tvl-depot/absl/synchronization/internal/waiter.cc

473 lines
14 KiB
C++
Raw Normal View History

2017-09-19 22:54:40 +02:00
// 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
2017-09-19 22:54:40 +02:00
//
// 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.
#include "absl/synchronization/internal/waiter.h"
#include "absl/base/config.h"
#ifdef _WIN32
#include <windows.h>
#else
#include <pthread.h>
#include <sys/time.h>
#include <unistd.h>
#endif
#ifdef __linux__
#include <linux/futex.h>
#include <sys/syscall.h>
#endif
#ifdef ABSL_HAVE_SEMAPHORE_H
#include <semaphore.h>
#endif
#include <errno.h>
#include <stdio.h>
#include <time.h>
#include <atomic>
#include <cassert>
#include <cstdint>
#include <new>
#include <type_traits>
Export of internal Abseil changes. -- bdce7e57e9e886eff1114d0266781b443f7ec639 by Derek Mauro <dmauro@google.com>: Change {Get|Set}EnvironmentVariable to {Get|Set}EnvironmentVariableA for compatibility with /DUNICODE. PiperOrigin-RevId: 239229514 -- 2276ed502326a044a84060d34eb19d499e3a3be2 by Derek Mauro <dmauro@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 239228622 -- a462efb970ff43b08a362ef2343fb75ac1295a50 by Derek Mauro <dmauro@google.com>: Adding linking of CoreFoundation to CMakeLists in absl/time. Import https://github.com/abseil/abseil-cpp/pull/280. Fix #283 PiperOrigin-RevId: 239220785 -- fc23327b97f940c682aae1956cf7a1bf87f88c06 by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of gcc (8.3.0 today) with libstdc++ and bazel. PiperOrigin-RevId: 239220448 -- 418c08a8f6a53e63b84e39473035774417ca3aa7 by Derek Mauro <dmauro@google.com>: Disable part of the variant exeception safety test on move assignment when using versions of libstd++ that contain a bug. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87431#c7 PiperOrigin-RevId: 239062455 -- 799722217aeda79679577843c91d5be62cbcbb42 by Matt Calabrese <calabrese@google.com>: Add internal-only IsSwappable traits corresponding to std::is_swappable and std::is_nothrow_swappable, which are used with the swap implementations of optional and variant. PiperOrigin-RevId: 239049448 -- aa46a036038a3de5c68ac5e5d3b4bf76f818d2ea by CJ Johnson <johnsoncj@google.com>: Make InlinedVectorStorage constructor explicit PiperOrigin-RevId: 239044361 -- 17949715b3aa21c794701f69f2154e91b6acabc3 by CJ Johnson <johnsoncj@google.com>: Add absl namesapce to internal/inlined_vector.h PiperOrigin-RevId: 239030789 -- 834628325953078cc08ed10d23bb8890e5bec897 by Derek Mauro <dmauro@google.com>: Add test script that uses Docker to build Abseil with gcc-4.8, libstdc++, and cmake. PiperOrigin-RevId: 239028433 -- 80fe24149ed73ed2ced995ad1e372fb060c60427 by CJ Johnson <johnsoncj@google.com>: Factors data members of InlinedVector into an impl type called InlinedVectorStorage so that (in future changes) the contents of a vector can be grouped together with a single pointer. PiperOrigin-RevId: 239021086 -- 585331436d5d4d79f845e45dcf79d918a0dc6169 by Derek Mauro <dmauro@google.com>: Add -Wno-missing-field-initializers to gcc compiler flags. gcc-4.x has spurious missing field initializer warnings. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36750 PiperOrigin-RevId: 239017217 -- 94602fe4e33ee3a552a7f2939c0f57a992f55075 by Abseil Team <absl-team@google.com>: Formatting fixes. PiperOrigin-RevId: 238983038 -- a1c1b63c08505574e0a8c491561840cecb2bb93e by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libc++ and bazel. PiperOrigin-RevId: 238669118 -- e525f8d20bc2f79a0d69336b902f63858f3bff9d by Derek Mauro <dmauro@google.com>: Disable the test optionalTest.InPlaceTSFINAEBug until libc++ is updated. PiperOrigin-RevId: 238661703 -- f99a2a0b5ec424a059678f7f226600f137b4c74e by Derek Mauro <dmauro@google.com>: Correct the check for the FlatHashMap-Any test bug (list conditions instead of platforms when possible) PiperOrigin-RevId: 238653344 -- 777928035dbcbf39f361eb7d10dc3696822f692f by Jon Cohen <cohenjon@google.com>: Add install rules for Abseil CMake. These are attempted to be limited to in-project installation. This serves two purposes -- first it's morally the same as using Abseil in-source, except you don't have to rebuild us every time. Second, the presence of an install rule makes life massively simpler for package manager maintainers. Currently this doesn't install absl tests or testonly libraries. This can be added in a follow-up patch. Fixes #38, Fixes #80, Closes #182 PiperOrigin-RevId: 238645836 -- ded1c6ce697c191b7a6ff14572b3e6d183117b2c by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libstdc++ and bazel. PiperOrigin-RevId: 238517815 GitOrigin-RevId: bdce7e57e9e886eff1114d0266781b443f7ec639 Change-Id: I6f745869cb8ef63851891ccac05ae9a7dd241c4f
2019-03-19 19:14:01 +01:00
2017-09-19 22:54:40 +02:00
#include "absl/base/internal/raw_logging.h"
#include "absl/base/internal/thread_identity.h"
#include "absl/base/optimization.h"
2017-09-19 22:54:40 +02:00
#include "absl/synchronization/internal/kernel_timeout.h"
namespace absl {
namespace synchronization_internal {
static void MaybeBecomeIdle() {
base_internal::ThreadIdentity *identity =
base_internal::CurrentThreadIdentityIfPresent();
assert(identity != nullptr);
const bool is_idle = identity->is_idle.load(std::memory_order_relaxed);
const int ticker = identity->ticker.load(std::memory_order_relaxed);
const int wait_start = identity->wait_start.load(std::memory_order_relaxed);
if (!is_idle && ticker - wait_start > Waiter::kIdlePeriods) {
identity->is_idle.store(true, std::memory_order_relaxed);
}
}
#if ABSL_WAITER_MODE == ABSL_WAITER_MODE_FUTEX
// Some Android headers are missing these definitions even though they
// support these futex operations.
#ifdef __BIONIC__
#ifndef SYS_futex
#define SYS_futex __NR_futex
#endif
#ifndef FUTEX_WAIT_BITSET
#define FUTEX_WAIT_BITSET 9
#endif
#ifndef FUTEX_PRIVATE_FLAG
#define FUTEX_PRIVATE_FLAG 128
#endif
#ifndef FUTEX_CLOCK_REALTIME
#define FUTEX_CLOCK_REALTIME 256
#endif
#ifndef FUTEX_BITSET_MATCH_ANY
#define FUTEX_BITSET_MATCH_ANY 0xFFFFFFFF
#endif
#endif
Export of internal Abseil changes. -- bdce7e57e9e886eff1114d0266781b443f7ec639 by Derek Mauro <dmauro@google.com>: Change {Get|Set}EnvironmentVariable to {Get|Set}EnvironmentVariableA for compatibility with /DUNICODE. PiperOrigin-RevId: 239229514 -- 2276ed502326a044a84060d34eb19d499e3a3be2 by Derek Mauro <dmauro@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 239228622 -- a462efb970ff43b08a362ef2343fb75ac1295a50 by Derek Mauro <dmauro@google.com>: Adding linking of CoreFoundation to CMakeLists in absl/time. Import https://github.com/abseil/abseil-cpp/pull/280. Fix #283 PiperOrigin-RevId: 239220785 -- fc23327b97f940c682aae1956cf7a1bf87f88c06 by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of gcc (8.3.0 today) with libstdc++ and bazel. PiperOrigin-RevId: 239220448 -- 418c08a8f6a53e63b84e39473035774417ca3aa7 by Derek Mauro <dmauro@google.com>: Disable part of the variant exeception safety test on move assignment when using versions of libstd++ that contain a bug. