/per_thread_sem.h"
#include "absl/time/time.h"
namespace absl {
ABSL_NAMESPACE_BEGIN
class Condition;
struct SynchWaitParams;
// -----------------------------------------------------------------------------
// Mutex
// -----------------------------------------------------------------------------
//
// A `Mutex` is a non-reentrant (aka non-recursive) Mutually Exclusive lock
// on some resource, typically a variable or data structure with associated
// invariants. Proper usage of mutexes prevents concurrent access by different
// threads to the same resource.
//
// A `Mutex` has two basic operations: `Mutex::Lock()` and `Mutex::Unlock()`.
// The `Lock()` operation *acquires* a `Mutex` (in a state known as an
// *exclusive* -- or write -- lock), while the `Unlock()` operation *releases* a
// Mutex. During the span of time between the Lock() and Unlock() operations,
// a mutex is said to be *held*. By design all mutexes support exclusive/write
// locks, as this is the most common way to use a mutex.
//
// The `Mutex` state machine for basic lock/unlock operations is quite simple:
//
// | | Lock() | Unlock() |
// |----------------+------------+----------|
// | Free | Exclusive | invalid |
// | Exclusive | blocks | Free |
//
// Attempts to `Unlock()` must originate from the thread that performed the
// corresponding `Lock()` operation.
//
// An "invalid" operation is disallowed by the API. The `Mutex` implementation
// is allowed to do anything on an invalid call, including but not limited to
// crashing with a useful error message, silently succeeding, or corrupting
// data structures. In debug mode, the implementation attempts to crash with a
// useful error message.
//
// `Mutex` is not guaranteed to be "fair" in prioritizing waiting threads; it
// is, however, approximately fair over long periods, and starvation-free for
// threads at the same priority.
//
// The lock/unlock primitives are now annotated with lock annotations
// defined in (base/thread_annotations.h). When writing multi-threaded code,
// you should use lock annotations whenever possible to document your lock
// synchronization policy. Besides acting as documentation, these annotations
// also help compilers or static analysis tools to identify and warn about
// issues that could potentially result in race conditions and deadlocks.
//
// For more information about the lock annotations, please see
// [Thread Safety Analysis](http://clang.llvm.org/docs/ThreadSafetyAnalysis.html)
// in the Clang documentation.
//
// See also `MutexLock`, below, for scoped `Mutex` acquisition.
class ABSL_LOCKABLE Mutex {
public:
// Creates a `Mutex` that is not held by anyone. This constructor is
// typically used for Mutexes allocated on the heap or the stack.
//
// To create `Mutex` instances with static storage duration
// (e.g. a namespace-scoped or global variable), see
// `Mutex::Mutex(absl::kConstInit)` below instead.
Mutex();
// Creates a mutex with static storage duration. A global variable
// constructed this way avoids the lifetime issues that can occur on program
// startup and shutdown. (See absl/base/const_init.h.)
//
// For Mutexes allocated on the heap and stack, instead use the default
// constructor, which can interact more fully with the thread sanitizer.
//
// Example usage:
// namespace foo {
// ABSL_CONST_INIT absl::Mutex mu(absl::kConstInit);
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// }
explicit constexpr Mutex(absl::ConstInitType);
~Mutex();
// Mutex::Lock()
//
// Blocks the calling thread, if necessary, until this `Mutex` is free, and
// then acquires it exclusively. (This lock is also known as a "write lock.")
void Lock() ABSL_EXCLUSIVE_LOCK_FUNCTION();
// Mutex::Unlock()
//
// Releases this `Mutex` and returns it from the exclusive/write state to the
Export of internal Abseil changes -- 642ab296a2c9629c44f3f2ce6911cd2488bcf416 by Derek Mauro <dmauro@google.com>: Remove an obsolete check in CMakeLists.txt PiperOrigin-RevId: 352852564 -- ce78cb96bcfd162737dbcf35005da3d1d6a3486b by Abseil Team <absl-team@google.com>: Clarify that the calling *thread* must have locked the mutex in order to unlock it. PiperOrigin-RevId: 352801804 -- 24e1f5f72756046f5265abf618e951c341f09b8d by Derek Mauro <dmauro@google.com>: Fixes failing CMake string comparisons https://cmake.org/cmake/help/latest/policy/CMP0054.html Fixes #791 PiperOrigin-RevId: 352791054 -- 0ac10bc3f4dca2c4c4b51d7b8196a2eaee9537a1 by Abseil Team <absl-team@google.com>: Introduce CordRepRing class This change introduces the CordRepRing class that implements all the lower level / internal implementation for upcoming CordRepRing ring buffer support in cord. PiperOrigin-RevId: 352771994 -- 4bd36dda61760785844f0f29f26d90cc18046f75 by Abseil Team <absl-team@google.com>: Optimize InlineData representation for cord sampling (cordz) This CL changes InlineData to allow us to store a (future) Cordz Info pointer directly into the inline representation: - make InlineData a class that provides a public API to set the active union members (tree or chars) and safely access that data. - change 'tree' and 'profiled' bits to be the 2 least significant bits, allowing us 62 continquous bits for storing a Cordz Info pointer. PiperOrigin-RevId: 352642411 -- dc55ba71bbce0e6a83e05a453990c51ac3d68426 by Mark Barolak <mbar@google.com>: Add unit test coverage for the mutating overload of absl::AsciiStrToLower. PiperOrigin-RevId: 352626006 GitOrigin-RevId: 642ab296a2c9629c44f3f2ce6911cd2488bcf416 Change-Id: I6c5929dd830d3c630e14e7fd5387fc3e25a69100
4 years ago
// free state. Calling thread must hold the `Mutex` exclusively.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
void Unlock() ABSL_UNLOCK_FUNCTION();
// Mutex::TryLock()
//
// If the mutex can be acquired without blocking, does so exclusively and
// returns `true`. Otherwise, returns `false`. Returns `true` with high
// probability if the `Mutex` was free.
bool TryLock() ABSL_EXCLUSIVE_TRYLOCK_FUNCTION(true);
// Mutex::AssertHeld()
//
// Require that the mutex be held exclusively (write mode) by this thread.
//
// If the mutex is not currently held by this thread, this function may report
// an error (typically by crashing with a diagnostic) or it may do nothing.
// This function is intended only as a tool to assist debugging; it doesn't
// guarantee correctness.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
void AssertHeld() const ABSL_ASSERT_EXCLUSIVE_LOCK();
// ---------------------------------------------------------------------------
// Reader-Writer Locking
// ---------------------------------------------------------------------------
// A Mutex can also be used as a starvation-free reader-writer lock.
// Neither read-locks nor write-locks are reentrant/recursive to avoid
// potential client programming errors.
//
// The Mutex API provides `Writer*()` aliases for the existing `Lock()`,
// `Unlock()` and `TryLock()` methods for use within applications mixing
// reader/writer locks. Using `Reader*()` and `Writer*()` operations in this
// manner can make locking behavior clearer when mixing read and write modes.
//
// Introducing reader locks necessarily complicates the `Mutex` state
// machine somewhat. The table below illustrates the allowed state transitions
// of a mutex in such cases. Note that ReaderLock() may block even if the lock
// is held in shared mode; this occurs when another thread is blocked on a
// call to WriterLock().
//
// ---------------------------------------------------------------------------
// Operation: WriterLock() Unlock() ReaderLock() ReaderUnlock()
// ---------------------------------------------------------------------------
// State
// ---------------------------------------------------------------------------
// Free Exclusive invalid Shared(1) invalid
// Shared(1) blocks invalid Shared(2) or blocks Free
// Shared(n) n>1 blocks invalid Shared(n+1) or blocks Shared(n-1)
// Exclusive blocks Free blocks invalid
// ---------------------------------------------------------------------------
//
// In comments below, "shared" refers to a state of Shared(n) for any n > 0.
