Improve KeepRunning loop performance to be similar to the range-based for. (#460)

This patch improves the performance of the KeepRunning loop in two ways: (A) it removes the dependency on the max_iterations variable, preventing it from being loaded every iteration. (B) it loops to zero, instead of to an upper bound. This allows a single decrement instruction to be used instead of a arithmetic op followed by a comparison.

Improve KeepRunning loop performance to be similar to the range-based for. (#460)
a37fc0c4 · Eric · Dominic Hamon · 2fc2ea0e · a37fc0c4 · a37fc0c4
Commit a37fc0c4 authored Oct 17, 2017 by Eric Committed by Dominic Hamon Oct 17, 2017
Hide whitespace changes
Inline Side-by-side

Showing with 7 additions and 6 deletions

benchmark.h include/benchmark/benchmark.h +2 -2

benchmark.cc src/benchmark.cc +5 -4

No files found.
--- a/include/benchmark/benchmark.h
+++ b/include/benchmark/benchmark.h
@@ -432,7 +432,7 @@ class State {
    if (BENCHMARK_BUILTIN_EXPECT(!started_, false)) {
      StartKeepRunning();
    }
-    bool const res = total_iterations_++ < max_iterations;
+    bool const res = --total_iterations_;
    if (BENCHMARK_BUILTIN_EXPECT(!res, false)) {
      FinishKeepRunning();
    }
@@ -559,7 +559,7 @@ class State {
  int range_y() const { return range(1); }

  BENCHMARK_ALWAYS_INLINE
-  size_t iterations() const { return total_iterations_; }
+  size_t iterations() const { return (max_iterations - total_iterations_) + 1; }

 private:
  bool started_;

--- a/src/benchmark.cc
+++ b/src/benchmark.cc
@@ -397,7 +397,7 @@ State::State(size_t max_iters, const std::vector<int>& ranges, int thread_i,
             internal::ThreadManager* manager)
    : started_(false),
      finished_(false),
-      total_iterations_(0),
+      total_iterations_(max_iters + 1),
      range_(ranges),
      bytes_processed_(0),
      items_processed_(0),
@@ -410,6 +410,7 @@ State::State(size_t max_iters, const std::vector<int>& ranges, int thread_i,
      timer_(timer),
      manager_(manager) {
  CHECK(max_iterations != 0) << "At least one iteration must be run";
+  CHECK(total_iterations_ != 0) << "max iterations wrapped around";
  CHECK_LT(thread_index, threads) << "thread_index must be less than threads";
 }

@@ -434,7 +435,7 @@ void State::SkipWithError(const char* msg) {
      manager_->results.has_error_ = true;
    }
  }
-  total_iterations_ = max_iterations;
+  total_iterations_ = 1;
  if (timer_->running()) timer_->StopTimer();
 }

@@ -459,8 +460,8 @@ void State::FinishKeepRunning() {
  if (!error_occurred_) {
    PauseTiming();
  }
-  // Total iterations now is one greater than max iterations. Fix this.
-  total_iterations_ = max_iterations;
+  // Total iterations has now wrapped around zero. Fix this.
+  total_iterations_ = 1;
  finished_ = true;
  manager_->StartStopBarrier();
 }