Intermediate changes

Author: robot-piglet

contrib/restricted/google/benchmark/include/benchmark/benchmark.h

@@ -1004,7 +1004,8 @@ class BENCHMARK_EXPORT State {
   State(std::string name, IterationCount max_iters,
         const std::vector<int64_t>& ranges, int thread_i, int n_threads,
         internal::ThreadTimer* timer, internal::ThreadManager* manager,
-        internal::PerfCountersMeasurement* perf_counters_measurement);
+        internal::PerfCountersMeasurement* perf_counters_measurement,
+        ProfilerManager* profiler_manager);
 
   void StartKeepRunning();
   // Implementation of KeepRunning() and KeepRunningBatch().
@@ -1019,6 +1020,7 @@ class BENCHMARK_EXPORT State {
   internal::ThreadTimer* const timer_;
   internal::ThreadManager* const manager_;
   internal::PerfCountersMeasurement* const perf_counters_measurement_;
+  ProfilerManager* const profiler_manager_;
 
   friend class internal::BenchmarkInstance;
 };
@@ -1832,14 +1834,11 @@ class BENCHMARK_EXPORT BenchmarkReporter {
     internal::Skipped skipped;
     std::string skip_message;
 
-    // Total iterations across all threads.
     IterationCount iterations;
     int64_t threads;
     int64_t repetition_index;
     int64_t repetitions;
     TimeUnit time_unit;
-
-    // Total time across all threads.
     double real_accumulated_time;
     double cpu_accumulated_time;
 

contrib/restricted/google/benchmark/src/benchmark.cc

@@ -168,7 +168,8 @@ void UseCharPointer(char const volatile* const v) {
 State::State(std::string name, IterationCount max_iters,
              const std::vector<int64_t>& ranges, int thread_i, int n_threads,
              internal::ThreadTimer* timer, internal::ThreadManager* manager,
-             internal::PerfCountersMeasurement* perf_counters_measurement)
+             internal::PerfCountersMeasurement* perf_counters_measurement,
+             ProfilerManager* profiler_manager)
     : total_iterations_(0),
       batch_leftover_(0),
       max_iterations(max_iters),
@@ -182,7 +183,8 @@ State::State(std::string name, IterationCount max_iters,
       threads_(n_threads),
       timer_(timer),
       manager_(manager),
-      perf_counters_measurement_(perf_counters_measurement) {
+      perf_counters_measurement_(perf_counters_measurement),
+      profiler_manager_(profiler_manager) {
   BM_CHECK(max_iterations != 0) << "At least one iteration must be run";
   BM_CHECK_LT(thread_index_, threads_)
       << "thread_index must be less than threads";
@@ -207,7 +209,7 @@ State::State(std::string name, IterationCount max_iters,
 #if defined(__INTEL_COMPILER)
 #pragma warning push
 #pragma warning(disable : 1875)
-#elif defined(__GNUC__)
+#elif defined(__GNUC__) || defined(__clang__)
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Winvalid-offsetof"
 #endif
@@ -225,7 +227,7 @@ State::State(std::string name, IterationCount max_iters,
       offsetof(State, skipped_) <= (cache_line_size - sizeof(skipped_)), "");
 #if defined(__INTEL_COMPILER)
 #pragma warning pop
-#elif defined(__GNUC__)
+#elif defined(__GNUC__) || defined(__clang__)
 #pragma GCC diagnostic pop
 #endif
 #if defined(__NVCC__)
@@ -302,6 +304,8 @@ void State::StartKeepRunning() {
   BM_CHECK(!started_ && !finished_);
   started_ = true;
   total_iterations_ = skipped() ? 0 : max_iterations;
+  if (BENCHMARK_BUILTIN_EXPECT(profiler_manager_ != nullptr, false))
+    profiler_manager_->AfterSetupStart();
   manager_->StartStopBarrier();
   if (!skipped()) ResumeTiming();
 }
@@ -315,6 +319,8 @@ void State::FinishKeepRunning() {
   total_iterations_ = 0;
   finished_ = true;
   manager_->StartStopBarrier();
+  if (BENCHMARK_BUILTIN_EXPECT(profiler_manager_ != nullptr, false))
+    profiler_manager_->BeforeTeardownStop();
 }
 