// Mutex::ReaderLock()
//
// Blocks the calling thread, if necessary, until this `Mutex` is either free,
// or in shared mode, and then acquires a share of it. Note that
// `ReaderLock()` will block if some other thread has an exclusive/writer lock
// on the mutex.
void ReaderLock() ABSL_SHARED_LOCK_FUNCTION();
// Mutex::ReaderUnlock()
//
// Releases a read share of this `Mutex`. `ReaderUnlock` may return a mutex to
// the free state if this thread holds the last reader lock on the mutex. Note
// that you cannot call `ReaderUnlock()` on a mutex held in write mode.
void ReaderUnlock() ABSL_UNLOCK_FUNCTION();
// Mutex::ReaderTryLock()
//
// If the mutex can be acquired without blocking, acquires this mutex for
// shared access and returns `true`. Otherwise, returns `false`. Returns
// `true` with high probability if the `Mutex` was free or shared.
bool ReaderTryLock() ABSL_SHARED_TRYLOCK_FUNCTION(true);
// Mutex::AssertReaderHeld()
//
// Require that the mutex be held at least in shared mode (read mode) by this
// thread.
//
// If the mutex is not currently held by this thread, this function may report
// an error (typically by crashing with a diagnostic) or it may do nothing.
// This function is intended only as a tool to assist debugging; it doesn't
// guarantee correctness.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
void AssertReaderHeld() const ABSL_ASSERT_SHARED_LOCK();
// Mutex::WriterLock()
// Mutex::WriterUnlock()
// Mutex::WriterTryLock()
//
// Aliases for `Mutex::Lock()`, `Mutex::Unlock()`, and `Mutex::TryLock()`.
//
// These methods may be used (along with the complementary `Reader*()`
// methods) to distingish simple exclusive `Mutex` usage (`Lock()`,
// etc.) from reader/writer lock usage.
void WriterLock() ABSL_EXCLUSIVE_LOCK_FUNCTION() { this->Lock(); }
void WriterUnlock() ABSL_UNLOCK_FUNCTION() { this->Unlock(); }
bool WriterTryLock() ABSL_EXCLUSIVE_TRYLOCK_FUNCTION(true) {
return this->TryLock();
}
// ---------------------------------------------------------------------------
// Conditional Critical Regions
// ---------------------------------------------------------------------------
// Conditional usage of a `Mutex` can occur using two distinct paradigms:
//
// * Use of `Mutex` member functions with `Condition` objects.
// * Use of the separate `CondVar` abstraction.
//
// In general, prefer use of `Condition` and the `Mutex` member functions
// listed below over `CondVar`. When there are multiple threads waiting on
// distinctly different conditions, however, a battery of `CondVar`s may be
// more efficient. This section discusses use of `Condition` objects.
//
// `Mutex` contains member functions for performing lock operations only under
// certain conditions, of class `Condition`. For correctness, the `Condition`
// must return a boolean that is a pure function, only of state protected by
// the `Mutex`. The condition must be invariant w.r.t. environmental state
// such as thread, cpu id, or time, and must be `noexcept`. The condition will
// always be invoked with the mutex held in at least read mode, so you should
// not block it for long periods or sleep it on a timer.
//
// Since a condition must not depend directly on the current time, use
// `*WithTimeout()` member function variants to make your condition
// effectively true after a given duration, or `*WithDeadline()` variants to
// make your condition effectively true after a given time.
//
// The condition function should have no side-effects aside from debug
// logging; as a special exception, the function may acquire other mutexes
// provided it releases all those that it acquires. (This exception was
// required to allow logging.)
// Mutex::Await()
//
// Unlocks this `Mutex` and blocks until simultaneously both `cond` is `true`
// and this `Mutex` can be reacquired, then reacquires this `Mutex` in the
// same mode in which it was previously held. If the condition is initially
// `true`, `Await()` *may* skip the release/re-acquire step.
//
// `Await()` requires that this thread holds this `Mutex` in some mode.
void Await(const Condition &cond);
// Mutex::LockWhen()
// Mutex::ReaderLockWhen()
// Mutex::WriterLockWhen()
//
// Blocks until simultaneously both `cond` is `true` and this `Mutex` can
// be acquired, then atomically acquires this `Mutex`. `LockWhen()` is
// logically equivalent to `*Lock(); Await();` though they may have different
// performance characteristics.
void LockWhen(const Condition &cond) ABSL_EXCLUSIVE_LOCK_FUNCTION();
void ReaderLockWhen(const Condition &cond) ABSL_SHARED_LOCK_FUNCTION();
void WriterLockWhen(const Condition &cond) ABSL_EXCLUSIVE_LOCK_FUNCTION() {
this->LockWhen(cond);
}
// ---------------------------------------------------------------------------
// Mutex Variants with Timeouts/Deadlines
// ---------------------------------------------------------------------------
// Mutex::AwaitWithTimeout()
// Mutex::AwaitWithDeadline()
//
// Unlocks this `Mutex` and blocks until simultaneously:
// - either `cond` is true or the {timeout has expired, deadline has passed}
// and
// - this `Mutex` can be reacquired,
// then reacquire this `Mutex` in the same mode in which it was previously
// held, returning `true` iff `cond` is `true` on return.
//
// If the condition is initially `true`, the implementation *may* skip the
// release/re-acquire step and return immediately.
//
// Deadlines in the past are equivalent to an immediate deadline.
// Negative timeouts are equivalent to a zero timeout.
//
// This method requires that this thread holds this `Mutex` in some mode.
bool AwaitWithTimeout(const Condition &cond, absl::Duration timeout);
bool AwaitWithDeadline(const Condition &cond, absl::Time deadline);
// Mutex::LockWhenWithTimeout()
// Mutex::ReaderLockWhenWithTimeout()
// Mutex::WriterLockWhenWithTimeout()
//
// Blocks until simultaneously both:
// - either `cond` is `true` or the timeout has expired, and
// - this `Mutex` can be acquired,
// then atomically acquires this `Mutex`, returning `true` iff `cond` is
// `true` on return.
//
// Negative timeouts are equivalent to a zero timeout.
bool LockWhenWithTimeout(const Condition &cond, absl::Duration timeout)
ABSL_EXCLUSIVE_LOCK_FUNCTION();
bool ReaderLockWhenWithTimeout(const Condition &cond, absl::Duration timeout)
ABSL_SHARED_LOCK_FUNCTION();
bool WriterLockWhenWithTimeout(const Condition &cond, absl::Duration timeout)
ABSL_EXCLUSIVE_LOCK_FUNCTION() {
return this->LockWhenWithTimeout(cond, timeout);
}
// Mutex::LockWhenWithDeadline()
// Mutex::ReaderLockWhenWithDeadline()
// Mutex::WriterLockWhenWithDeadline()
//
// Blocks until simultaneously both:
// - either `cond` is `true` or the deadline has been passed, and
// - this `Mutex` can be acquired,
// then atomically acquires this Mutex, returning `true` iff `cond` is `true`
// on return.
//
// Deadlines in the past are equivalent to an immediate deadline.
bool LockWhenWithDeadline(const Condition &cond, absl::Time deadline)
ABSL_EXCLUSIVE_LOCK_FUNCTION();
bool ReaderLockWhenWithDeadline(const Condition &cond, absl::Time deadline)
ABSL_SHARED_LOCK_FUNCTION();
bool WriterLockWhenWithDeadline(const Condition &cond, absl::Time deadline)
ABSL_EXCLUSIVE_LOCK_FUNCTION() {
return this->LockWhenWithDeadline(cond, deadline);
}
// ---------------------------------------------------------------------------
// Debug Support: Invariant Checking, Deadlock Detection, Logging.
