Small reduction clean-ups

Removed unnecessary copies in custom_reduce_over_group

Sequential reduction now casts before calling operator (makes behavior explicit rather than implicit)

Author: ndgrigorian

dpctl/tensor/libtensor/include/kernels/reductions.hpp

@@ -97,7 +97,9 @@ struct SequentialReduction
             const py::ssize_t inp_offset =
                 inp_iter_offset + inp_reduction_offset;
 
-            red_val = reduction_op_(red_val, inp_[inp_offset]);
+            using dpctl::tensor::type_utils::convert_impl;
+            outT val = convert_impl<outT, argT>(inp_[inp_offset]);
+            red_val = reduction_op_(red_val, val);
         }
 
         out_[out_iter_offset] = red_val;

dpctl/tensor/libtensor/include/utils/sycl_utils.hpp

@@ -132,10 +132,10 @@ size_t choose_workgroup_size(const size_t nelems,
 }
 
 template <typename T, typename GroupT, typename LocAccT, typename OpT>
-T custom_reduce_over_group(GroupT wg,
+T custom_reduce_over_group(const GroupT &wg,
                            LocAccT local_mem_acc,
-                           T local_val,
-                           OpT op)
+                           const T &local_val,
+                           const OpT &op)
 {
     size_t wgs = wg.get_local_linear_range();
     local_mem_acc[wg.get_local_linear_id()] = local_val;