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87431#c7 PiperOrigin-RevId: 239062455 -- 799722217aeda79679577843c91d5be62cbcbb42 by Matt Calabrese <calabrese@google.com>: Add internal-only IsSwappable traits corresponding to std::is_swappable and std::is_nothrow_swappable, which are used with the swap implementations of optional and variant. PiperOrigin-RevId: 239049448 -- aa46a036038a3de5c68ac5e5d3b4bf76f818d2ea by CJ Johnson <johnsoncj@google.com>: Make InlinedVectorStorage constructor explicit PiperOrigin-RevId: 239044361 -- 17949715b3aa21c794701f69f2154e91b6acabc3 by CJ Johnson <johnsoncj@google.com>: Add absl namesapce to internal/inlined_vector.h PiperOrigin-RevId: 239030789 -- 834628325953078cc08ed10d23bb8890e5bec897 by Derek Mauro <dmauro@google.com>: Add test script that uses Docker to build Abseil with gcc-4.8, libstdc++, and cmake. PiperOrigin-RevId: 239028433 -- 80fe24149ed73ed2ced995ad1e372fb060c60427 by CJ Johnson <johnsoncj@google.com>: Factors data members of InlinedVector into an impl type called InlinedVectorStorage so that (in future changes) the contents of a vector can be grouped together with a single pointer. PiperOrigin-RevId: 239021086 -- 585331436d5d4d79f845e45dcf79d918a0dc6169 by Derek Mauro <dmauro@google.com>: Add -Wno-missing-field-initializers to gcc compiler flags. gcc-4.x has spurious missing field initializer warnings. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36750 PiperOrigin-RevId: 239017217 -- 94602fe4e33ee3a552a7f2939c0f57a992f55075 by Abseil Team <absl-team@google.com>: Formatting fixes. PiperOrigin-RevId: 238983038 -- a1c1b63c08505574e0a8c491561840cecb2bb93e by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libc++ and bazel. PiperOrigin-RevId: 238669118 -- e525f8d20bc2f79a0d69336b902f63858f3bff9d by Derek Mauro <dmauro@google.com>: Disable the test optionalTest.InPlaceTSFINAEBug until libc++ is updated. PiperOrigin-RevId: 238661703 -- f99a2a0b5ec424a059678f7f226600f137b4c74e by Derek Mauro <dmauro@google.com>: Correct the check for the FlatHashMap-Any test bug (list conditions instead of platforms when possible) PiperOrigin-RevId: 238653344 -- 777928035dbcbf39f361eb7d10dc3696822f692f by Jon Cohen <cohenjon@google.com>: Add install rules for Abseil CMake. These are attempted to be limited to in-project installation. This serves two purposes -- first it's morally the same as using Abseil in-source, except you don't have to rebuild us every time. Second, the presence of an install rule makes life massively simpler for package manager maintainers. Currently this doesn't install absl tests or testonly libraries. This can be added in a follow-up patch. Fixes #38, Fixes #80, Closes #182 PiperOrigin-RevId: 238645836 -- ded1c6ce697c191b7a6ff14572b3e6d183117b2c by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libstdc++ and bazel. PiperOrigin-RevId: 238517815 GitOrigin-RevId: bdce7e57e9e886eff1114d0266781b443f7ec639 Change-Id: I6f745869cb8ef63851891ccac05ae9a7dd241c4f
2019-03-19 19:14:01 +01:00
class Futex {
public:
static int WaitUntil(std::atomic<int32_t> *v, int32_t val,
KernelTimeout t) {
int err = 0;
if (t.has_timeout()) {
// https://locklessinc.com/articles/futex_cheat_sheet/
// Unlike FUTEX_WAIT, FUTEX_WAIT_BITSET uses absolute time.
struct timespec abs_timeout = t.MakeAbsTimespec();
// Atomically check that the futex value is still 0, and if it
// is, sleep until abs_timeout or until woken by FUTEX_WAKE.
err = syscall(
SYS_futex, reinterpret_cast<int32_t *>(v),
FUTEX_WAIT_BITSET | FUTEX_PRIVATE_FLAG | FUTEX_CLOCK_REALTIME, val,
&abs_timeout, nullptr, FUTEX_BITSET_MATCH_ANY);
} else {
// Atomically check that the futex value is still 0, and if it
// is, sleep until woken by FUTEX_WAKE.
err = syscall(SYS_futex, reinterpret_cast<int32_t *>(v),
FUTEX_WAIT | FUTEX_PRIVATE_FLAG, val, nullptr);
}
if (err != 0) {
err = -errno;
}
return err;
}
static int Wake(std::atomic<int32_t> *v, int32_t count) {
int err = syscall(SYS_futex, reinterpret_cast<int32_t *>(v),
FUTEX_WAKE | FUTEX_PRIVATE_FLAG, count);
if (ABSL_PREDICT_FALSE(err < 0)) {
err = -errno;
}
return err;
}
};
2017-09-19 22:54:40 +02:00
void Waiter::Init() {
futex_.store(0, std::memory_order_relaxed);
}
bool Waiter::Wait(KernelTimeout t) {
// Loop until we can atomically decrement futex from a positive
// value, waiting on a futex while we believe it is zero.