// ---------------------------------------------------------------------------
// Mutex::EnableInvariantDebugging()
//
// If `invariant`!=null and if invariant debugging has been enabled globally,
// cause `(*invariant)(arg)` to be called at moments when the invariant for
// this `Mutex` should hold (for example: just after acquire, just before
// release).
//
// The routine `invariant` should have no side-effects since it is not
// guaranteed how many times it will be called; it should check the invariant
// and crash if it does not hold. Enabling global invariant debugging may
// substantially reduce `Mutex` performance; it should be set only for
// non-production runs. Optimization options may also disable invariant
// checks.
void EnableInvariantDebugging(void (*invariant)(void *), void *arg);
// Mutex::EnableDebugLog()
//
// Cause all subsequent uses of this `Mutex` to be logged via
// `ABSL_RAW_LOG(INFO)`. Log entries are tagged with `name` if no previous
// call to `EnableInvariantDebugging()` or `EnableDebugLog()` has been made.
//
// Note: This method substantially reduces `Mutex` performance.
void EnableDebugLog(const char *name);
// Deadlock detection
// Mutex::ForgetDeadlockInfo()
//
// Forget any deadlock-detection information previously gathered
// about this `Mutex`. Call this method in debug mode when the lock ordering
// of a `Mutex` changes.
void ForgetDeadlockInfo();
// Mutex::AssertNotHeld()
//
// Return immediately if this thread does not hold this `Mutex` in any
// mode; otherwise, may report an error (typically by crashing with a
// diagnostic), or may return immediately.
//
// Currently this check is performed only if all of:
// - in debug mode
// - SetMutexDeadlockDetectionMode() has been set to kReport or kAbort
// - number of locks concurrently held by this thread is not large.
// are true.
void AssertNotHeld() const;
// Special cases.
// A `MuHow` is a constant that indicates how a lock should be acquired.
// Internal implementation detail. Clients should ignore.
typedef const struct MuHowS *MuHow;
// Mutex::InternalAttemptToUseMutexInFatalSignalHandler()
//
// Causes the `Mutex` implementation to prepare itself for re-entry caused by
// future use of `Mutex` within a fatal signal handler. This method is
// intended for use only for last-ditch attempts to log crash information.
// It does not guarantee that attempts to use Mutexes within the handler will
// not deadlock; it merely makes other faults less likely.
//
// WARNING: This routine must be invoked from a signal handler, and the
// signal handler must either loop forever or terminate the process.
// Attempts to return from (or `longjmp` out of) the signal handler once this
// call has been made may cause arbitrary program behaviour including
// crashes and deadlocks.
static void InternalAttemptToUseMutexInFatalSignalHandler();
private:
std::atomic<intptr_t> mu_; // The Mutex state.
// Post()/Wait() versus associated PerThreadSem; in class for required
// friendship with PerThreadSem.
Export of internal Abseil changes -- 5ed5dc9e17c66c298ee31cefc941a46348d8ad34 by Abseil Team <absl-team@google.com>: Fix typo. PiperOrigin-RevId: 362040582 -- ac704b53a49becc42f77e4529d3952f8e7d18ce4 by Abseil Team <absl-team@google.com>: Fix a typo in a comment. PiperOrigin-RevId: 361576641 -- d20ccb27b7e9b53481e9192c1aae5202c06bfcb1 by Derek Mauro <dmauro@google.com>: Remove the inline keyword from functions that aren't defined in the header. This may fix #910. PiperOrigin-RevId: 361551300 -- aed9ae1dffa7b228dcb6ffbeb2fe06a13970c72b by Laramie Leavitt <lar@google.com>: Propagate nice/strict/naggy state on absl::MockingBitGen. Allowing NiceMocks reduces the log spam for un-mocked calls, and it enables nicer setup with ON_CALL, so it is desirable to support it in absl::MockingBitGen. Internally, gmock tracks object "strictness" levels using an internal API; in order to achieve the same results we detect when the MockingBitGen is wrapped in a Nice/Naggy/Strict and wrap the internal implementation MockFunction in the same type. This is achieved by providing overloads to the Call() function, and passing the mock object type down into it's own RegisterMock call, where a compile-time check verifies the state and creates the appropriate mock function. PiperOrigin-RevId: 361233484 -- 96186023fabd13d01d32d60d9c7ac4ead1aeb989 by Abseil Team <absl-team@google.com>: Ensure that trivial types are passed by value rather than reference PiperOrigin-RevId: 361217450 -- e1135944835d27f77e8119b8166d8fb6aa25f906 by Evan Brown <ezb@google.com>: Internal change. PiperOrigin-RevId: 361215882 -- 583fe6c94c1c2ef757ef6e78292a15fbe4030e35 by Evan Brown <ezb@google.com>: Increase the minimum number of slots per node from 3 to 4. We also rename kNodeValues (and related names) to kNodeSlots to make it clear that they are about the number of slots per node rather than the number of values per node - kMinNodeValues keeps the same name because it's actually about the number of values rather than the number of slots. Motivation: I think the expected number of values per node, assuming random insertion order, is the average of the maximum and minimum numbers of values per node (kNodeSlots and kMinNodeValues). For large and/or even kNodeSlots, this is ~75% of kNodeSlots, but for kNodeSlots=3, this is ~67% of kNodeSlots. kMinNodeValues (which corresponds to worst-case occupancy) is ~33% of kNodeSlots, when kNodeSlots=3, compared to 50% for even kNodeSlots. This results in higher memory overhead per value, and since this case (kNodeSlots=3) is used when values are large, it seems worth fixing. PiperOrigin-RevId: 361171495 GitOrigin-RevId: 5ed5dc9e17c66c298ee31cefc941a46348d8ad34 Change-Id: I8e33b5df1f987a77112093821085c410185ab51a
4 years ago
static void IncrementSynchSem(Mutex *mu, base_internal::PerThreadSynch *w);
static bool DecrementSynchSem(Mutex *mu, base_internal::PerThreadSynch *w,
synchronization_internal::KernelTimeout t);
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// slow path acquire
void LockSlowLoop(SynchWaitParams *waitp, int flags);
// wrappers around LockSlowLoop()
bool LockSlowWithDeadline(MuHow how, const Condition *cond,
synchronization_internal::KernelTimeout t,
int flags);
void LockSlow(MuHow how, const Condition *cond,
int flags) ABSL_ATTRIBUTE_COLD;
// slow path release
void UnlockSlow(SynchWaitParams *waitp) ABSL_ATTRIBUTE_COLD;
// Common code between Await() and AwaitWithTimeout/Deadline()
bool AwaitCommon(const Condition &cond,
synchronization_internal::KernelTimeout t);
// Attempt to remove thread s from queue.
void TryRemove(base_internal::PerThreadSynch *s);
// Block a thread on mutex.
void Block(base_internal::PerThreadSynch *s);
// Wake a thread; return successor.
base_internal::PerThreadSynch *Wakeup(base_internal::PerThreadSynch *w);
friend class CondVar; // for access to Trans()/Fer().
void Trans(MuHow how); // used for CondVar->Mutex transfer
void Fer(
base_internal::PerThreadSynch *w); // used for CondVar->Mutex transfer
// Catch the error of writing Mutex when intending MutexLock.
