Compilation optimizations - batch 1

cpp/include/cugraph/arithmetic_variant_types.hpp

@@ -0,0 +1,249 @@
+/*
+ * Copyright (c) 2025, NVIDIA CORPORATION.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#pragma once
+
+#include <cugraph/edge_property.hpp>
+#include <cugraph/utilities/error.hpp>
+
+#include <raft/core/device_span.hpp>
+
+#include <rmm/device_uvector.hpp>
+
+#include <variant>
+
+namespace cugraph {
+
+using arithmetic_device_uvector_t    = std::variant<std::monostate,
+                                                    rmm::device_uvector<float>,
+                                                    rmm::device_uvector<double>,
+                                                    rmm::device_uvector<int32_t>,
+                                                    rmm::device_uvector<int64_t>,
+                                                    rmm::device_uvector<size_t>>;
+using arithmetic_device_span_t       = std::variant<std::monostate,
+                                                    raft::device_span<float>,
+                                                    raft::device_span<double>,
+                                                    raft::device_span<int32_t>,
+                                                    raft::device_span<int64_t>,
+                                                    raft::device_span<size_t>>;
+using const_arithmetic_device_span_t = std::variant<std::monostate,
+                                                    raft::device_span<float const>,
+                                                    raft::device_span<double const>,
+                                                    raft::device_span<int32_t const>,
+                                                    raft::device_span<int64_t const>,
+                                                    raft::device_span<size_t const>>;
+
+template <typename edge_t>
+using edge_arithmetic_property_view_t =
+  std::variant<std::monostate,
+               cugraph::edge_property_view_t<edge_t, float const*>,
+               cugraph::edge_property_view_t<edge_t, double const*>,
+               cugraph::edge_property_view_t<edge_t, int32_t const*>,
+               cugraph::edge_property_view_t<edge_t, int64_t const*>,
+               cugraph::edge_property_view_t<edge_t, size_t const*>>;
+
+template <typename edge_t>
+using edge_arithmetic_property_mutable_view_t =
+  std::variant<std::monostate,
+               cugraph::edge_property_view_t<edge_t, float*>,
+               cugraph::edge_property_view_t<edge_t, double*>,
+               cugraph::edge_property_view_t<edge_t, int32_t*>,
+               cugraph::edge_property_view_t<edge_t, int64_t*>,
+               cugraph::edge_property_view_t<edge_t, size_t*>>;
+
+template <typename func_t>
+auto variant_type_dispatch(arithmetic_device_uvector_t& property, func_t func)
+{
+  if (std::holds_alternative<rmm::device_uvector<float>>(property)) {
+    auto& prop = std::get<rmm::device_uvector<float>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<rmm::device_uvector<double>>(property)) {
+    auto& prop = std::get<rmm::device_uvector<double>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<rmm::device_uvector<int32_t>>(property)) {
+    auto& prop = std::get<rmm::device_uvector<int32_t>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<rmm::device_uvector<int64_t>>(property)) {
+    auto& prop = std::get<rmm::device_uvector<int64_t>>(property);
+    return func(prop);
+  } else {
+    CUGRAPH_EXPECTS(std::holds_alternative<rmm::device_uvector<size_t>>(property),
+                    "unsupported variant type -- shouldn't happen");
+
+    auto& prop = std::get<rmm::device_uvector<size_t>>(property);
+    return func(prop);
+  }
+}
+
+template <typename func_t>
+auto variant_type_dispatch(arithmetic_device_uvector_t const& property, func_t func)
+{
+  if (std::holds_alternative<rmm::device_uvector<float>>(property)) {
+    auto& prop = std::get<rmm::device_uvector<float>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<rmm::device_uvector<double>>(property)) {
+    auto& prop = std::get<rmm::device_uvector<double>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<rmm::device_uvector<int32_t>>(property)) {
+    auto& prop = std::get<rmm::device_uvector<int32_t>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<rmm::device_uvector<int64_t>>(property)) {
+    auto& prop = std::get<rmm::device_uvector<int64_t>>(property);
+    return func(prop);
+  } else {
+    CUGRAPH_EXPECTS(std::holds_alternative<rmm::device_uvector<size_t>>(property),
+                    "unsupported variant type -- shouldn't happen");
+    auto& prop = std::get<rmm::device_uvector<size_t>>(property);
+    return func(prop);
+  }
+}
+
+template <typename func_t>
+auto variant_type_dispatch(arithmetic_device_span_t& property, func_t func)
+{
+  if (std::holds_alternative<raft::device_span<float>>(property)) {
+    auto& prop = std::get<raft::device_span<float>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<raft::device_span<double>>(property)) {
+    auto& prop = std::get<raft::device_span<double>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<raft::device_span<int32_t>>(property)) {
+    auto& prop = std::get<raft::device_span<int32_t>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<raft::device_span<int64_t>>(property)) {
+    auto& prop = std::get<raft::device_span<int64_t>>(property);
+    return func(prop);
+  } else {
+    CUGRAPH_EXPECTS(std::holds_alternative<raft::device_span<size_t>>(property),
+                    "unsupported variant type -- shouldn't happen");
+
+    auto& prop = std::get<raft::device_span<size_t>>(property);
+    return func(prop);
+  }
+}
+
+template <typename func_t>
+auto variant_type_dispatch(const_arithmetic_device_span_t& property, func_t func)
+{
+  if (std::holds_alternative<raft::device_span<float const>>(property)) {
+    auto& prop = std::get<raft::device_span<float const>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<raft::device_span<double const>>(property)) {
+    auto& prop = std::get<raft::device_span<double const>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<raft::device_span<int32_t const>>(property)) {
+    auto& prop = std::get<raft::device_span<int32_t const>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<raft::device_span<int64_t const>>(property)) {
+    auto& prop = std::get<raft::device_span<int64_t const>>(property);
+    return func(prop);
+  } else {
+    CUGRAPH_EXPECTS(std::holds_alternative<raft::device_span<size_t const>>(property),
+                    "unsupported variant type -- shouldn't happen");
+
+    auto& prop = std::get<raft::device_span<size_t const>>(property);
+    return func(prop);
+  }
+}
+
+template <typename edge_t, typename func_t>
+auto variant_type_dispatch(edge_arithmetic_property_view_t<edge_t>& property, func_t func)
+{
+  if (std::holds_alternative<cugraph::edge_property_view_t<edge_t, float const*>>(property)) {
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, float const*>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<cugraph::edge_property_view_t<edge_t, double const*>>(
+               property)) {
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, double const*>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<cugraph::edge_property_view_t<edge_t, int32_t const*>>(
+               property)) {
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, int32_t const*>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<cugraph::edge_property_view_t<edge_t, int64_t const*>>(
+               property)) {
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, int64_t const*>>(property);
+    return func(prop);
+  } else {
+    CUGRAPH_EXPECTS(
+      (std::holds_alternative<cugraph::edge_property_view_t<edge_t, size_t const*>>(property)),
+      "unsupported variant type -- shouldn't happen");
+
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, size_t const*>>(property);
+    return func(prop);
+  }
+}
+
+template <typename edge_t, typename func_t>
+auto variant_type_dispatch(edge_arithmetic_property_mutable_view_t<edge_t>& property, func_t func)
+{
+  if (std::holds_alternative<cugraph::edge_property_view_t<edge_t, float*>>(property)) {
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, float*>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<cugraph::edge_property_view_t<edge_t, double*>>(property)) {
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, double*>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<cugraph::edge_property_view_t<edge_t, int32_t*>>(property)) {
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, int32_t*>>(property);
+    return func(prop);
+  } else if (std::holds_alternative<cugraph::edge_property_view_t<edge_t, int64_t*>>(property)) {
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, int64_t*>>(property);
+    return func(prop);
+  } else {
+    CUGRAPH_EXPECTS(
+      (std::holds_alternative<cugraph::edge_property_view_t<edge_t, size_t*>>(property)),
+      "unsupported variant type -- shouldn't happen");
+
+    auto& prop = std::get<cugraph::edge_property_view_t<edge_t, size_t const*>>(property);
+    return func(prop);
+  }
+}
+
+struct sizeof_arithmetic_element {
+  template <typename T>
+  size_t operator()(rmm::device_uvector<T> const&) const
+  {
+    return sizeof(T);
+  }
+  template <typename T>
+  size_t operator()(raft::device_span<T> const&) const
+  {
+    return sizeof(T);
+  }
+  template <typename T>
+  size_t operator()(raft::device_span<T const> const&) const
+  {
+    return sizeof(T);
+  }
+};
+
+inline arithmetic_device_span_t make_arithmetic_device_span(arithmetic_device_uvector_t& v)
+{
+  return variant_type_dispatch(v, [](auto& v) {
+    using T = typename std::remove_reference<decltype(v)>::type::value_type;
+    return static_cast<arithmetic_device_span_t>(raft::device_span<T>(v.data(), v.size()));
+  });
+}
+
+inline std::vector<arithmetic_device_span_t> make_arithmetic_device_span_vector(
+  std::vector<arithmetic_device_uvector_t>& v)
+{
+  std::vector<arithmetic_device_span_t> results(v.size());
+  std::transform(
+    v.begin(), v.end(), results.begin(), [](auto& c) { return make_arithmetic_device_span(c); });
+  return results;
+}
+
+}  // namespace cugraph

