Changeset 1232 for XIOS/dev/XIOS_DEV_CMIP6/src/server_distribution_description.cpp

Timestamp:

07/31/17 17:59:25 (7 years ago)

Author:

mhnguyen

Message:

Fixing the blocking problem where there are more servers than the number of grid band distribution

+) Correct this problem not only for writing but also for reading
+) Allow "zero-size" domain, axis (i.e: domain, axis with ni = 0, and/or nj=0)

Test
+) On Curie
+) Work in both cases: Read and Write data

File:

• XIOS/dev/XIOS_DEV_CMIP6/src/server_distribution_description.cpp (modified) (9 diffs)

XIOS/dev/XIOS_DEV_CMIP6/src/server_distribution_description.cpp

@@ -100 +100 @@
   \param [in] positionDimensionDistributed dimension of server on which we make the cut.
 */
-void CServerDistributionDescription::computeServerGlobalIndexInRange(const std::pair<size_t, size_t>& indexBeginEnd,
+std::vector<int> CServerDistributionDescription::computeServerGlobalIndexInRange(const std::pair<size_t, size_t>& indexBeginEnd,
                                                                      int positionDimensionDistributed)
 {
+  int nBand  = 0;
   switch (serverType_) {
     case BAND_DISTRIBUTION:
-      computeBandDistribution(nServer_, positionDimensionDistributed);
+      nBand = computeBandDistribution(nServer_, positionDimensionDistributed);
       break;
     default:
@@ -122 +123 @@
   std::vector<int> currentIndex(dim);
 
-  for (int idxServer = 0; idxServer < nServer_; ++idxServer)
+  for (int idxServer = 0; idxServer < nBand; ++idxServer)
   {
     size_t ssize = 1, idx = 0;
@@ -161 +162 @@
     }
   }
+
+    // List of servers without distribution (cause total number of server is greater than number of bands, for example)
+  std::vector<int> zeroIndexServer(nServer_-nBand); 
+  for (int idxServer = nBand; idxServer < nServer_; ++idxServer)
+    zeroIndexServer[idxServer-nBand] = idxServer;
+
+  return zeroIndexServer;
 }
 
@@ -172 +180 @@
   \param [in] positionDimensionDistributed dimension of server on which we make the cut.
 */
-void CServerDistributionDescription::computeServerGlobalByElement(std::vector<boost::unordered_map<size_t,std::vector<int> > >& indexServerOnElement,
-                                                                  int clientRank,
-                                                                  int clientSize,
-                                                                  const CArray<int,1>& axisDomainOrder,
-                                                                  int positionDimensionDistributed)
-{
+std::vector<int> CServerDistributionDescription::computeServerGlobalByElement(std::vector<boost::unordered_map<size_t,std::vector<int> > >& indexServerOnElement,
+                                                                              int clientRank,
+                                                                              int clientSize,
+                                                                              const CArray<int,1>& axisDomainOrder,
+                                                                              int positionDimensionDistributed)
+{
+  int nBand  = 0;
   switch (serverType_) {
     case BAND_DISTRIBUTION:
-      computeBandDistribution(nServer_, positionDimensionDistributed);
+      nBand = computeBandDistribution(nServer_, positionDimensionDistributed);
       break;
     default:
@@ -197 +206 @@
   }
 
-  for (int idxServer = 0; idxServer < nServer_; ++idxServer)
+  for (int idxServer = 0; idxServer < nBand; ++idxServer)
   {
     std::vector<int> elementDimension(4);
@@ -247 +256 @@
     }
   }
+
+  // List of servers without distribution (cause total number of server is greater than number of bands, for example)
+  std::vector<int> zeroIndexServer(nServer_-nBand); 
+  for (int idxServer = nBand; idxServer < nServer_; ++idxServer)
+    zeroIndexServer[idxServer-nBand] = idxServer;
+
+  return zeroIndexServer;
 }
 
@@ -297 +313 @@
   \param [in] nServer number of server
 */
-void CServerDistributionDescription::computeBandDistribution(int nServer, int positionDimensionDistributed)
+int CServerDistributionDescription::computeBandDistribution(int nServer, int positionDimensionDistributed)
 {
   int dim = nGlobal_.size();
@@ -324 +340 @@
   int positionDistributed = (1<dim) ? positionDimensionDistributed_ : 0;
   nGlobTemp = nGlobal_[positionDistributed];
-
-  for (int i = 0; i < nServer; ++i)
+  int nbBand = std::min(nGlobTemp, nServer);
+
+  for (int i = 0; i < nbBand; ++i)
   {
     if (0 < i) njRangeBegin[i] = njRangeEnd[i-1];
-    njRangeSize = nGlobTemp / nServer;
-    if (i < nGlobTemp%nServer) ++njRangeSize;
+    njRangeSize = nGlobTemp / nbBand;
+    if (i < nGlobTemp%nbBand) ++njRangeSize;
     njRangeEnd[i] = njRangeSize + njRangeBegin[i];
   }
-  njRangeEnd[nServer-1] = nGlobTemp;
+  njRangeEnd[nbBand-1] = nGlobTemp;
+
+  for (int i = nbBand; i < nServer; ++i)
+  {
+    njRangeBegin[i] = njRangeEnd[i] = 0;
+  }
 
   for (int i = 0; i < nServer; ++i)
@@ -358 +380 @@
     }
   }
+
+  return nbBand;
 }