Add the checkpoint restart feature

project/src/Main.cpp

@@ -74,13 +74,26 @@ void exchangeHalos(int rank, int nsize, Problem &p) {
 
 
 int main(int argc, char **argv) {
-  MPI_Init(nullptr, nullptr);
+  // Process arguments for checkpoint facility
+  const char *pauseFile = nullptr;
+  const char *restartFile = nullptr;
+
+  for (int i = 1; i < argc; ++i) {
+    if (!strncmp(argv[i], "-p=", 3)) {
+      pauseFile = argv[i] + 3;
+    } else if (!strncmp(argv[i], "-r=", 3)) {
+      restartFile = argv[i] + 3;
+    }
+  }
 
+  // Initialize the MPI machinery
+  MPI_Init(nullptr, nullptr);
   int nsize, rank;
 
   MPI_Comm_size(DEFAULT_COMM, &nsize);
   MPI_Comm_rank(DEFAULT_COMM, &rank);
 
+  // Distribute the problem parameters
   GlobalProblemConfig config;
   MPI_Datatype GlobalProblemConfig_t;
   MPI_Type_create_struct(5, GlobalProblemConfig::block_lengths(), GlobalProblemConfig::displacements(),
@@ -92,26 +105,50 @@ int main(int argc, char **argv) {
   }
   MPI_Bcast(&config, 1, GlobalProblemConfig_t, 0, DEFAULT_COMM);
 
-  printf("%d: %d, %d\n", rank, config.xdim, config.ydim);
-
+  // Set up the local problem
   unsigned localXdim = config.xdim;
   unsigned localYdim = config.ydim / nsize;
 
   assert(config.ydim % nsize == 0);
 
-  int upNeighbour = rank + 1;
-  int dnNeighbour = rank - 1;
-
+  // Hardcoded parameters
   double gateVal = 100;
   unsigned nGates = 2;
 
   Problem p(nsize, rank, localXdim, localYdim, nGates, nGates, gateVal);
+  if (restartFile) {
+    printf("%d: Restarting from file %s\n", rank, restartFile);
+    MPI_File fh;
+    int dataSize = p.getField().numCols() * localYdim;
+    double *data = p.getField().firstRow().raw();
+
+    MPI_File_open(DEFAULT_COMM, restartFile, MPI_MODE_RDONLY, MPI_INFO_NULL, &fh);
+    MPI_File_set_view(fh, rank * dataSize * sizeof(double),
+                      MPI_DOUBLE, MPI_DOUBLE, "native", MPI_INFO_NULL);
+    MPI_File_read(fh, data, dataSize, MPI_DOUBLE, MPI_STATUS_IGNORE);
+    MPI_File_close(&fh);
+  }
 
+  // Simulate
   for (int i = 0; i < config.maxiter; ++i) {
     exchangeHalos(rank, nsize, p);
     p.step(config.cx, config.cy);
   }
 
+  if (pauseFile) {
+    // Write the problem out, for later restart
+    MPI_File fh;
+    int dataSize = p.getField().numCols() * localYdim;
+    double *data = p.getField().firstRow().raw();
+
+    MPI_File_open(DEFAULT_COMM, pauseFile, MPI_MODE_WRONLY | MPI_MODE_CREATE, MPI_INFO_NULL, &fh);
+    MPI_File_set_view(fh, rank * dataSize * sizeof(double),
+                      MPI_DOUBLE, MPI_DOUBLE, "native", MPI_INFO_NULL);
+    MPI_File_write(fh, data, dataSize, MPI_DOUBLE, MPI_STATUS_IGNORE);
+    MPI_File_close(&fh);
+
+    printf("%d: Problem written out to file %s\n", rank, pauseFile);
+  }
 
   // Compute the global min and max
   double localMin, localMax;

project/src/Problem.cpp

@@ -35,10 +35,8 @@ void Problem::addGates(Field &f, unsigned xGates, unsigned yGates, double value)
     unsigned gateStart = gateCentre - 0.5 * GATE_WIDTH * nFields * ydim;
     unsigned gateEnd = gateCentre + 0.5 * GATE_WIDTH * nFields * ydim;
 
-    if (gateStart >= starty && gateStart < endy) {
-      for (unsigned j = gateStart; j < std::min(gateEnd, endy); j++) {
-        f[j - starty + 1][xdim + 1] = value;
-      }
+    for (unsigned j = std::max(gateStart, starty); j < std::min(gateEnd, endy); j++) {
+      f[j - starty + 1][xdim + 1] = value;
     }
   }
 }

project/src/Problem.h

@@ -15,6 +15,10 @@ class Problem {
   Field *currentf;
   Field *nextf;
 
+  unsigned xgates;
+  unsigned ygates;
+  double gateVal;
+
   void initField(Field &f);
   void addGates(Field &f, unsigned xgates, unsigned ygates, double value);
 
@@ -28,13 +32,12 @@ public:
     f0(xdim, ydim),
     f1(xdim, ydim),
     currentf(&f0),
-    nextf(&f1) {
+    nextf(&f1), xgates(xgates), ygates(ygates), gateVal(gateVal) {
 
     initField(f0);
     initField(f1);
     addGates(f0, xgates, ygates, gateVal);
     addGates(f1, xgates, ygates, gateVal);
-
   }
 
   void step(double cx, double cy);