Evaluate more terms

Now the calculation runs for about 2 seconds. Previously it was a
fraction of a second, so GFLOPS between runs varied wildly.

mpi_pi_cpu/mpi_pi_cpu.c

@@ -10,17 +10,19 @@
 #include <sys/time.h>
 
 // number of terms evaluated by each process
-#define NPTS 1000000
+#define NPTS 250000000
 
 void main(int argc, char** argv) {
     // variables used by all ranks
-    int rank, i, istart, iend;
+    int rank;
+    long unsigned int i, istart, iend;
     double sum, pi;
 
     // variables used only by rank 0
     unsigned long int start_time, end_time;
     struct timeval start, end;
     int nproc;
+    unsigned long int total_terms;
     double mflops;
 
     MPI_Init(&argc, &argv);
@@ -28,11 +30,13 @@ void main(int argc, char** argv) {
 
     if (rank == 0) {
         MPI_Comm_size(MPI_COMM_WORLD, &nproc);
+        total_terms = nproc;
+        total_terms *= NPTS;
         gettimeofday(&start, NULL);
     }
 
-    istart = NPTS * rank + 1;
-    iend = NPTS * (rank + 1) + 1;
+    istart = (long unsigned int)NPTS * rank + 1;
+    iend = istart + NPTS;
     sum = 0.0;
     for (i = istart; i <= iend; ++i) {
         sum += 0.5 / ((i - 0.75) * (i - 0.25));
@@ -44,8 +48,8 @@ void main(int argc, char** argv) {
         start_time = start.tv_sec * 1e6 + start.tv_usec;
         end_time = end.tv_sec * 1e6 + end.tv_usec;
         mflops = (double)nproc * NPTS * 5.0 / (end_time - start_time);
-        printf("processes = %d, terms per process = %ld, total terms = %ld\n", nproc, NPTS, nproc * NPTS);
-        printf("time = %f, estimated MFlops = %f\n", (end_time - start_time) / 1.0e6, mflops);
+        printf("processes = %d, terms per process = %ldM, total terms = %ldM\n", nproc, NPTS / 1000000, total_terms / 1000000);
+        printf("time = %fs, estimated GFlops = %f\n", (end_time - start_time) / 1.0e6, mflops / 1000);
         printf("pi ~ %f\n", pi);
     }