Mutex(const volatile Mutex * /*ignored*/) {} // NOLINT(runtime/explicit)
Mutex(const Mutex&) = delete;
Mutex& operator=(const Mutex&) = delete;
};
// -----------------------------------------------------------------------------
// Mutex RAII Wrappers
// -----------------------------------------------------------------------------
// MutexLock
//
// `MutexLock` is a helper class, which acquires and releases a `Mutex` via
// RAII.
//
// Example:
//
// Class Foo {
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// public:
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// Foo::Bar* Baz() {
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// MutexLock lock(&mu_);
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// ...
// return bar;
// }
//
// private:
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// Mutex mu_;
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// };
class ABSL_SCOPED_LOCKABLE MutexLock {
public:
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// Constructors
// Calls `mu->Lock()` and returns when that call returns. That is, `*mu` is
// guaranteed to be locked when this object is constructed. Requires that
// `mu` be dereferenceable.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
explicit MutexLock(Mutex *mu) ABSL_EXCLUSIVE_LOCK_FUNCTION(mu) : mu_(mu) {
this->mu_->Lock();
}
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// Like above, but calls `mu->LockWhen(cond)` instead. That is, in addition to
// the above, the condition given by `cond` is also guaranteed to hold when
// this object is constructed.
explicit MutexLock(Mutex *mu, const Condition &cond)
ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
: mu_(mu) {
this->mu_->LockWhen(cond);
}
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
MutexLock(const MutexLock &) = delete; // NOLINT(runtime/mutex)
MutexLock(MutexLock&&) = delete; // NOLINT(runtime/mutex)
MutexLock& operator=(const MutexLock&) = delete;
MutexLock& operator=(MutexLock&&) = delete;
~MutexLock() ABSL_UNLOCK_FUNCTION() { this->mu_->Unlock(); }
private:
Mutex *const mu_;
};
// ReaderMutexLock
//
// The `ReaderMutexLock` is a helper class, like `MutexLock`, which acquires and
// releases a shared lock on a `Mutex` via RAII.
class ABSL_SCOPED_LOCKABLE ReaderMutexLock {
public:
explicit ReaderMutexLock(Mutex *mu) ABSL_SHARED_LOCK_FUNCTION(mu) : mu_(mu) {
mu->ReaderLock();
}
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
explicit ReaderMutexLock(Mutex *mu, const Condition &cond)
ABSL_SHARED_LOCK_FUNCTION(mu)
: mu_(mu) {
mu->ReaderLockWhen(cond);
}
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
ReaderMutexLock(const ReaderMutexLock&) = delete;
ReaderMutexLock(ReaderMutexLock&&) = delete;
ReaderMutexLock& operator=(const ReaderMutexLock&) = delete;
ReaderMutexLock& operator=(ReaderMutexLock&&) = delete;
~ReaderMutexLock() ABSL_UNLOCK_FUNCTION() { this->mu_->ReaderUnlock(); }
private:
Mutex *const mu_;
};
// WriterMutexLock
//
// The `WriterMutexLock` is a helper class, like `MutexLock`, which acquires and
// releases a write (exclusive) lock on a `Mutex` via RAII.
class ABSL_SCOPED_LOCKABLE WriterMutexLock {
public:
explicit WriterMutexLock(Mutex *mu) ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
: mu_(mu) {
mu->WriterLock();
}
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
explicit WriterMutexLock(Mutex *mu, const Condition &cond)
ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
: mu_(mu) {
mu->WriterLockWhen(cond);
}
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
WriterMutexLock(const WriterMutexLock&) = delete;
WriterMutexLock(WriterMutexLock&&) = delete;
WriterMutexLock& operator=(const WriterMutexLock&) = delete;
WriterMutexLock& operator=(WriterMutexLock&&) = delete;
~WriterMutexLock() ABSL_UNLOCK_FUNCTION() { this->mu_->WriterUnlock(); }
private:
Mutex *const mu_;
};
// -----------------------------------------------------------------------------
// Condition
// -----------------------------------------------------------------------------
//
// `Mutex` contains a number of member functions which take a `Condition` as an
// argument; clients can wait for conditions to become `true` before attempting
// to acquire the mutex. These sections are known as "condition critical"
// sections. To use a `Condition`, you simply need to construct it, and use
// within an appropriate `Mutex` member function; everything else in the
// `Condition` class is an implementation detail.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
// A `Condition` is specified as a function pointer which returns a boolean.
// `Condition` functions should be pure functions -- their results should depend
// only on passed arguments, should not consult any external state (such as
// clocks), and should have no side-effects, aside from debug logging. Any
// objects that the function may access should be limited to those which are
// constant while the mutex is blocked on the condition (e.g. a stack variable),
// or objects of state protected explicitly by the mutex.
//
// No matter which construction is used for `Condition`, the underlying
// function pointer / functor / callable must not throw any
// exceptions. Correctness of `Mutex` / `Condition` is not guaranteed in
// the face of a throwing `Condition`. (When Abseil is allowed to depend
// on C++17, these function pointers will be explicitly marked
// `noexcept`; until then this requirement cannot be enforced in the
// type system.)
//
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// Note: to use a `Condition`, you need only construct it and pass it to a
// suitable `Mutex' member function, such as `Mutex::Await()`, or to the
// constructor of one of the scope guard classes.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// Example using LockWhen/Unlock:
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
// // assume count_ is not internal reference count
// int count_ ABSL_GUARDED_BY(mu_);
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// Condition count_is_zero(+[](int *count) { return *count == 0; }, &count_);
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
// mu_.LockWhen(count_is_zero);
// // ...
// mu_.Unlock();
//
// Example using a scope guard:
//
// {
// MutexLock lock(&mu_, count_is_zero);
// // ...
// }
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
// When multiple threads are waiting on exactly the same condition, make sure
// that they are constructed with the same parameters (same pointer to function
// + arg, or same pointer to object + method), so that the mutex implementation
// can avoid redundantly evaluating the same condition for each thread.
class Condition {
public:
// A Condition that returns the result of "(*func)(arg)"
Condition(bool (*func)(void *), void *arg);
// Templated version for people who are averse to casts.
//
// To use a lambda, prepend it with unary plus, which converts the lambda
// into a function pointer:
// Condition(+[](T* t) { return ...; }, arg).
//
// Note: lambdas in this case must contain no bound variables.
//
// See class comment for performance advice.
template<typename T>
Condition(bool (*func)(T *), T *arg);
// Templated version for invoking a method that returns a `bool`.
//
// `Condition(object, &Class::Method)` constructs a `Condition` that evaluates
// `object->Method()`.
//
// Implementation Note: `absl::internal::identity` is used to allow methods to
// come from base classes. A simpler signature like
// `Condition(T*, bool (T::*)())` does not suffice.
template<typename T>
Condition(T *object, bool (absl::internal::identity<T>::type::* method)());
// Same as above, for const members
template<typename T>
Condition(const T *object,
bool (absl::internal::identity<T>::type::* method)() const);
// A Condition that returns the value of `*cond`
explicit Condition(const bool *cond);
// Templated version for invoking a functor that returns a `bool`.
// This approach accepts pointers to non-mutable lambdas, `std::function`,
// the result of` std::bind` and user-defined functors that define
// `bool F::operator()() const`.
//
// Example:
//
// auto reached = [this, current]() {
// mu_.AssertReaderHeld(); // For annotalysis.
