Skip to content

Commit

Permalink
Attempt to fix cuda CI errors
Browse files Browse the repository at this point in the history
  • Loading branch information
e10harvey committed Oct 24, 2023
1 parent 1bae37a commit dd0ad97
Showing 1 changed file with 3 additions and 3 deletions.
6 changes: 3 additions & 3 deletions blas/impl/KokkosBlas2_gemv_impl.hpp
Original file line number Diff line number Diff line change
Expand Up @@ -636,7 +636,7 @@ struct TwoLevelTransposeGEMV {
// Two-level parallel version of GEMV.
template <class ExecutionSpace, class AViewType, class XViewType,
class YViewType, class IndexType = typename AViewType::size_type>
void twoLevelGemv(ExecutionSpace& space, const char trans[],
void twoLevelGemv(const ExecutionSpace& space, const char trans[],
typename AViewType::const_value_type& alpha,
const AViewType& A, const XViewType& x,
typename YViewType::const_value_type& beta,
Expand Down Expand Up @@ -704,14 +704,14 @@ void twoLevelGemv(ExecutionSpace& space, const char trans[],
YViewType, IndexType>;
functor_type functor(alpha, A, x, beta, y);
tagged_policy team;
if (isLayoutLeft) {
if constexpr (isLayoutLeft) {
using AccumScalar = typename std::conditional<
std::is_same<y_value_type, Kokkos::Experimental::half_t>::value ||
std::is_same<y_value_type, Kokkos::Experimental::bhalf_t>::value,
float, y_value_type>::type;
size_t sharedPerTeam = 32 * sizeof(AccumScalar);
IndexType numTeams = (A.extent(0) + 31) / 32;
tagged_policy temp(1, 1);
tagged_policy temp(space, 1, 1);
temp.set_scratch_size(0, Kokkos::PerTeam(sharedPerTeam));
int teamSize =
temp.team_size_recommended(functor, Kokkos::ParallelForTag());
Expand Down

0 comments on commit dd0ad97

Please sign in to comment.