Vector implementation for floor_divide() function has been changed.

Author: npolina4

dpctl/tensor/libtensor/include/kernels/elementwise_functions/floor_divide.hpp

@@ -81,24 +81,24 @@ struct FloorDivideFunctor
         {
             return tmp;
         }
-        if constexpr (std::is_integral_v<typename decltype(tmp)::element_type>)
-        {
+        else if constexpr (std::is_integral_v<typename decltype(
+                               tmp)::element_type>) {
             using dpctl::tensor::type_utils::vec_cast;
             return vec_cast<resT, typename decltype(tmp)::element_type, vec_sz>(
                 tmp);
         }
         else {
-            sycl::vec<resT, vec_sz> res;
-            for (int i = 0; i < vec_sz; i++) {
-                auto tmp2 = sycl::floor(tmp[i]);
-                if constexpr (std::is_same_v<resT, decltype(tmp2)>) {
-                    res[i] = tmp2;
-                }
-                else {
-                    res[i] = static_cast<resT>(tmp2);
-                }
+            sycl::vec<resT, vec_sz> res = sycl::floor(tmp);
+            if constexpr (std::is_same_v<resT,
+                                         typename decltype(res)::element_type>)
+            {
+                return res;
+            }
+            else {
+                using dpctl::tensor::type_utils::vec_cast;
+                return vec_cast<resT, typename decltype(res)::element_type,
+                                vec_sz>(res);
             }
-            return res;
         }
     }
 };

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

@@ -64,7 +64,7 @@ template <class V, class U> sink_t __appender(V &lhs, U &&rhs)
 }
 
 template <typename T, typename A, typename... Vs>
-std::vector<T, A> concat(std::vector<T, A> lhs, Vs &&... vs)
+std::vector<T, A> concat(std::vector<T, A> lhs, Vs &&...vs)
 {
     std::size_t s = lhs.size();
     {
@@ -87,7 +87,7 @@ template <typename indT, typename... Vs>
 std::tuple<indT *, size_t, sycl::event>
 device_allocate_and_pack(sycl::queue q,
                          std::vector<sycl::event> &host_task_events,
-                         Vs &&... vs)
+                         Vs &&...vs)
 {
 
     // memory transfer optimization, use USM-host for temporary speeds up