// return processed_ >= current;
// };
// mu_.Await(Condition(&reached));
Export of internal Abseil changes -- 4833151c207fac9f57a735efe6d5db4c83368415 by Gennadiy Rozental <rogeeff@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 320398694 -- a1becb36b223230f0a45f204a5fb33b83d2deffe by Gennadiy Rozental <rogeeff@google.com>: Update CMakeLists.txt Import of https://github.com/abseil/abseil-cpp/pull/737 PiperOrigin-RevId: 320391906 -- b529c45856fe7a3447f1f3259286d57e13b1f292 by Abseil Team <absl-team@google.com>: Improves a comment about use of absl::Condition. PiperOrigin-RevId: 320384329 -- c7b1dacda2739c10dc1ccbfb56b07ed7fe2464a4 by Laramie Leavitt <lar@google.com>: Improve FastUniformBits performance for std::minstd_rand. The rejection algorithm was too pessimistic before, and not in line with the [rand.adapt.ibits]. Specifically, when sampling from an URBG with a non power of 2 range, FastUniformBits constructed a rejection threshold with a power-of-2 range that was too restrictive. For example, minstd_rand has a range of [1, 2147483646], which has a range of 2145386495, or about 30.999 bits. Before FastUniformBits rejected values between 1<<30 and 2145386495, which includes approximately 50% of the generated values. However, since a minimum of 3 calls are required to generate a full 64-bit value from an entropy pool of 30.9 bits, the correct value for rejection sampling is the range value which masks 21 (0x7fe00000) or 22 bits and rejects values greater than that. This reduces the probability of rejecting a sample to about 0.1% NOTE: Abseil random does not guarantee sequence stability over time, and this is expected to change sequences in some cases. PiperOrigin-RevId: 320285836 -- 15800a39557a07dd52e0add66a0ab67aed00590b by Gennadiy Rozental <rogeeff@google.com>: Internal change. PiperOrigin-RevId: 320220913 -- ef39348360873f6d19669755fe0b5d09a945a501 by Gennadiy Rozental <rogeeff@google.com>: Internal change PiperOrigin-RevId: 320181729 -- 4f9f6ef8034a24da1832e4c838c72f80fc2ea062 by Gennadiy Rozental <rogeeff@google.com>: Internal change PiperOrigin-RevId: 320176084 -- 6bfc8008462801657d231585bd5c37fc18bb25b6 by Gennadiy Rozental <rogeeff@google.com>: Internal change PiperOrigin-RevId: 320176070 -- b35b055ab1f41e6056031ff0641cabab23530027 by Abseil Team <absl-team@google.com>: Disabling using header module as well as building one for randen_hwaes_impl PiperOrigin-RevId: 320024299 GitOrigin-RevId: 4833151c207fac9f57a735efe6d5db4c83368415 Change-Id: I9cf102dbf46ed07752a508b7cda3ab3858857d0d
4 years ago
//
Export of internal Abseil changes -- a5af5874c1c5cc02bd2a748d455321f82b6f2a93 by Andy Getzendanner <durandal@google.com>: fix compile fails with asan and -Wredundant-decls Import of https://github.com/abseil/abseil-cpp/pull/801 PiperOrigin-RevId: 336693223 -- ed9df42ab2b742386c6692c2bed015374c919d9c by Derek Mauro <dmauro@google.com>: Fix integer conversion warning Fixes #814 PiperOrigin-RevId: 336651814 -- 0ab4c23884e72dce17b67c1eb520f9dbb802565d by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 336585378 -- eba0e3dccd52a6e91bcff84075bef0affc650b74 by Matt Kulukundis <kfm@google.com>: Add bitset operations to Futex helper. PiperOrigin-RevId: 336409368 -- 8b0709a8b4500bf5f0af4b602d76a298d81645e8 by Abseil Team <absl-team@google.com>: Fix code indentation in a comment. PiperOrigin-RevId: 336368167 -- bc3961c87a7e7760c10319a5b0349c279f7ae3ad by Samuel Benzaquen <sbenza@google.com>: Improve performance of the registry: - Reduce contention - Reduce memory usage for each flag by `6*sizeof(void*)`. - Replace one immortal allocation per-flag with a single one for all the flags - Slightly improve single-threaded performance by avoiding the std::map indirections. PiperOrigin-RevId: 336365904 -- 264ad9f28f935aad8b6b1437f8bf804fa9104346 by Abseil Team <absl-team@google.com>: Fix typo in comment on absl::Condition. PiperOrigin-RevId: 336311680 -- b5b808a8c75ca0df7b09eff9a423ec171d80f771 by Derek Mauro <dmauro@google.com>: Add missing Apache license headers PiperOrigin-RevId: 336294980 -- 89446c3a4793df8b95060385cf3e219357c3db1d by Andy Soffer <asoffer@google.com>: Internal changes PiperOrigin-RevId: 336287465 -- 57c8be4e294881bc79a6a44b8e4bf7ecbb19b9b9 by Matt Kulukundis <kfm@google.com>: Extract Futex from an implementation detail of Wait to a private interface. PiperOrigin-RevId: 336123209 GitOrigin-RevId: a5af5874c1c5cc02bd2a748d455321f82b6f2a93 Change-Id: Ie5a0ebe28e571814e3e11d4c05ca308523ccf311
4 years ago
// NOTE: never use "mu_.AssertHeld()" instead of "mu_.AssertReaderHeld()" in
// the lambda as it may be called when the mutex is being unlocked from a
// scope holding only a reader lock, which will make the assertion not
// fulfilled and crash the binary.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// See class comment for performance advice. In particular, if there
// might be more than one waiter for the same condition, make sure
// that all waiters construct the condition with the same pointers.
// Implementation note: The second template parameter ensures that this
// constructor doesn't participate in overload resolution if T doesn't have
// `bool operator() const`.
template <typename T, typename E = decltype(
static_cast<bool (T::*)() const>(&T::operator()))>
explicit Condition(const T *obj)
: Condition(obj, static_cast<bool (T::*)() const>(&T::operator())) {}
// A Condition that always returns `true`.
static const Condition kTrue;
// Evaluates the condition.
bool Eval() const;
// Returns `true` if the two conditions are guaranteed to return the same
// value if evaluated at the same time, `false` if the evaluation *may* return
// different results.
//
// Two `Condition` values are guaranteed equal if both their `func` and `arg`
// components are the same. A null pointer is equivalent to a `true`
// condition.
static bool GuaranteedEqual(const Condition *a, const Condition *b);
private:
// Sizing an allocation for a method pointer can be subtle. In the Itanium
// specifications, a method pointer has a predictable, uniform size. On the
// other hand, MSVC ABI, method pointer sizes vary based on the
// inheritance of the class. Specifically, method pointers from classes with
// multiple inheritance are bigger than those of classes with single
// inheritance. Other variations also exist.
// A good way to allocate enough space for *any* pointer in these ABIs is to
// employ a class declaration with no definition. Because the inheritance
// structure is not available for this declaration, the compiler must
// assume, conservatively, that its method pointers have the largest possible
// size.
class OpaqueClass;
using ConservativeMethodPointer = bool (OpaqueClass::*)();
static_assert(sizeof(bool(OpaqueClass::*)()) >= sizeof(bool (*)(void *)),
"Unsupported platform.");
// Allocation for a function pointer or method pointer.
