Optimize hnsw::from_cagra<GPU> (#826)

Reduce the CAGRA-for-HNSW build times by:
 - avoiding unnecessary copies of the data between cagra::build and hnsw::from_cagra in the benchmarks
 - avoiding unnecessary temporary data buffers in hnsw::from_cagra<GPU>
 - reducing random reads via forcing 1-1 mapping between the internal indices and external labels during HNSW import

As a side-effect, this PR also fixes the bug where hnsw::from_cagra segfaults in benchmarks if the dataset is passed in device memory (and incorrectly wrapped in a host_matrix_view).

In addition, this PR adds a bit more verbose NVTX reporting of different stages during the CAGRA/HNSW index build.

Authors:
  - Artem M. Chirkin (https://github.com/achirkin)
  - Corey J. Nolet (https://github.com/cjnolet)

Approvers:
  - Corey J. Nolet (https://github.com/cjnolet)

URL: #826

Author: achirkin

cpp/bench/ann/src/cuvs/cuvs_cagra_hnswlib_wrapper.h

@@ -39,9 +39,7 @@ class cuvs_cagra_hnswlib : public algo<T>, public algo_gpu {
   };
 
   cuvs_cagra_hnswlib(Metric metric, int dim, const build_param& param, int concurrent_searches = 1)
-    : algo<T>(metric, dim),
-      build_param_{param},
-      cagra_build_{metric, dim, param.cagra_build_param, concurrent_searches}
+    : algo<T>(metric, dim), build_param_{param}
   {
   }
 
@@ -57,7 +55,13 @@ class cuvs_cagra_hnswlib : public algo<T>, public algo_gpu {
 
   [[nodiscard]] auto get_sync_stream() const noexcept -> cudaStream_t override
   {
-    return cagra_build_.get_sync_stream();
+    return handle_.get_sync_stream();
+  }
+
+  [[nodiscard]] auto uses_stream() const noexcept -> bool override
+  {
+    // there's no need to synchronize with the GPU neither on build nor on search
+    return false;
   }
 
   // to enable dataset access from GPU memory
@@ -77,28 +81,37 @@ class cuvs_cagra_hnswlib : public algo<T>, public algo_gpu {
   }
 
  private:
-  raft::resources handle_{};
+  configured_raft_resources handle_{};
   build_param build_param_;
   search_param search_param_;
-  cuvs_cagra<T, IdxT> cagra_build_;
   std::shared_ptr<cuvs::neighbors::hnsw::index<T>> hnsw_index_;
 };
 
 template <typename T, typename IdxT>
 void cuvs_cagra_hnswlib<T, IdxT>::build(const T* dataset, size_t nrow)
 {
-  cagra_build_.build(dataset, nrow);
-  auto* cagra_index      = cagra_build_.get_index();
-  auto host_dataset_view = raft::make_host_matrix_view<const T, int64_t>(dataset, nrow, this->dim_);
-  auto opt_dataset_view =
-    std::optional<raft::host_matrix_view<const T, int64_t>>(std::move(host_dataset_view));
-  const auto start_clock = std::chrono::system_clock::now();
-  hnsw_index_            = cuvs::neighbors::hnsw::from_cagra(
-    handle_, build_param_.hnsw_index_params, *cagra_index, opt_dataset_view);
-  int time =
-    std::chrono::duration_cast<std::chrono::seconds>(std::chrono::system_clock::now() - start_clock)
-      .count();
-  RAFT_LOG_DEBUG("Graph saved to HNSW format in %d:%d min", time / 60, time % 60);
+  // when the data set is on host, we can pass it directly to HNSW
+  bool dataset_is_on_host = raft::get_device_for_address(dataset) == -1;
+
+  // re-use the CAGRA wrapper to parse build params
+  auto bps                                 = build_param_.cagra_build_param;
+  bps.cagra_params.attach_dataset_on_build = !dataset_is_on_host;
+  cuvs_cagra<T, IdxT> cagra_wrapper{this->metric_, this->dim_, bps};
+
+  // build the CAGRA index
+  cagra_wrapper.build(dataset, nrow);
+  auto& cagra_index = *cagra_wrapper.get_index();
+
+  // pass the dataset directly to HNSW if it's on the host
+  std::optional<raft::host_matrix_view<const T, int64_t>> opt_dataset_view = std::nullopt;
+  if (dataset_is_on_host) {
+    opt_dataset_view.emplace(
+      raft::make_host_matrix_view<const T, int64_t>(dataset, nrow, this->dim_));
+  }
+
+  // convert the index to HNSW format
+  hnsw_index_ = cuvs::neighbors::hnsw::from_cagra(
+    handle_, build_param_.hnsw_index_params, cagra_index, opt_dataset_view);
 }
 
 template <typename T, typename IdxT>

cpp/src/neighbors/detail/cagra/cagra_build.cuh

@@ -574,6 +574,13 @@ index<T, IdxT> build(
 {
   size_t intermediate_degree = params.intermediate_graph_degree;
   size_t graph_degree        = params.graph_degree;
+  common::nvtx::range<common::nvtx::domain::cuvs> function_scope(
+    "cagra::build<%s>(%zu, %zu)",
+    Accessor::is_managed_type::value ? "managed"
+    : Accessor::is_host_type::value  ? "host"
+                                     : "device",
+    intermediate_degree,
+    graph_degree);
   if (intermediate_degree >= static_cast<size_t>(dataset.extent(0))) {
     RAFT_LOG_WARN(
       "Intermediate graph degree cannot be larger than dataset size, reducing it to %lu",