Merge pull request #1678 from lcheng9/develop

pcarruscag · web-flow · commit eed1e67174d6 · 2022-07-21T20:31:54.000-07:00
Fix #1667 Potential crash for the OpenMP multi-thread running.
diff --git a/Common/src/linear_algebra/CSysMatrix.cpp b/Common/src/linear_algebra/CSysMatrix.cpp
@@ -190,6 +190,7 @@ void CSysMatrix<ScalarType>::Initialize(unsigned long npoint, unsigned long npoi
 
   /*--- This is akin to the row_ptr. ---*/
   omp_partitions = new unsigned long [omp_num_parts+1];
+  for (unsigned long i = 0; i <= omp_num_parts; ++i) omp_partitions[i] = nPointDomain;
 
   /*--- Work estimate based on non-zeros to produce balanced partitions. ---*/
 
@@ -202,7 +203,15 @@ void CSysMatrix<ScalarType>::Initialize(unsigned long npoint, unsigned long npoi
     if (row_ptr_prec[iPoint] >= part*nnz_per_part)
       omp_partitions[part++] = iPoint;
   }
-  omp_partitions[omp_num_parts] = nPointDomain;
+
+  for (unsigned long thread = 0; thread < omp_num_parts; ++thread) {
+    const auto begin = omp_partitions[thread];
+    const auto end = omp_partitions[thread + 1];
+    if (begin == end) {
+      cout << "WARNING: Redundant thread has been detected. Performance could be impacted due to low number of nodes per thread." << endl;
+      break;
+    }
+  }
 
   /*--- Generate MKL Kernels ---*/
 
@@ -700,6 +709,7 @@ void CSysMatrix<ScalarType>::BuildILUPreconditioner() {
   {
     const auto begin = omp_partitions[thread];
     const auto end = omp_partitions[thread+1];
+    if (begin == end) continue;
 
     /*--- Each thread will work on the submatrix defined from row/col "begin"
      *    to row/col "end-1" (i.e. the range [begin,end[). Which is exactly
@@ -777,6 +787,7 @@ void CSysMatrix<ScalarType>::ComputeILUPreconditioner(const CSysVector<ScalarTyp
   {
     const auto begin = omp_partitions[thread];
     const auto end = omp_partitions[thread+1];
+    if (begin == end) continue;
 
     ScalarType aux_vec[MAXNVAR];
 
@@ -839,6 +850,7 @@ void CSysMatrix<ScalarType>::ComputeLU_SGSPreconditioner(const CSysVector<Scalar
   {
     const auto begin = omp_partitions[thread];
     const auto end = omp_partitions[thread+1];
+    if (begin == end) continue;
 
     /*--- Each thread will work on the submatrix defined from row/col "begin"
      *    to row/col "end-1", except the last thread that also considers halos.
@@ -869,6 +881,8 @@ void CSysMatrix<ScalarType>::ComputeLU_SGSPreconditioner(const CSysVector<Scalar
   {
     const auto begin = omp_partitions[thread];
     const auto row_end = omp_partitions[thread+1];
+    if (begin == row_end) continue;
+
     /*--- On the last thread partition the upper
      *    product should consider halo columns. ---*/
     const auto col_end = (row_end==nPointDomain)? nPoint : row_end;
