Add pre-allocation to memory measures.

This avoids issues with doubling the measurement vector sizes influencing memory measures.

app/benchmark_fft/main.cpp

@@ -39,6 +39,7 @@ int main(int argc, char **argv) {
   string full_directory = directory + "/PLS_v3/";
   benchmark_runner runner{full_directory, test_name};
   runner.enable_memory_stats();
+  runner.pre_allocate_stats();
 
   fft::complex_vector data(fft::SIZE);
   fft::complex_vector swap_array(fft::SIZE);
@@ -47,12 +48,12 @@ int main(int argc, char **argv) {
   pls::scheduler scheduler{(unsigned) num_threads, MAX_NUM_TASKS, MAX_STACK_SIZE};
 
 //  scheduler.get_profiler().disable_memory_measure();
-  runner.run_iterations(10, [&]() {
+  runner.run_iterations(fft::NUM_ITERATIONS, [&]() {
     scheduler.perform_work([&]() {
       pls_conquer(data.begin(), swap_array.begin(), fft::SIZE);;
     });
 //    scheduler.get_profiler().current_run().print_stats();
-  }, 1, [&]() {
+  }, fft::NUM_WARMUP_ITERATIONS, [&]() {
     fft::fill_input(data); // Reset data before each run
   });
 //  scheduler.get_profiler().current_run().print_dag(std::cout);

app/benchmark_fib/main.cpp

@@ -37,6 +37,7 @@ int main(int argc, char **argv) {
   string full_directory = directory + "/PLS_v3/";
   benchmark_runner runner{full_directory, test_name};
   runner.enable_memory_stats();
+  runner.pre_allocate_stats();
 
   pls::scheduler scheduler{(unsigned) num_threads, MAX_NUM_TASKS, MAX_STACK_SIZE};
 

app/benchmark_matrix/main.cpp

@@ -32,6 +32,7 @@ int main(int argc, char **argv) {
   string full_directory = directory + "/PLS_v3/";
   benchmark_runner runner{full_directory, test_name};
   runner.enable_memory_stats();
+  runner.pre_allocate_stats();
 
   pls_matrix<double, matrix::MATRIX_SIZE> a;
   pls_matrix<double, matrix::MATRIX_SIZE> b;

app/benchmark_matrix_div_conquer/main.cpp

@@ -113,6 +113,7 @@ int main(int argc, char **argv) {
   string full_directory = directory + "/PLS_v3/";
   benchmark_runner runner{full_directory, test_name};
   runner.enable_memory_stats();
+  runner.pre_allocate_stats();
 
   // Only run on one version to avoid copy
   std::unique_ptr<double[]> result_data{new double[size * size]};

app/benchmark_unbalanced/main.cpp

@@ -43,6 +43,7 @@ int main(int argc, char **argv) {
   string full_directory = directory + "/PLS_v3/";
   benchmark_runner runner{full_directory, test_name};
   runner.enable_memory_stats();
+  runner.pre_allocate_stats();
 
   scheduler scheduler{(unsigned) num_threads, MAX_NUM_TASKS, MAX_STACK_SIZE};
 

extern/benchmark_base/include/benchmark_base/fft.h

@@ -11,7 +11,7 @@ namespace fft {
 
 const int SIZE = 8192;
 const int NUM_ITERATIONS = 1000;
-const int NUM_WARMUP_ITERATIONS = 100;
+const int NUM_WARMUP_ITERATIONS = 0;
 
 const int RECURSIVE_CUTOFF = 32;
 typedef std::vector<std::complex<double>> complex_vector;

extern/benchmark_runner/benchmark_runner.h

@@ -108,6 +108,15 @@ class benchmark_runner {
     add_custom_stats_field(MEMORY_POST_RUN);
   }
 
+  void pre_allocate_stats(size_t num = 100000) {
+    times_.reserve(num);
+    memset(times_.data(), 'a', num * sizeof(long));
+    for (auto &iter : custom_stats_) {
+      iter.second.reserve(num);
+      memset(iter.second.data(), 'a', num * sizeof(long));
+    }
+  }
+
   static void read_args(int argc, char **argv, int &num_threads, string &path) {
     if (argc < 3) {
       cout << "Must Specifiy concurrency and output directory! (usage: `benchmark <output_directory> <num_threads>`)"