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
// Note that, since the thread ticker is just reset, we don't need to check
// whether the thread is idle on the very first pass of the loop.
bool first_pass = true;
2017-09-19 22:54:40 +02:00
while (true) {
int32_t x = futex_.load(std::memory_order_relaxed);
2017-09-19 22:54:40 +02:00
if (x != 0) {
if (!futex_.compare_exchange_weak(x, x - 1,
std::memory_order_acquire,
std::memory_order_relaxed)) {
continue; // Raced with someone, retry.
}
return true; // Consumed a wakeup, we are done.
}
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
if (!first_pass) MaybeBecomeIdle();
const int err = Futex::WaitUntil(&futex_, 0, t);
2017-09-19 22:54:40 +02:00
if (err != 0) {
if (err == -EINTR || err == -EWOULDBLOCK) {
2017-09-19 22:54:40 +02:00
// Do nothing, the loop will retry.
} else if (err == -ETIMEDOUT) {
return false;
2017-09-19 22:54:40 +02:00
} else {
ABSL_RAW_LOG(FATAL, "Futex operation failed with error %d\n", err);
2017-09-19 22:54:40 +02:00
}
}
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
first_pass = false;
2017-09-19 22:54:40 +02:00
}
}
void Waiter::Post() {
if (futex_.fetch_add(1, std::memory_order_release) == 0) {
// We incremented from 0, need to wake a potential waker.
Poke();
}
}
void Waiter::Poke() {
// Wake one thread waiting on the futex.
const int err = Futex::Wake(&futex_, 1);
if (ABSL_PREDICT_FALSE(err < 0)) {
ABSL_RAW_LOG(FATAL, "Futex operation failed with error %d\n", err);
2017-09-19 22:54:40 +02:00
}
}
#elif ABSL_WAITER_MODE == ABSL_WAITER_MODE_CONDVAR
class PthreadMutexHolder {
public:
explicit PthreadMutexHolder(pthread_mutex_t *mu) : mu_(mu) {
const int err = pthread_mutex_lock(mu_);
if (err != 0) {
ABSL_RAW_LOG(FATAL, "pthread_mutex_lock failed: %d", err);
}
}
PthreadMutexHolder(const PthreadMutexHolder &rhs) = delete;
PthreadMutexHolder &operator=(const PthreadMutexHolder &rhs) = delete;
~PthreadMutexHolder() {
const int err = pthread_mutex_unlock(mu_);
if (err != 0) {
ABSL_RAW_LOG(FATAL, "pthread_mutex_unlock failed: %d", err);
}
}
private:
pthread_mutex_t *mu_;
};
void Waiter::Init() {
const int err = pthread_mutex_init(&mu_, 0);
if (err != 0) {
ABSL_RAW_LOG(FATAL, "pthread_mutex_init failed: %d", err);
}
const int err2 = pthread_cond_init(&cv_, 0);
if (err2 != 0) {
ABSL_RAW_LOG(FATAL, "pthread_cond_init failed: %d", err2);
}
waiter_count_.store(0, std::memory_order_relaxed);
wakeup_count_.store(0, std::memory_order_relaxed);
}
bool Waiter::Wait(KernelTimeout t) {
struct timespec abs_timeout;
if (t.has_timeout()) {
abs_timeout = t.MakeAbsTimespec();
}
PthreadMutexHolder h(&mu_);
waiter_count_.fetch_add(1, std::memory_order_relaxed);
// Loop until we find a wakeup to consume or timeout.
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
// Note that, since the thread ticker is just reset, we don't need to check
// whether the thread is idle on the very first pass of the loop.
bool first_pass = true;
2017-09-19 22:54:40 +02:00
while (true) {
int x = wakeup_count_.load(std::memory_order_relaxed);
if (x != 0) {
if (!wakeup_count_.compare_exchange_weak(x, x - 1,
std::memory_order_acquire,
std::memory_order_relaxed)) {
continue; // Raced with someone, retry.