cpp/include/cugraph/detail/shuffle_wrappers.hpp

@@ -15,6 +15,8 @@
  */
 #pragma once
 
+#include <cugraph/arithmetic_variant_types.hpp>
+
 #include <raft/core/handle.hpp>
 #include <raft/core/host_span.hpp>
 #include <raft/random/rng_state.hpp>
@@ -231,20 +233,12 @@ shuffle_int_vertex_value_pairs_to_local_gpu_by_vertex_partitioning(
  * groupby_and_count_local_partition is false) or in each segment with the same (local partition ID,
  * GPU ID) pair.
  */
-template <typename vertex_t,
-          typename edge_t,
-          typename weight_t,
-          typename edge_type_t,
-          typename edge_time_t>
+template <typename vertex_t>
 rmm::device_uvector<size_t> groupby_and_count_edgelist_by_local_partition_id(
   raft::handle_t const& handle,
-  rmm::device_uvector<vertex_t>& d_edgelist_majors,
-  rmm::device_uvector<vertex_t>& d_edgelist_minors,
-  std::optional<rmm::device_uvector<weight_t>>& d_edgelist_weights,
-  std::optional<rmm::device_uvector<edge_t>>& d_edgelist_edge_ids,
-  std::optional<rmm::device_uvector<edge_type_t>>& d_edgelist_edge_types,
-  std::optional<rmm::device_uvector<edge_time_t>>& d_edgelist_edge_start_times,
-  std::optional<rmm::device_uvector<edge_time_t>>& d_edgelist_edge_end_times,
+  raft::device_span<vertex_t> edgelist_majors,
+  raft::device_span<vertex_t> edgelist_minors,
+  raft::host_span<cugraph::arithmetic_device_span_t> edgelist_properties,
   bool groupby_and_count_local_partition_by_minor = false);
 
 /**