made toAppendDelim better

[folly.git] / folly / Benchmark.cpp

diff --git a/folly/Benchmark.cpp b/folly/Benchmark.cpp
index 545cefccb574a08cc065b132a97980eaf7876647..6259970872fe7ad26370672be7b2d95cb3546359 100644 (file)
--- a/folly/Benchmark.cpp
+++ b/folly/Benchmark.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2012 Facebook, Inc.
+ * Copyright 2014 Facebook, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -40,6 +40,9 @@ DEFINE_string(bm_regex, "",
 DEFINE_int64(bm_min_usec, 100,
              "Minimum # of microseconds we'll accept for each benchmark.");
 
+DEFINE_int64(bm_min_iters, 1,
+             "Minimum # of iterations we'll try for each benchmark.");
+
 DEFINE_int32(bm_max_secs, 1,
              "Maximum # of seconds we'll spend on each benchmark.");
 
@@ -48,12 +51,16 @@ namespace folly {
 
 BenchmarkSuspender::NanosecondsSpent BenchmarkSuspender::nsSpent;
 
-typedef function<uint64_t(unsigned int)> BenchmarkFun;
+typedef function<detail::TimeIterPair(unsigned int)> BenchmarkFun;
 static vector<tuple<const char*, const char*, BenchmarkFun>> benchmarks;
 
 // Add the global baseline
 BENCHMARK(globalBenchmarkBaseline) {
+#ifdef _MSC_VER
+  _ReadWriteBarrier();
+#else
   asm volatile("");
+#endif
 }
 
 void detail::addBenchmarkImpl(const char* file, const char* name,
@@ -207,7 +214,8 @@ static double runBenchmarkGetNSPerIteration(const BenchmarkFun& fun,
   // the clock resolution is worse than that, it will be larger. In
   // essence we're aiming at making the quantization noise 0.01%.
   static const auto minNanoseconds =
-    max(FLAGS_bm_min_usec * 1000UL, min(resolutionInNs * 100000, 1000000000UL));
+    max<uint64_t>(FLAGS_bm_min_usec * 1000UL,
+        min<uint64_t>(resolutionInNs * 100000, 1000000000ULL));
 
   // We do measurements in several epochs and take the minimum, to
   // account for jitter.
@@ -222,14 +230,15 @@ static double runBenchmarkGetNSPerIteration(const BenchmarkFun& fun,
   size_t actualEpochs = 0;
 
   for (; actualEpochs < epochs; ++actualEpochs) {
-    for (unsigned int n = 1; n < (1UL << 30); n *= 2) {
-      auto const nsecs = fun(n);
-      if (nsecs < minNanoseconds) {
+    for (unsigned int n = FLAGS_bm_min_iters; n < (1UL << 30); n *= 2) {
+      auto const nsecsAndIter = fun(n);
+      if (nsecsAndIter.first < minNanoseconds) {
         continue;
       }
       // We got an accurate enough timing, done. But only save if
       // smaller than the current result.
-      epochResults[actualEpochs] = max(0.0, double(nsecs) / n - globalBaseline);
+      epochResults[actualEpochs] = max(0.0, double(nsecsAndIter.first) /
+                                       nsecsAndIter.second - globalBaseline);
       // Done with the current epoch, we got a meaningful timing.
       break;
     }
@@ -263,7 +272,7 @@ static const ScaleInfo kTimeSuffixes[] {
   { 1E-9, "ns" },
   { 1E-12, "ps" },
   { 1E-15, "fs" },
-  { 0, NULL },
+  { 0, nullptr },
 };
 
 static const ScaleInfo kMetricSuffixes[] {
@@ -285,7 +294,7 @@ static const ScaleInfo kMetricSuffixes[] {
   { 1E-18, "a" }, // atto
   { 1E-21, "z" }, // zepto
   { 1E-24, "y" }, // yocto
-  { 0, NULL },
+  { 0, nullptr },
 };
 
 static string humanReadable(double n, unsigned int decimals,
@@ -296,7 +305,7 @@ static string humanReadable(double n, unsigned int decimals,
 
   const double absValue = fabs(n);
   const ScaleInfo* scale = scales;
-  while (absValue < scale[0].boundary && scale[1].suffix != NULL) {
+  while (absValue < scale[0].boundary && scale[1].suffix != nullptr) {
     ++scale;
   }
 
@@ -315,7 +324,7 @@ static string metricReadable(double n, unsigned int decimals) {
 static void printBenchmarkResultsAsTable(
   const vector<tuple<const char*, const char*, double> >& data) {
   // Width available
-  static const uint columns = 76;
+  static const unsigned int columns = 76;
 
   // Compute the longest benchmark name
   size_t longestName = 0;
@@ -420,7 +429,7 @@ void runBenchmarks() {
     get<2>(benchmarks.front()), 0);
   FOR_EACH_RANGE (i, 1, benchmarks.size()) {
     double elapsed = 0.0;
-    if (!strcmp(get<1>(benchmarks[i]), "-") == 0) { // skip separators
+    if (strcmp(get<1>(benchmarks[i]), "-") != 0) { // skip separators
       if (bmRegex && !boost::regex_search(get<1>(benchmarks[i]), *bmRegex)) {
         continue;
       }