}
// Successfully consumed a wakeup, we're done.
waiter_count_.fetch_sub(1, std::memory_order_relaxed);
return true;
}
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
if (!first_pass) MaybeBecomeIdle();
2017-09-19 22:54:40 +02:00
// No wakeups available, time to wait.
if (!t.has_timeout()) {
const int err = pthread_cond_wait(&cv_, &mu_);
if (err != 0) {
ABSL_RAW_LOG(FATAL, "pthread_cond_wait failed: %d", err);
}
} else {
const int err = pthread_cond_timedwait(&cv_, &mu_, &abs_timeout);
if (err == ETIMEDOUT) {
waiter_count_.fetch_sub(1, std::memory_order_relaxed);
return false;
}
if (err != 0) {
ABSL_RAW_LOG(FATAL, "pthread_cond_wait failed: %d", err);
}
}
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
first_pass = false;
2017-09-19 22:54:40 +02:00
}
}
void Waiter::Post() {
wakeup_count_.fetch_add(1, std::memory_order_release);
Poke();
}
void Waiter::Poke() {
if (waiter_count_.load(std::memory_order_relaxed) == 0) {
return;
}
// Potentially a waker. Take the lock and check again.
PthreadMutexHolder h(&mu_);
if (waiter_count_.load(std::memory_order_relaxed) == 0) {
return;
}
const int err = pthread_cond_signal(&cv_);
if (err != 0) {
ABSL_RAW_LOG(FATAL, "pthread_cond_signal failed: %d", err);
}
}
#elif ABSL_WAITER_MODE == ABSL_WAITER_MODE_SEM
void Waiter::Init() {
if (sem_init(&sem_, 0, 0) != 0) {
ABSL_RAW_LOG(FATAL, "sem_init failed with errno %d\n", errno);
}
wakeups_.store(0, std::memory_order_relaxed);
}
bool Waiter::Wait(KernelTimeout t) {
struct timespec abs_timeout;
if (t.has_timeout()) {
abs_timeout = t.MakeAbsTimespec();
}
// Loop until we timeout or consume a wakeup.
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
// Note that, since the thread ticker is just reset, we don't need to check
// whether the thread is idle on the very first pass of the loop.
bool first_pass = true;
2017-09-19 22:54:40 +02:00
while (true) {
int x = wakeups_.load(std::memory_order_relaxed);
if (x != 0) {
if (!wakeups_.compare_exchange_weak(x, x - 1,
std::memory_order_acquire,
std::memory_order_relaxed)) {
continue; // Raced with someone, retry.
}
// Successfully consumed a wakeup, we're done.
return true;
}
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
if (!first_pass) MaybeBecomeIdle();
2017-09-19 22:54:40 +02:00
// Nothing to consume, wait (looping on EINTR).
while (true) {
if (!t.has_timeout()) {
if (sem_wait(&sem_) == 0) break;
if (errno == EINTR) continue;
ABSL_RAW_LOG(FATAL, "sem_wait failed: %d", errno);
} else {
if (sem_timedwait(&sem_, &abs_timeout) == 0) break;
if (errno == EINTR) continue;
if (errno == ETIMEDOUT) return false;
ABSL_RAW_LOG(FATAL, "sem_timedwait failed: %d", errno);
}
}
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
first_pass = false;
2017-09-19 22:54:40 +02:00
}
}
void Waiter::Post() {
wakeups_.fetch_add(1, std::memory_order_release); // Post a wakeup.
Poke();
}
void Waiter::Poke() {
if (sem_post(&sem_) != 0) { // Wake any semaphore waiter.
