Avoid the SIMD code branch if the batched size is not a multiple of t…

…he vector length
kokkos · Sep 27, 2022 · 3562c23 · 3562c23
1 parent caf47dc
commit 3562c23
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/batched/sparse/impl/KokkosBatched_Spmv_TeamVector_Impl.hpp b/batched/sparse/impl/KokkosBatched_Spmv_TeamVector_Impl.hpp
@@ -219,7 +219,7 @@ KOKKOS_INLINE_FUNCTION int TeamVectorSpmvInternal::invoke(
     const OrdinalType ys1) {
 #if !defined(__CUDA_ARCH__) && !defined(__HIP_DEVICE_COMPILE__)
   if (member.team_size() == 1) {
-    if (N_team > 1 && valuess0 == 1) {
+    if (N_team > 1 && valuess0 == 1 && valuess1 % N_team == 0) {
       /*
         Left layout as valuess0 = 1 and non-zero vector length given at
         compilation time Here we use the SIMD data type which is using Intel