Add demo for reductions

mpi/demos/reduce.c

@@ -0,0 +1,25 @@
+#include <stdio.h>
+#include <mpi.h>
+
+int main(int argc, char *argv[])
+{
+  int size, rank;
+  MPI_Init(&argc, &argv);
+  MPI_Comm_size(MPI_COMM_WORLD, &size);
+  MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+
+  printf("Hello from rank %d of %d\n", rank, size);
+
+  double data[2] = {1. * rank, -1. * rank};
+  printf("Rank %d data is                     (%.1f, %.1f)\n", rank, data[0], data[1]);
+
+  double sum_of_data[2];
+  MPI_Reduce(data, sum_of_data, 2, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+//  MPI_Allreduce(data, sum_of_data, 2, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
+//  MPI_Allreduce(MPI_IN_PLACE, data, 2, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
+
+  printf("Rank %d data after reduce is        (%.1f, %.1f)\n", rank, data[0], data[1]);
+  printf("Rank %d sum_of_data after reduce is (%.1f, %.1f)\n", rank, sum_of_data[0], sum_of_data[1]);
+
+  MPI_Finalize();
+}

mpi/docs/05-collective-reductions.md

@@ -45,6 +45,10 @@ lang:   en
 MPI_Reduce(`sendbuf`{.input}, `recvbuf`{.output}, `count`{.input}, `datatype`{.input}, `op`{.input}, `root`{.input}, `comm`{.input})
   : Combines values to the root process from all processes of the group
 
+<p>
+- Demo: `reduce.c`
+
+
 # Global reduction
 
 MPI_Allreduce(`sendbuf`{.input}, `recvbuf`{.output}, `count`{.input}, `datatype`{.input}, `op`{.input}, `comm`{.input})
@@ -53,6 +57,9 @@ MPI_Allreduce(`sendbuf`{.input}, `recvbuf`{.output}, `count`{.input}, `datatype`
 <p>
 - Similar to `MPI_Reduce` + `MPI_Bcast` but more efficient
 
+<p>
+- Demo: `reduce.c`
+
 
 # Allreduce example: parallel dot product
 
@@ -93,6 +100,9 @@ call mpi_allreduce(rloc, r, 1, MPI_REAL, &
 `call mpi_allreduce(a, a, n, mpi_real,...`
     - One should employ `MPI_IN_PLACE` for this purpose
 - Replacing the send buffer with the `MPI_IN_PLACE` clause informs MPI that the buffers are the same
+<p>
+- Demo: `reduce.c`
+
 
 # Summary