Optimizations for Python bindings and sum_of_squares (#102)

This PR has two small optimizations:
* Passes `indices` and `ids` by reference rather than by value in `module.cc`.  This seems to give a rather substantial speedup for Python ivf query times.
* Removes explicit casts to `float` in `sum_of_squares`.  This gives a 30% speedup (tested on EC2 r6id.2xlarge).

Author: lums658

apis/python/src/tiledb/vector_search/module.cc

@@ -109,8 +109,8 @@ static void declare_qv_query_heap_infinite_ram(py::module& m, const std::string&
       [](const ColMajorMatrix<T>& parts,
          const ColMajorMatrix<float>& centroids,
          const ColMajorMatrix<float>& query_vectors,
-         std::vector<Id_Type> indices,
-         std::vector<Id_Type> ids,
+         std::vector<Id_Type>& indices,
+         std::vector<Id_Type>& ids,
          size_t nprobe,
          size_t k_nn,
          bool nth,
@@ -137,7 +137,7 @@ static void declare_qv_query_heap_finite_ram(py::module& m, const std::string& s
          const std::string& parts_uri,
          const ColMajorMatrix<float>& centroids,
          const ColMajorMatrix<float>& query_vectors,
-         std::vector<Id_Type> indices,
+         std::vector<Id_Type>& indices,
          const std::string& ids_uri,
          size_t nprobe,
          size_t k_nn,
@@ -167,8 +167,8 @@ static void declare_nuv_query_heap_infinite_ram(py::module& m, const std::string
       [](const ColMajorMatrix<T>& parts,
          const ColMajorMatrix<float>& centroids,
          const ColMajorMatrix<float>& query_vectors,
-         std::vector<Id_Type> indices,
-         std::vector<Id_Type> ids,
+         std::vector<Id_Type>& indices,
+         std::vector<Id_Type>& ids,
          size_t nprobe,
          size_t k_nn,
          bool nth,
@@ -195,7 +195,7 @@ static void declare_nuv_query_heap_finite_ram(py::module& m, const std::string&
          const std::string& parts_uri,
          const ColMajorMatrix<float>& centroids,
          const ColMajorMatrix<float>& query_vectors,
-         std::vector<Id_Type> indices,
+         std::vector<Id_Type>& indices,
          const std::string& ids_uri,
          size_t nprobe,
          size_t k_nn,

src/include/defs.h

@@ -57,17 +57,41 @@
  * @param b
  * @return
  */
+#if 0
 template <class V, class U>
 inline auto sum_of_squares(V const& a, U const& b) {
   float sum{0.0};
   size_t size_a = size(a);
+
+  if constexpr (std::is_same_v<decltype(a[0]),decltype(b[0])>) {
+    for (size_t i = 0; i < size_a; ++i) {
+      float diff = a[i]- b[i];
+      sum += diff * diff;
+    }
+  } else {
+    for (size_t i = 0; i < size_a; ++i) {
+      float diff = ((float)a[i]) - ((float)b[i]);
+      sum += diff * diff;
+    }
+  }
+  return sum;
+}
+#else
+template <class V, class U>
+inline auto sum_of_squares(V const& a, U const& b) {
+  float sum{0.0};
+  size_t size_a = size(a);
+
   for (size_t i = 0; i < size_a; ++i) {
-    float diff = ((float)a[i]) - ((float)b[i]);
+    // float diff = ((float)a[i]) - ((float)b[i]);
+    float diff = a[i] - b[i];
     sum += diff * diff;
   }
   return sum;
 }
 
+#endif
+
 /**
  * @brief Compute L2 distance between two vectors.
  * @tparam V