ABSL_RAW_LOG(FATAL, "sem_post failed with errno %d\n", errno);
}
}
#elif ABSL_WAITER_MODE == ABSL_WAITER_MODE_WIN32
class Waiter::WinHelper {
public:
static SRWLOCK *GetLock(Waiter *w) {
return reinterpret_cast<SRWLOCK *>(&w->mu_storage_);
}
static CONDITION_VARIABLE *GetCond(Waiter *w) {
return reinterpret_cast<CONDITION_VARIABLE *>(&w->cv_storage_);
}
static_assert(sizeof(SRWLOCK) == sizeof(Waiter::SRWLockStorage),
"SRWLockStorage does not have the same size as SRWLOCK");
static_assert(
alignof(SRWLOCK) == alignof(Waiter::SRWLockStorage),
"SRWLockStorage does not have the same alignment as SRWLOCK");
static_assert(sizeof(CONDITION_VARIABLE) ==
sizeof(Waiter::ConditionVariableStorage),
"ABSL_CONDITION_VARIABLE_STORAGE does not have the same size "
"as CONDITION_VARIABLE");
static_assert(alignof(CONDITION_VARIABLE) ==
alignof(Waiter::ConditionVariableStorage),
"ConditionVariableStorage does not have the same "
"alignment as CONDITION_VARIABLE");
// The SRWLOCK and CONDITION_VARIABLE types must be trivially constructible
// and destructible because we never call their constructors or destructors.
static_assert(std::is_trivially_constructible<SRWLOCK>::value,
"The SRWLOCK type must be trivially constructible");
static_assert(std::is_trivially_constructible<CONDITION_VARIABLE>::value,
"The CONDITION_VARIABLE type must be trivially constructible");
static_assert(std::is_trivially_destructible<SRWLOCK>::value,
"The SRWLOCK type must be trivially destructible");
static_assert(std::is_trivially_destructible<CONDITION_VARIABLE>::value,
"The CONDITION_VARIABLE type must be trivially destructible");
};
2017-09-19 22:54:40 +02:00
class LockHolder {
public:
explicit LockHolder(SRWLOCK* mu) : mu_(mu) {
AcquireSRWLockExclusive(mu_);
}
LockHolder(const LockHolder&) = delete;
LockHolder& operator=(const LockHolder&) = delete;
~LockHolder() {
ReleaseSRWLockExclusive(mu_);
}
private:
SRWLOCK* mu_;
};
void Waiter::Init() {
auto *mu = ::new (static_cast<void *>(&mu_storage_)) SRWLOCK;
auto *cv = ::new (static_cast<void *>(&cv_storage_)) CONDITION_VARIABLE;
InitializeSRWLock(mu);
InitializeConditionVariable(cv);
2017-09-19 22:54:40 +02:00
waiter_count_.store(0, std::memory_order_relaxed);
wakeup_count_.store(0, std::memory_order_relaxed);
}
bool Waiter::Wait(KernelTimeout t) {
SRWLOCK *mu = WinHelper::GetLock(this);
CONDITION_VARIABLE *cv = WinHelper::GetCond(this);
LockHolder h(mu);
2017-09-19 22:54:40 +02:00
waiter_count_.fetch_add(1, std::memory_order_relaxed);
// Loop until we find a wakeup to consume or timeout.
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
// Note that, since the thread ticker is just reset, we don't need to check
// whether the thread is idle on the very first pass of the loop.
bool first_pass = true;
2017-09-19 22:54:40 +02:00
while (true) {
int x = wakeup_count_.load(std::memory_order_relaxed);
if (x != 0) {
if (!wakeup_count_.compare_exchange_weak(x, x - 1,
std::memory_order_acquire,
std::memory_order_relaxed)) {
continue; // Raced with someone, retry.
}
// Successfully consumed a wakeup, we're done.
waiter_count_.fetch_sub(1, std::memory_order_relaxed);
return true;
}
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
if (!first_pass) MaybeBecomeIdle();
2017-09-19 22:54:40 +02:00
// No wakeups available, time to wait.
if (!SleepConditionVariableSRW(cv, mu, t.InMillisecondsFromNow(), 0)) {
2017-09-19 22:54:40 +02:00
// GetLastError() returns a Win32 DWORD, but we assign to
// unsigned long to simplify the ABSL_RAW_LOG case below. The uniform
// initialization guarantees this is not a narrowing conversion.