 namespace internal {

contrib/restricted/google/benchmark/src/benchmark_api_internal.cc

@@ -92,25 +92,26 @@ BenchmarkInstance::BenchmarkInstance(Benchmark* benchmark, int family_idx,
 State BenchmarkInstance::Run(
     IterationCount iters, int thread_id, internal::ThreadTimer* timer,
     internal::ThreadManager* manager,
-    internal::PerfCountersMeasurement* perf_counters_measurement) const {
+    internal::PerfCountersMeasurement* perf_counters_measurement,
+    ProfilerManager* profiler_manager) const {
   State st(name_.function_name, iters, args_, thread_id, threads_, timer,
-           manager, perf_counters_measurement);
+           manager, perf_counters_measurement, profiler_manager);
   benchmark_.Run(st);
   return st;
 }
 
 void BenchmarkInstance::Setup() const {
   if (setup_) {
     State st(name_.function_name, /*iters*/ 1, args_, /*thread_id*/ 0, threads_,
-             nullptr, nullptr, nullptr);
+             nullptr, nullptr, nullptr, nullptr);
     setup_(st);
   }
 }
 
 void BenchmarkInstance::Teardown() const {
   if (teardown_) {
     State st(name_.function_name, /*iters*/ 1, args_, /*thread_id*/ 0, threads_,
-             nullptr, nullptr, nullptr);
+             nullptr, nullptr, nullptr, nullptr);
     teardown_(st);
   }
 }

contrib/restricted/google/benchmark/src/benchmark_api_internal.h

@@ -44,7 +44,8 @@ class BenchmarkInstance {
 
   State Run(IterationCount iters, int thread_id, internal::ThreadTimer* timer,
             internal::ThreadManager* manager,
-            internal::PerfCountersMeasurement* perf_counters_measurement) const;
+            internal::PerfCountersMeasurement* perf_counters_measurement,
+            ProfilerManager* profiler_manager) const;
 
  private:
   BenchmarkName name_;

contrib/restricted/google/benchmark/src/benchmark_runner.cc

@@ -93,7 +93,6 @@ BenchmarkReporter::Run CreateRunReport(
   report.repetitions = repeats;
 
   if (!report.skipped) {
-    // This is the total time across all threads.
     if (b.use_manual_time()) {
       report.real_accumulated_time = results.manual_time_used;
     } else {
@@ -126,14 +125,15 @@ BenchmarkReporter::Run CreateRunReport(
 // Adds the stats collected for the thread into manager->results.
 void RunInThread(const BenchmarkInstance* b, IterationCount iters,
                  int thread_id, ThreadManager* manager,
-                 PerfCountersMeasurement* perf_counters_measurement) {
+                 PerfCountersMeasurement* perf_counters_measurement,
+                 ProfilerManager* profiler_manager) {
   internal::ThreadTimer timer(
       b->measure_process_cpu_time()
           ? internal::ThreadTimer::CreateProcessCpuTime()
           : internal::ThreadTimer::Create());
 
-  State st =
-      b->Run(iters, thread_id, &timer, manager, perf_counters_measurement);
+  State st = b->Run(iters, thread_id, &timer, manager,
+                    perf_counters_measurement, profiler_manager);
   BM_CHECK(st.skipped() || st.iterations() >= st.max_iterations)
       << "Benchmark returned before State::KeepRunning() returned false!";
   {
@@ -269,12 +269,14 @@ BenchmarkRunner::IterationResults BenchmarkRunner::DoNIterations() {
   // Run all but one thread in separate threads
   for (std::size_t ti = 0; ti < pool.size(); ++ti) {
     pool[ti] = std::thread(&RunInThread, &b, iters, static_cast<int>(ti + 1),
-                           manager.get(), perf_counters_measurement_ptr);
+                           manager.get(), perf_counters_measurement_ptr,
+                           /*profiler_manager=*/nullptr);
   }
   // And run one thread here directly.
   // (If we were asked to run just one thread, we don't create new threads.)
   // Yes, we need to do this here *after* we start the separate threads.
-  RunInThread(&b, iters, 0, manager.get(), perf_counters_measurement_ptr);
+  RunInThread(&b, iters, 0, manager.get(), perf_counters_measurement_ptr,
+              /*profiler_manager=*/nullptr);
 
   // The main thread has finished. Now let's wait for the other threads.
   manager->WaitForAllThreads();
@@ -290,10 +292,6 @@ BenchmarkRunner::IterationResults BenchmarkRunner::DoNIterations() {
   // And get rid of the manager.
   manager.reset();
 
-  // If we were measuring whole-process CPU usage then each thread reports
-  // total CPU time of all threads. Divide by threads to get real value.
-  if (b.measure_process_cpu_time()) i.results.cpu_time_used /= b.threads();
-
   BM_VLOG(2) << "Ran in " << i.results.cpu_time_used << "/"
              << i.results.real_time_used << "\n";
 
