Fix runtime error with complex double and Intel compiler 18.2.199

kokkos · Nov 5, 2018 · 8dcd972 · 8dcd972
1 parent ec3142e
commit 8dcd972
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 7 deletions.
diff --git a/src/batched/KokkosBatched_InverseLU_Serial_Impl.hpp b/src/batched/KokkosBatched_InverseLU_Serial_Impl.hpp
@@ -88,7 +88,7 @@ namespace KokkosBatched {
         auto B = Kokkos::View<ScalarType**, Kokkos::LayoutLeft, typename WViewType::memory_space, Kokkos::MemoryTraits<Kokkos::Unmanaged> >(W.data(), A.extent(0), A.extent(1));
 
         const ScalarType one(1.0);
-        
+
 #if defined(KOKKOS_ENABLE_PRAGMA_UNROLL)
 #pragma unroll
 #endif

diff --git a/src/batched/KokkosBatched_SolveLU_Serial_Impl.hpp b/src/batched/KokkosBatched_SolveLU_Serial_Impl.hpp
@@ -40,7 +40,7 @@ namespace KokkosBatched {
         SerialTrsm<Side::Left,Uplo::Lower,Trans::NoTranspose,Diag::Unit,Algo::Trsm::Unblocked>::invoke(one, A, B);
         //Second, compute X by solving the system U*X = Y for X
         SerialTrsm<Side::Left,Uplo::Upper,Trans::NoTranspose,Diag::NonUnit,Algo::Trsm::Unblocked>::invoke(one, A, B);
-        
+
         return 0;
     }
 

diff --git a/src/batched/KokkosBatched_Trsm_Serial_Internal.hpp b/src/batched/KokkosBatched_Trsm_Serial_Internal.hpp
@@ -180,8 +180,8 @@ namespace KokkosBatched {
           const int iend = p, jend = n;
 
           const ValueType *__restrict__ a01 = A+p*as1;
-          /**/  ValueType *__restrict__ b1t = B+p*bs0;
-            
+                ValueType *__restrict__ b1t = B+p*bs0;
+
           if (!use_unit_diag) {
             const ValueType alpha11 = A[p*as0+p*as1];
 
@@ -191,13 +191,16 @@ namespace KokkosBatched {
             for (int j=0;j<n;++j)
               b1t[j*bs1] = b1t[j*bs1] / alpha11;
           }
-          for (int i=0;i<iend;++i)
+
+          if (p>0){
+            for (int i=0;i<iend;++i)
 
 #if defined(KOKKOS_ENABLE_PRAGMA_UNROLL)
 #pragma unroll
 #endif
-            for (int j=0;j<jend;++j)
-              B0[i*bs0+j*bs1] -= a01[i*as0] * b1t[j*bs1];
+              for (int j=0;j<jend;++j)
+                B0[i*bs0+j*bs1] -= a01[i*as0] * b1t[j*bs1];
+          }
         }
       }
       return 0;