const unsigned long err{GetLastError()}; // NOLINT(runtime/int)
if (err == ERROR_TIMEOUT) {
waiter_count_.fetch_sub(1, std::memory_order_relaxed);
return false;
} else {
ABSL_RAW_LOG(FATAL, "SleepConditionVariableSRW failed: %lu", err);
}
}
Export of internal Abseil changes -- f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>: Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS. ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel). PiperOrigin-RevId: 274932312 -- c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>: Silence MSAN in the stack consumption test utility PiperOrigin-RevId: 274912950 -- 2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>: Internal change PiperOrigin-RevId: 274847103 -- 75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>: Reformat BUILD file to current standards. PiperOrigin-RevId: 274815392 -- a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>: Fix invalid result regarding leading zeros in the exponent. PiperOrigin-RevId: 274808017 -- dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>: Make string_view's relational operators constexpr when possible. PiperOrigin-RevId: 274807873 -- b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>: Internal rework. PiperOrigin-RevId: 274787159 -- 70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>: Internal rework. PiperOrigin-RevId: 274715557 -- 14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>: Make deprecated statements about ParseFLag/UnparseFlag consistent in a file. PiperOrigin-RevId: 274668123 -- 2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>: Allow absl::c_equal to be used with arrays. This is achieved by allowing container size computation for arrays. PiperOrigin-RevId: 274426830 -- 219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>: Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG PiperOrigin-RevId: 274273788 -- 47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>: Rework of flags persistence/FlagSaver internals. PiperOrigin-RevId: 274225213 -- 7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>: Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems. PiperOrigin-RevId: 274061877 -- 561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>: Fix ABSL_WAITER_MODE detection for mingw Import of https://github.com/abseil/abseil-cpp/pull/342 PiperOrigin-RevId: 274030071 -- 9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>: Support using Abseil with the musl libc implementation. Only test changes were required: * Workaround for a bug in sigaltstack() on musl * printf-style pointer formatting (%p) is implementation defined, so verify StrFromat produces something compatible * Fix detection of feenableexcept() PiperOrigin-RevId: 274011666 -- 73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>: nvcc workaround: explicitly specify the definition of node_handle::Base PiperOrigin-RevId: 274011392 -- ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>: Internal change PiperOrigin-RevId: 273996318 -- e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>: Introduce a type alias to work around an nvcc bug. On the previous code, nvcc gets confused thinking that T has to be a parameter pack, as IsDecomposable accepts one. PiperOrigin-RevId: 273980472 -- 105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 273955589 -- 8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>: Avoid dual-exporting scheduling_mode.h PiperOrigin-RevId: 273825112 -- fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>: Fix ABSL_HAVE_ALARM check on mingw Import of https://github.com/abseil/abseil-cpp/pull/341 PiperOrigin-RevId: 273817839 -- 6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>: Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h Tested: TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5 PiperOrigin-RevId: 273810331 -- 6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 273589963 -- 91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>: Update macro name for `ABSL_GUARDED_BY()` in the example section. PiperOrigin-RevId: 273286983 -- 0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>: Fix potential integer overflow in the absl time library. In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation. PiperOrigin-RevId: 273092952 -- b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>: Correctly Unlock the global mutex in case of concurrent flag initialization. Fixes #386 PiperOrigin-RevId: 272979749 -- c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>: Try to become idle only when there is no wakeup. Immediately after waking up (when futex wait returns), the current thread tries to become idle doing bunch of memory loads and a branch. Problem is that there is a good chance that we woke up due to a wakeup, especially for actively used threads. For such wakeups, calling MaybeBecomeIdle() would be a waste of cycles. Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup. For idle threads the net effect should be the same. For active, threads this will be more efficient. Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the current thread will try to become idle before sleeping. This should result in more accurate idleness and more efficient release of thread resources. PiperOrigin-RevId: 272940381 GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85 Change-Id: I36de05aec12595183725652dd362dfa58fb095d0
2019-10-16 03:18:40 +02:00
first_pass = false;
2017-09-19 22:54:40 +02:00
}
}
void Waiter::Post() {
wakeup_count_.fetch_add(1, std::memory_order_release);
Poke();
}
void Waiter::Poke() {
if (waiter_count_.load(std::memory_order_relaxed) == 0) {
return;
}
// Potentially a waker. Take the lock and check again.
LockHolder h(WinHelper::GetLock(this));
2017-09-19 22:54:40 +02:00
if (waiter_count_.load(std::memory_order_relaxed) == 0) {
return;
}
WakeConditionVariable(WinHelper::GetCond(this));
2017-09-19 22:54:40 +02:00
}
#else
#error Unknown ABSL_WAITER_MODE
#endif
} // namespace synchronization_internal
} // namespace absl