Add simple example using smpi_execute_* calls inside MPI code to simulate computation.

[simgrid.git] / examples / smpi / simple-execute / simple-execute.c

diff --git a/examples/smpi/simple-execute/simple-execute.c b/examples/smpi/simple-execute/simple-execute.c
new file mode 100644 (file)
index 0000000..1b5fb81
--- /dev/null
+++ b/examples/smpi/simple-execute/simple-execute.c
@@ -0,0 +1,70 @@
+/* A simple example ping-pong program to test MPI_Send and MPI_Recv */
+
+/* Copyright (c) 2009-2021. The SimGrid Team.
+ * All rights reserved.                                                     */
+
+/* This program is free software; you can redistribute it and/or modify it
+ * under the terms of the license (GNU LGPL) which comes with this package. */
+
+#include <stdio.h>
+#include <mpi.h>
+
+/* This test performs a simple pingpong between 2 processes
+   Each process calls smpi_execute or smpi_execute_flops */
+
+int main(int argc, char *argv[])
+{
+  const int tag1 = 42;
+  const int tag2 = 43; /* Message tag */
+  int size;
+  int rank;
+  int msg = 99;
+  MPI_Status status;
+  int err = MPI_Init(&argc, &argv); /* Initialize MPI */
+
+  if (err != MPI_SUCCESS) {
+    printf("MPI initialization failed!\n");
+    exit(1);
+  }
+  MPI_Comm_size(MPI_COMM_WORLD, &size);   /* Get nr of tasks */
+  MPI_Comm_rank(MPI_COMM_WORLD, &rank);   /* Get id of this process */
+  if (size != 2) {
+    printf("run this program with exactly 2 processes (-np 2)\n");
+    MPI_Finalize();
+    exit(0);
+  }
+  if (0 == rank) {
+    printf("\n    *** Ping-pong test (MPI_Send/MPI_Recv) ***\n\n");
+  }
+
+  /* start ping-pong tests between several pairs */
+  if ( rank == 0) {
+    int dst = 1;
+    printf("[%d] About to send 1st message '%d' to process [%d]\n", rank, msg, dst);
+    MPI_Send(&msg, 1, MPI_INT, dst, tag1, MPI_COMM_WORLD);
+
+    /* Inject five seconds of fake computation time */
+    /* We are in a public file, not internal to simgrid, so _benched flavour is preferred, as it protects against accidental skip */
+    /* smpi_execute_benched here is mostly equivalent to sleep, which is intercepted by smpi and turned into smpi_sleep */
+    /* Difference with sleep is only for energy consumption */
+    smpi_execute_benched(5.0);
+
+    MPI_Recv(&msg, 1, MPI_INT, dst, tag2, MPI_COMM_WORLD, &status);     /* Receive a message */
+    printf("[%d] Received reply message '%d' from process [%d]\n", rank, msg, dst);
+  } else {
+    int src = 0;
+    MPI_Recv(&msg, 1, MPI_INT, src, tag1, MPI_COMM_WORLD, &status);     /* Receive a message */
+    printf("[%d] Received 1st message '%d' from process [%d]\n", rank, msg, src);
+    msg++;
+
+    /* Inject 762.96 Mflops of computation time - Host Jupiter is 76.296Mf per second, so this should amount to 10s */
+    /* We are in a public file, not internal to simgrid, so _benched flavour is preferred, as it protects against accidental skip */
+    smpi_execute_flops_benched(762960000);
+
+    printf("[%d] After a nap, increment message's value to  '%d'\n", rank, msg);
+    printf("[%d] About to send back message '%d' to process [%d]\n", rank, msg, src);
+    MPI_Send(&msg, 1, MPI_INT, src, tag2, MPI_COMM_WORLD);
+  }
+  MPI_Finalize();
+  return 0;
+}