// The {0} initializer ensures that all unused bytes of this buffer are
// always zeroed out. This is necessary, because GuaranteedEqual() compares
// all of the bytes, unaware of which bytes are relevant to a given `eval_`.
char callback_[sizeof(ConservativeMethodPointer)] = {0};
// Function with which to evaluate callbacks and/or arguments.
bool (*eval_)(const Condition*);
// Either an argument for a function call or an object for a method call.
void *arg_;
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// Various functions eval_ can point to:
static bool CallVoidPtrFunction(const Condition*);
template <typename T> static bool CastAndCallFunction(const Condition* c);
template <typename T> static bool CastAndCallMethod(const Condition* c);
// Helper methods for storing, validating, and reading callback arguments.
template <typename T>
inline void StoreCallback(T callback) {
static_assert(
sizeof(callback) <= sizeof(callback_),
"An overlarge pointer was passed as a callback to Condition.");
std::memcpy(callback_, &callback, sizeof(callback));
}
template <typename T>
inline void ReadCallback(T *callback) const {
std::memcpy(callback, callback_, sizeof(*callback));
}
Condition(); // null constructor used only to create kTrue
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
};
// -----------------------------------------------------------------------------
// CondVar
// -----------------------------------------------------------------------------
//
// A condition variable, reflecting state evaluated separately outside of the
// `Mutex` object, which can be signaled to wake callers.
// This class is not normally needed; use `Mutex` member functions such as
// `Mutex::Await()` and intrinsic `Condition` abstractions. In rare cases
// with many threads and many conditions, `CondVar` may be faster.
//
// The implementation may deliver signals to any condition variable at
// any time, even when no call to `Signal()` or `SignalAll()` is made; as a
// result, upon being awoken, you must check the logical condition you have
// been waiting upon.
//
// Examples:
//
// Usage for a thread waiting for some condition C protected by mutex mu:
// mu.Lock();
// while (!C) { cv->Wait(&mu); } // releases and reacquires mu
// // C holds; process data
// mu.Unlock();
//
// Usage to wake T is:
// mu.Lock();
// // process data, possibly establishing C
// if (C) { cv->Signal(); }
// mu.Unlock();
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
// If C may be useful to more than one waiter, use `SignalAll()` instead of
// `Signal()`.
//
// With this implementation it is efficient to use `Signal()/SignalAll()` inside
// the locked region; this usage can make reasoning about your program easier.
//
class CondVar {
public:
// A `CondVar` allocated on the heap or on the stack can use the this
// constructor.
CondVar();
~CondVar();
// CondVar::Wait()
//
// Atomically releases a `Mutex` and blocks on this condition variable.
// Waits until awakened by a call to `Signal()` or `SignalAll()` (or a
// spurious wakeup), then reacquires the `Mutex` and returns.
//
// Requires and ensures that the current thread holds the `Mutex`.
void Wait(Mutex *mu);
// CondVar::WaitWithTimeout()
//
// Atomically releases a `Mutex` and blocks on this condition variable.
// Waits until awakened by a call to `Signal()` or `SignalAll()` (or a
// spurious wakeup), or until the timeout has expired, then reacquires
// the `Mutex` and returns.
//
// Returns true if the timeout has expired without this `CondVar`
// being signalled in any manner. If both the timeout has expired
// and this `CondVar` has been signalled, the implementation is free
// to return `true` or `false`.
//
// Requires and ensures that the current thread holds the `Mutex`.
bool WaitWithTimeout(Mutex *mu, absl::Duration timeout);
// CondVar::WaitWithDeadline()
//
// Atomically releases a `Mutex` and blocks on this condition variable.
// Waits until awakened by a call to `Signal()` or `SignalAll()` (or a
// spurious wakeup), or until the deadline has passed, then reacquires
// the `Mutex` and returns.
//
// Deadlines in the past are equivalent to an immediate deadline.
//
// Returns true if the deadline has passed without this `CondVar`
// being signalled in any manner. If both the deadline has passed
// and this `CondVar` has been signalled, the implementation is free
// to return `true` or `false`.
//
// Requires and ensures that the current thread holds the `Mutex`.
bool WaitWithDeadline(Mutex *mu, absl::Time deadline);
// CondVar::Signal()
//
// Signal this `CondVar`; wake at least one waiter if one exists.
void Signal();
// CondVar::SignalAll()
//
// Signal this `CondVar`; wake all waiters.
void SignalAll();
// CondVar::EnableDebugLog()
//
// Causes all subsequent uses of this `CondVar` to be logged via
// `ABSL_RAW_LOG(INFO)`. Log entries are tagged with `name` if `name != 0`.
// Note: this method substantially reduces `CondVar` performance.
void EnableDebugLog(const char *name);
private:
bool WaitCommon(Mutex *mutex, synchronization_internal::KernelTimeout t);
void Remove(base_internal::PerThreadSynch *s);
void Wakeup(base_internal::PerThreadSynch *w);
std::atomic<intptr_t> cv_; // Condition variable state.
CondVar(const CondVar&) = delete;
CondVar& operator=(const CondVar&) = delete;
};
// Variants of MutexLock.
//
// If you find yourself using one of these, consider instead using
// Mutex::Unlock() and/or if-statements for clarity.
// MutexLockMaybe
//
// MutexLockMaybe is like MutexLock, but is a no-op when mu is null.
class ABSL_SCOPED_LOCKABLE MutexLockMaybe {
public:
explicit MutexLockMaybe(Mutex *mu) ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
: mu_(mu) {
if (this->mu_ != nullptr) {
this->mu_->Lock();
}
}
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
explicit MutexLockMaybe(Mutex *mu, const Condition &cond)
ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
: mu_(mu) {
if (this->mu_ != nullptr) {
this->mu_->LockWhen(cond);
}
}
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
~MutexLockMaybe() ABSL_UNLOCK_FUNCTION() {
if (this->mu_ != nullptr) { this->mu_->Unlock(); }
}
private:
Mutex *const mu_;
MutexLockMaybe(const MutexLockMaybe&) = delete;
MutexLockMaybe(MutexLockMaybe&&) = delete;
MutexLockMaybe& operator=(const MutexLockMaybe&) = delete;
MutexLockMaybe& operator=(MutexLockMaybe&&) = delete;
};
// ReleasableMutexLock
//
// ReleasableMutexLock is like MutexLock, but permits `Release()` of its
// mutex before destruction. `Release()` may be called at most once.