@@ -309,9 +307,6 @@ BenchmarkRunner::IterationResults BenchmarkRunner::DoNIterations() {
     i.seconds = i.results.real_time_used;
   }
 
-  // Adjust time stats to average since they were reported by all threads.
-  i.seconds /= b.threads();
-
   return i;
 }
 
@@ -417,7 +412,8 @@ MemoryManager::Result* BenchmarkRunner::RunMemoryManager(
   manager.reset(new internal::ThreadManager(1));
   b.Setup();
   RunInThread(&b, memory_iterations, 0, manager.get(),
-              perf_counters_measurement_ptr);
+              perf_counters_measurement_ptr,
+              /*profiler_manager=*/nullptr);
   manager->WaitForAllThreads();
   manager.reset();
   b.Teardown();
@@ -431,11 +427,10 @@ void BenchmarkRunner::RunProfilerManager() {
   std::unique_ptr<internal::ThreadManager> manager;
   manager.reset(new internal::ThreadManager(1));
   b.Setup();
-  profiler_manager->AfterSetupStart();
   RunInThread(&b, profile_iterations, 0, manager.get(),
-              /*perf_counters_measurement_ptr=*/nullptr);
+              /*perf_counters_measurement_ptr=*/nullptr,
+              /*profiler_manager=*/profiler_manager);
   manager->WaitForAllThreads();
-  profiler_manager->BeforeTeardownStop();
   manager.reset();
   b.Teardown();
 }

contrib/restricted/google/benchmark/src/complexity.cc

@@ -27,7 +27,6 @@ namespace benchmark {
 
 // Internal function to calculate the different scalability forms
 BigOFunc* FittingCurve(BigO complexity) {
-  static const double kLog2E = 1.44269504088896340736;
   switch (complexity) {
     case oN:
       return [](IterationCount n) -> double { return static_cast<double>(n); };
@@ -36,15 +35,12 @@ BigOFunc* FittingCurve(BigO complexity) {
     case oNCubed:
       return [](IterationCount n) -> double { return std::pow(n, 3); };
     case oLogN:
-      /* Note: can't use log2 because Android's GNU STL lacks it */
-      return [](IterationCount n) {
-        return kLog2E * std::log(static_cast<double>(n));
+      return [](IterationCount n) -> double {
+        return std::log2(static_cast<double>(n));
       };
     case oNLogN:
-      /* Note: can't use log2 because Android's GNU STL lacks it */
-      return [](IterationCount n) {
-        return kLog2E * static_cast<double>(n) *
-               std::log(static_cast<double>(n));
+      return [](IterationCount n) -> double {
+        return static_cast<double>(n) * std::log2(static_cast<double>(n));
       };
     case o1:
     default:

contrib/restricted/google/benchmark/src/perf_counters.cc

@@ -218,7 +218,7 @@ PerfCounters PerfCounters::Create(
       GetErrorLogInstance() << "***WARNING*** Failed to start counters. "
                                "Claring out all counters.\n";
 
-      // Close all peformance counters
+      // Close all performance counters
       for (int id : counter_ids) {
         ::close(id);
       }

contrib/restricted/google/benchmark/test/ya.make

@@ -4,7 +4,7 @@ GTEST(benchmark_gtest)
 
 WITHOUT_LICENSE_TEXTS()
 
-VERSION(1.8.5)
+VERSION(1.9.0)
 
 LICENSE(Apache-2.0)
 

contrib/restricted/google/benchmark/tools/compare/ya.make

@@ -4,9 +4,9 @@ PY3_PROGRAM()
 
 WITHOUT_LICENSE_TEXTS()
 
-VERSION(1.8.5)
+VERSION(1.9.0)
 
-ORIGINAL_SOURCE(https://github.com/google/benchmark/archive/v1.8.5.tar.gz)
+ORIGINAL_SOURCE(https://github.com/google/benchmark/archive/v1.9.0.tar.gz)
 
 LICENSE(Apache-2.0)
 

contrib/restricted/google/benchmark/tools/ya.make

@@ -1,6 +1,6 @@
 # Generated by devtools/yamaker.
 
-VERSION(1.8.5)
+VERSION(1.9.0)
 
 IF (NOT USE_STL_SYSTEM)
     IF (NOT USE_SYSTEM_PYTHON OR NOT _SYSTEM_PYTHON27)