wip

hybrid/mpithreadpi.cpp

+//
+// mpithreadpi.cpp
+// Neil Gershenfeld 3/3/20
+// calculation of pi by an MPI thread sum
+// pi = 3.14159265358979323846 
+//
+#include <iostream>
+#include <chrono>
+#include <thread>
+#include <vector>
+#include <cstdint>
+#include <mpi.h>
+#define nloop 10
+#define npts 1e9
+unsigned int nthreads = std::thread::hardware_concurrency();
+std::vector<double> results(nthreads);
+void partialsum(int index, int rank) {
+   uint64_t start = npts*index+npts*nthreads*rank+1;
+   uint64_t end = npts*(index+1)+npts*nthreads*rank+1;
+   double result = 0;
+   for (uint64_t i = start; i < end; ++i)
+      result += 0.5/((i-0.75)*(i-0.25));
+   results[index] = result;
+   }
+int main(int argc, char** argv) {
+   double sum,pi,mflops,max;
+   int i,j,rank,nproc;
+   std::thread threads[nthreads];
+   MPI_Init(&argc,&argv);
+   MPI_Comm_rank(MPI_COMM_WORLD,&rank);
+   MPI_Comm_size(MPI_COMM_WORLD,&nproc);
+   if (rank == 0) {
+      for (j = 0; j < nloop; ++j) {
+         MPI_Barrier(MPI_COMM_WORLD);
+         auto tstart = std::chrono::high_resolution_clock::now();        
+         sum = 0.0;
+         for (int i = 0; i < nthreads; ++i)
+            threads[i] = std::thread(partialsum,i,rank);
+         for (int i = 0; i < nthreads; ++i) {
+            threads[i].join();
+            sum += results[i];
+            }
+         MPI_Reduce(&sum,&pi,1,MPI_DOUBLE,MPI_SUM,0,MPI_COMM_WORLD);
+         auto tend = std::chrono::high_resolution_clock::now();        
+         auto dt = std::chrono::duration_cast<std::chrono::microseconds>(tend-tstart).count();
+         auto mflops = npts*nthreads*nproc*5.0/dt;
+         if (mflops > max) max = mflops;
+         std::cout << "npts: " << npts << " nthreads: " << nthreads
+            << " nproc: " << nproc << " pi: " << pi << '\n';
+         std::cout << "time: " << 1e-6*dt << " estimated MFlops: " << mflops << " max: " << max << '\n';
+         }
+      }
+   else {
+      for (j = 0; j < nloop; ++j) {
+         MPI_Barrier(MPI_COMM_WORLD);
+         sum = 0.0;
+         for (int i = 0; i < nthreads; ++i)
+            threads[i] = std::thread(partialsum,i,rank);
+         for (int i = 0; i < nthreads; ++i) {
+            threads[i].join();
+            sum += results[i];
+            }
+         MPI_Reduce(&sum,&pi,1,MPI_DOUBLE,MPI_SUM,0,MPI_COMM_WORLD);
+         }
+      }
+   MPI_Finalize();
+   return 0;
+   }