class ABSL_SCOPED_LOCKABLE ReleasableMutexLock {
public:
explicit ReleasableMutexLock(Mutex *mu) ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
: mu_(mu) {
this->mu_->Lock();
}
Export of internal Abseil changes -- 0e3e8be75b3ab243991c9b28a27623d86e4511e6 by Abseil Team <absl-team@google.com>: Add constructor overloads with signature (Mutex*, const Condition&) to MutexLock, ReaderMutexLock, WriterMutexLock, ReleasableMutexLock, MaybeMutexLock. These overloads call Mutex::LockWhen, Mutex::ReaderLockWhen, Mutex::WriterLockWhen. Using the guard classes with these new constructors replaces both manual LockWhen/Unlock sequences and the less-efficient, but popular current pattern of "absl::MutexLock lock(&mu); mu.Await(cond);". PiperOrigin-RevId: 339480213 -- ff999bc08360f5bd95557147c97b0e7b200fe3a8 by Jorg Brown <jorg@google.com>: ConvertibleToStringView wastes a lot of cycles initializing members just to reset them immediately after. Only initialize the string storage when needed. This makes StrSplit() 0-30% faster depending on the use case. PiperOrigin-RevId: 339479046 -- 0a773bfb8bc141433a41388731357001fdb34881 by Derek Mauro <dmauro@google.com>: Remove the compiler upgrade fiasco inducing -Weverything -Werror. Switch to a curated set of warnings that may be expanded in the future. PiperOrigin-RevId: 339472677 -- eab54e3e11b126283d33f64c914b200038d215a4 by Abseil Team <absl-team@google.com>: Change execute permission to match presence of the shebang remove execute permission for cmake_common.sh add execute permission for conanfile.py PiperOrigin-RevId: 339453550 -- 7f9726fb605ed20f17f3e221dbce0df03d6904c6 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 339385761 -- f3210dbee3e8a719cf31706963721722203f90e0 by Derek Mauro <dmauro@google.com>: Switch clang compiler detection to use to the Bazel supported mechanism When Abseil launched, we relied on the compiler string "llvm", which we manually set when we used the automatic crosstool generation by using the environment variable BAZEL_COMPILER. Today, Bazel detects clang and automatically sets the compiler string to "clang". Fixes #732 PiperOrigin-RevId: 339360688 -- 413211f59e5e671bf5774efa63ab4df185c74248 by Abseil Team <absl-team@google.com>: Minor comment clarifications and cosmetic tweaks. PiperOrigin-RevId: 339344301 GitOrigin-RevId: 0e3e8be75b3ab243991c9b28a27623d86e4511e6 Change-Id: Ia5b7224cd3d274c79ec7f5514fef63014f458f0f
4 years ago
explicit ReleasableMutexLock(Mutex *mu, const Condition &cond)
ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
: mu_(mu) {
this->mu_->LockWhen(cond);
}
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
~ReleasableMutexLock() ABSL_UNLOCK_FUNCTION() {
if (this->mu_ != nullptr) { this->mu_->Unlock(); }
}
void Release() ABSL_UNLOCK_FUNCTION();
private:
Mutex *mu_;
ReleasableMutexLock(const ReleasableMutexLock&) = delete;
ReleasableMutexLock(ReleasableMutexLock&&) = delete;
ReleasableMutexLock& operator=(const ReleasableMutexLock&) = delete;
ReleasableMutexLock& operator=(ReleasableMutexLock&&) = delete;
};
inline Mutex::Mutex() : mu_(0) {
ABSL_TSAN_MUTEX_CREATE(this, __tsan_mutex_not_static);
}
inline constexpr Mutex::Mutex(absl::ConstInitType) : mu_(0) {}
inline CondVar::CondVar() : cv_(0) {}
// static
template <typename T>
bool Condition::CastAndCallMethod(const Condition *c) {
T *object = static_cast<T *>(c->arg_);
bool (T::*method_pointer)();
c->ReadCallback(&method_pointer);
return (object->*method_pointer)();
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
}
// static
template <typename T>
bool Condition::CastAndCallFunction(const Condition *c) {
bool (*function)(T *);
c->ReadCallback(&function);
T *argument = static_cast<T *>(c->arg_);
return (*function)(argument);
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
}
template <typename T>
inline Condition::Condition(bool (*func)(T *), T *arg)
: eval_(&CastAndCallFunction<T>),
arg_(const_cast<void *>(static_cast<const void *>(arg))) {
static_assert(sizeof(&func) <= sizeof(callback_),
"An overlarge function pointer was passed to Condition.");
StoreCallback(func);
}
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
template <typename T>
inline Condition::Condition(T *object,
bool (absl::internal::identity<T>::type::*method)())
: eval_(&CastAndCallMethod<T>),
arg_(object) {
static_assert(sizeof(&method) <= sizeof(callback_),
"An overlarge method pointer was passed to Condition.");
StoreCallback(method);
}
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
template <typename T>
inline Condition::Condition(const T *object,
bool (absl::internal::identity<T>::type::*method)()
const)
: eval_(&CastAndCallMethod<T>),
arg_(reinterpret_cast<void *>(const_cast<T *>(object))) {
StoreCallback(method);
}
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// Register hooks for profiling support.
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
// The function pointer registered here will be called whenever a mutex is
Export of internal Abseil changes -- 81f95fcf85b75b84f9892c73123501472b9cff33 by Martijn Vels <mvels@google.com>: Introduce GetEstimatedMemoryUsage(CordMemoryAccounting::kFairShare) Memory usage analysis is moved into a separate cord_analysis.cc source. PiperOrigin-RevId: 416370158 -- 6bc7b1348fd27fe53f100c9eabd47f4f2cb9c19c by Abseil Team <absl-team@google.com>: Support scoped enum in absl::Substitute. PiperOrigin-RevId: 416345422 -- 6399f4f6ae05ebcd67664ebd844902f699ab8ec7 by Abseil Team <absl-team@google.com>: Correct the computation of contention cycles Currently, we record contention cycles from the first time a thread started waiting on a mutex. Consider a situation in which two threads, T1 and T2, run a loop at the top of which they acquire a common mutex and release it at the end of the loop body. Further assume that T2 is never able to acquire the mutex as T1 repeatedly acquires and then releases the mutex. In this case, we would expect that the reported contention cycles would be increase linearly over time. But currently we observe a quadratic behavior in the reported waiting time as mentioned in b/14684244#comment10. To fix the issue, this CL records the contention cycles experienced by all the threads woken up when the mutex is released. Further, contention_start_cycles is set to the current time since the contention cycles for the time already passed has been taken into account. With this CL, we get a linear increase in the waiting time, the expected behavior. PiperOrigin-RevId: 416322593 -- 149c1637c8a0f1a38e5a8f9f27e5803a2015a554 by Jorg Brown <jorg@google.com>: Make Status::EmptyString more efficient by constructing it in global space, rather than on the heap. See https://godbolt.org/z/8M9n7YqcY for reduced code size. PiperOrigin-RevId: 416307833 -- 3b4562a8be5a3c80077cb67b0a32c97419058380 by Abseil Team <absl-team@google.com>: Clarify the usage of RegisterMutexProfiler PiperOrigin-RevId: 416146130 GitOrigin-RevId: 81f95fcf85b75b84f9892c73123501472b9cff33 Change-Id: Iccb72d7ee617e6ebe226a38170d62e0849b43480
3 years ago
// contended. The callback is given the cycles for which waiting happened (as
// measured by //absl/base/internal/cycleclock.h, and which may not
// be real "cycle" counts.)
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
//
// Calls to this function do not race or block, but there is no ordering
// guaranteed between calls to this function and call to the provided hook.
// In particular, the previously registered hook may still be called for some
// time after this function returns.
Export of internal Abseil changes -- 81f95fcf85b75b84f9892c73123501472b9cff33 by Martijn Vels <mvels@google.com>: Introduce GetEstimatedMemoryUsage(CordMemoryAccounting::kFairShare) Memory usage analysis is moved into a separate cord_analysis.cc source. PiperOrigin-RevId: 416370158 -- 6bc7b1348fd27fe53f100c9eabd47f4f2cb9c19c by Abseil Team <absl-team@google.com>: Support scoped enum in absl::Substitute. PiperOrigin-RevId: 416345422 -- 6399f4f6ae05ebcd67664ebd844902f699ab8ec7 by Abseil Team <absl-team@google.com>: Correct the computation of contention cycles Currently, we record contention cycles from the first time a thread started waiting on a mutex. Consider a situation in which two threads, T1 and T2, run a loop at the top of which they acquire a common mutex and release it at the end of the loop body. Further assume that T2 is never able to acquire the mutex as T1 repeatedly acquires and then releases the mutex. In this case, we would expect that the reported contention cycles would be increase linearly over time. But currently we observe a quadratic behavior in the reported waiting time as mentioned in b/14684244#comment10. To fix the issue, this CL records the contention cycles experienced by all the threads woken up when the mutex is released. Further, contention_start_cycles is set to the current time since the contention cycles for the time already passed has been taken into account. With this CL, we get a linear increase in the waiting time, the expected behavior. PiperOrigin-RevId: 416322593 -- 149c1637c8a0f1a38e5a8f9f27e5803a2015a554 by Jorg Brown <jorg@google.com>: Make Status::EmptyString more efficient by constructing it in global space, rather than on the heap. See https://godbolt.org/z/8M9n7YqcY for reduced code size. PiperOrigin-RevId: 416307833 -- 3b4562a8be5a3c80077cb67b0a32c97419058380 by Abseil Team <absl-team@google.com>: Clarify the usage of RegisterMutexProfiler PiperOrigin-RevId: 416146130 GitOrigin-RevId: 81f95fcf85b75b84f9892c73123501472b9cff33 Change-Id: Iccb72d7ee617e6ebe226a38170d62e0849b43480
3 years ago
void RegisterMutexProfiler(void (*fn)(int64_t wait_cycles));
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// Register a hook for Mutex tracing.
