Removed ParallelFor that depended on a static class function,

which is replaced by a new variadic templated ParallelFor. After testing the
performance of each, I was able to conclude that they perform identically, so
the earlier version is no longer needed.

Author: Chris Sullivan

example.cc

@@ -4,8 +4,6 @@
 
 #include <iostream>
 
-static double* a;
-static double* b;
 
 using namespace std;
 
@@ -19,30 +17,16 @@ int main () {
         ThreadPool pool(8);
 
         int N = 1e9;
-        a = (double*)calloc(N,sizeof(double));
-        b = (double*)calloc(N,sizeof(double));
+        auto a = (double*)calloc(N,sizeof(double));
+        auto b = (double*)calloc(N,sizeof(double));
         for (int i=0; i<N; i++) { b[i] = i; }
 
-        SERIAL_OPERATION(static_scale, a[i]=4*b[i]);
-        {
-                // This is a cold start loop, it is not timed
-                pool.ParallelFor<static_scale>(0,N);
-        }
+
+        pool.ParallelFor(0,N,raw_scale,a,b); // cold start
 
         int ntrials = 10;
         double tperformance = 0.0;
         for (int i=0; i<ntrials; i++)
-        {
-                Timer timer([&](int elapsed){
-                                cout << "Trial " << i << ": "<< elapsed*1e-6 << " ms\n";
-                                tperformance+=elapsed;
-                        });
-                pool.ParallelFor<static_scale>(0,N);
-        }
-        cout << "Average: " << tperformance*1e-6 / ntrials << " ms\n\n";
-
-        tperformance = 0.0;
-        for (int i=0; i<ntrials; i++)
         {
                 Timer timer([&](int elapsed){
                                 cout << "Trial " << i << ": "<< elapsed*1e-6 << " ms\n";
@@ -53,6 +37,6 @@ int main () {
         cout << "Average: " << tperformance*1e-6 / ntrials << " ms\n\n";
 
 
-        cin.get();
+
         return 0;
 }

include/ThreadPool.hh

@@ -23,24 +23,6 @@ public:
         void Worker();
         void JoinAll();
 
-        template <typename ClassFunction>
-        void ParallelFor(uint32_t begin, uint32_t end) {
-
-                int chunk = (end - begin) / m_nthreads;
-                for (int i = 0; i < m_nthreads; ++i) {
-                        m_promises.emplace_back();
-                        int mypromise = m_promises.size() - 1;
-                        Enqueue([=]{
-                                        int threadstart = begin + i*chunk;
-                                        int threadstop = (i == m_nthreads - 1) ? end : threadstart + chunk;
-                                        for (int it = threadstart; it < threadstop; ++it) {
-                                                ClassFunction::SerialFunction(it);
-                                        }
-                                        m_promises[mypromise].set_value();
-                                });
-                }
-                Finish();
-        }
         template <typename T, typename... Params>
         void ParallelFor(uint32_t begin, uint32_t end, T SerialFunction, Params&&... params) {
 
@@ -79,6 +61,3 @@ private:
         atomic<int> m_nrunning;
 
 };
-
-// Macro to define a static class function which can be called via ThreadPool::ParallelFor<T>
-#define SERIAL_OPERATION(name, function_kernal) class name { public: static void SerialFunction(const int& i) { function_kernal; } };