//
// The function pointer registered here will be called whenever a mutex is
// contended. The callback is given an opaque handle to the contended mutex,
// an event name, and the number of wait cycles (as measured by
// //absl/base/internal/cycleclock.h, and which may not be real
// "cycle" counts.)
//
// The only event name currently sent is "slow release".
//
// This has the same memory ordering concerns as RegisterMutexProfiler() above.
void RegisterMutexTracer(void (*fn)(const char *msg, const void *obj,
Export of internal Abseil changes -- a5af5874c1c5cc02bd2a748d455321f82b6f2a93 by Andy Getzendanner <durandal@google.com>: fix compile fails with asan and -Wredundant-decls Import of https://github.com/abseil/abseil-cpp/pull/801 PiperOrigin-RevId: 336693223 -- ed9df42ab2b742386c6692c2bed015374c919d9c by Derek Mauro <dmauro@google.com>: Fix integer conversion warning Fixes #814 PiperOrigin-RevId: 336651814 -- 0ab4c23884e72dce17b67c1eb520f9dbb802565d by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 336585378 -- eba0e3dccd52a6e91bcff84075bef0affc650b74 by Matt Kulukundis <kfm@google.com>: Add bitset operations to Futex helper. PiperOrigin-RevId: 336409368 -- 8b0709a8b4500bf5f0af4b602d76a298d81645e8 by Abseil Team <absl-team@google.com>: Fix code indentation in a comment. PiperOrigin-RevId: 336368167 -- bc3961c87a7e7760c10319a5b0349c279f7ae3ad by Samuel Benzaquen <sbenza@google.com>: Improve performance of the registry: - Reduce contention - Reduce memory usage for each flag by `6*sizeof(void*)`. - Replace one immortal allocation per-flag with a single one for all the flags - Slightly improve single-threaded performance by avoiding the std::map indirections. PiperOrigin-RevId: 336365904 -- 264ad9f28f935aad8b6b1437f8bf804fa9104346 by Abseil Team <absl-team@google.com>: Fix typo in comment on absl::Condition. PiperOrigin-RevId: 336311680 -- b5b808a8c75ca0df7b09eff9a423ec171d80f771 by Derek Mauro <dmauro@google.com>: Add missing Apache license headers PiperOrigin-RevId: 336294980 -- 89446c3a4793df8b95060385cf3e219357c3db1d by Andy Soffer <asoffer@google.com>: Internal changes PiperOrigin-RevId: 336287465 -- 57c8be4e294881bc79a6a44b8e4bf7ecbb19b9b9 by Matt Kulukundis <kfm@google.com>: Extract Futex from an implementation detail of Wait to a private interface. PiperOrigin-RevId: 336123209 GitOrigin-RevId: a5af5874c1c5cc02bd2a748d455321f82b6f2a93 Change-Id: Ie5a0ebe28e571814e3e11d4c05ca308523ccf311
4 years ago
int64_t wait_cycles));
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
// TODO(gfalcon): Combine RegisterMutexProfiler() and RegisterMutexTracer()
// into a single interface, since they are only ever called in pairs.
// Register a hook for CondVar tracing.
//
// The function pointer registered here will be called here on various CondVar
// events. The callback is given an opaque handle to the CondVar object and
// a string identifying the event. This is thread-safe, but only a single
// tracer can be registered.
//
// Events that can be sent are "Wait", "Unwait", "Signal wakeup", and
// "SignalAll wakeup".
//
// This has the same memory ordering concerns as RegisterMutexProfiler() above.
void RegisterCondVarTracer(void (*fn)(const char *msg, const void *cv));
// Register a hook for symbolizing stack traces in deadlock detector reports.
//
// 'pc' is the program counter being symbolized, 'out' is the buffer to write
// into, and 'out_size' is the size of the buffer. This function can return
// false if symbolizing failed, or true if a NUL-terminated symbol was written
// to 'out.'
//
// This has the same memory ordering concerns as RegisterMutexProfiler() above.
//
// DEPRECATED: The default symbolizer function is absl::Symbolize() and the
// ability to register a different hook for symbolizing stack traces will be
// removed on or after 2023-05-01.
ABSL_DEPRECATED("absl::RegisterSymbolizer() is deprecated and will be removed "
"on or after 2023-05-01")
void RegisterSymbolizer(bool (*fn)(const void *pc, char *out, int out_size));
// EnableMutexInvariantDebugging()
//
// Enable or disable global support for Mutex invariant debugging. If enabled,
// then invariant predicates can be registered per-Mutex for debug checking.
// See Mutex::EnableInvariantDebugging().
void EnableMutexInvariantDebugging(bool enabled);
// When in debug mode, and when the feature has been enabled globally, the
// implementation will keep track of lock ordering and complain (or optionally
// crash) if a cycle is detected in the acquired-before graph.
// Possible modes of operation for the deadlock detector in debug mode.
enum class OnDeadlockCycle {
kIgnore, // Neither report on nor attempt to track cycles in lock ordering
kReport, // Report lock cycles to stderr when detected
kAbort, // Report lock cycles to stderr when detected, then abort
};
// SetMutexDeadlockDetectionMode()
//
// Enable or disable global support for detection of potential deadlocks
// due to Mutex lock ordering inversions. When set to 'kIgnore', tracking of
// lock ordering is disabled. Otherwise, in debug builds, a lock ordering graph
// will be maintained internally, and detected cycles will be reported in
// the manner chosen here.
void SetMutexDeadlockDetectionMode(OnDeadlockCycle mode);
ABSL_NAMESPACE_END
} // namespace absl
// In some build configurations we pass --detect-odr-violations to the
// gold linker. This causes it to flag weak symbol overrides as ODR
// violations. Because ODR only applies to C++ and not C,
// --detect-odr-violations ignores symbols not mangled with C++ names.
// By changing our extension points to be extern "C", we dodge this
// check.
extern "C" {
void ABSL_INTERNAL_C_SYMBOL(AbslInternalMutexYield)();
Export of internal Abseil changes -- f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>: Slight restructuring of absl/random/internal randen implementation. Convert round-keys.inc into randen_round_keys.cc file. Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts. Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed. PiperOrigin-RevId: 312190313 -- dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 312167304 -- f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>: Cleanup StrFormat extensions a little. PiperOrigin-RevId: 312166336 -- 9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 312105213 -- 9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>: Complete IGNORE_TYPE macro renaming. PiperOrigin-RevId: 311999699 -- 64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>: Switch to fixed bytes specific default value. This fixes the Abseil Flags for big endian platforms. PiperOrigin-RevId: 311844448 -- bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>: Change many distribution tests to use pcg_engine as a deterministic source of entropy. It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests. Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values. This changelist does the latter. PiperOrigin-RevId: 311840096 GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266 Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
5 years ago
} // extern "C"
#endif // ABSL_SYNCHRONIZATION_MUTEX_H_