NVIDIA · AyodeAwe · May 19, 2025 · May 14, 2025 · May 15, 2025 · May 15, 2025
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -48,7 +48,7 @@ repos:
                 additional_dependencies: [toml]
                 args: ["--config=pyproject.toml"]
       - repo: https://github.com/pre-commit/mirrors-clang-format
-        rev: v16.0.6
+        rev: v20.1.4
         hooks:
               - id: clang-format
                 types_or: [c, c++, cuda]

diff --git a/conda/environments/all_cuda-118_arch-aarch64.yaml b/conda/environments/all_cuda-118_arch-aarch64.yaml
@@ -9,8 +9,8 @@ channels:
 dependencies:
 - breathe>=4.35.0
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-nvtx=11.8
 - cuda-profiler-api=11.8.86
@@ -26,7 +26,7 @@ dependencies:
 - go
 - graphviz
 - ipython
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev=11.11.3.6
 - libcublas=11.11.3.6
 - libcurand-dev=10.3.0.86

diff --git a/conda/environments/all_cuda-118_arch-x86_64.yaml b/conda/environments/all_cuda-118_arch-x86_64.yaml
@@ -9,8 +9,8 @@ channels:
 dependencies:
 - breathe>=4.35.0
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-nvtx=11.8
 - cuda-profiler-api=11.8.86
@@ -26,7 +26,7 @@ dependencies:
 - go
 - graphviz
 - ipython
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev=11.11.3.6
 - libcublas=11.11.3.6
 - libcurand-dev=10.3.0.86

diff --git a/conda/environments/all_cuda-128_arch-aarch64.yaml b/conda/environments/all_cuda-128_arch-aarch64.yaml
@@ -9,8 +9,8 @@ channels:
 dependencies:
 - breathe>=4.35.0
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-cudart-dev
 - cuda-nvcc
@@ -27,7 +27,7 @@ dependencies:
 - go
 - graphviz
 - ipython
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev
 - libcurand-dev
 - libcusolver-dev

diff --git a/conda/environments/all_cuda-128_arch-x86_64.yaml b/conda/environments/all_cuda-128_arch-x86_64.yaml
@@ -9,8 +9,8 @@ channels:
 dependencies:
 - breathe>=4.35.0
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-cudart-dev
 - cuda-nvcc
@@ -27,7 +27,7 @@ dependencies:
 - go
 - graphviz
 - ipython
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev
 - libcurand-dev
 - libcusolver-dev

diff --git a/conda/environments/bench_ann_cuda-118_arch-aarch64.yaml b/conda/environments/bench_ann_cuda-118_arch-aarch64.yaml
@@ -9,8 +9,8 @@ channels:
 dependencies:
 - benchmark>=1.8.2
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - click
 - cmake>=3.30.4
 - cuda-nvtx=11.8
@@ -26,7 +26,7 @@ dependencies:
 - gcc_linux-aarch64=11.*
 - glog>=0.6.0
 - h5py>=3.8.0
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev=11.11.3.6
 - libcublas=11.11.3.6
 - libcurand-dev=10.3.0.86

diff --git a/conda/environments/bench_ann_cuda-118_arch-x86_64.yaml b/conda/environments/bench_ann_cuda-118_arch-x86_64.yaml
@@ -9,8 +9,8 @@ channels:
 dependencies:
 - benchmark>=1.8.2
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - click
 - cmake>=3.30.4
 - cuda-nvtx=11.8
@@ -28,7 +28,7 @@ dependencies:
 - h5py>=3.8.0
 - libaio
 - libboost-devel=1.87
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev=11.11.3.6
 - libcublas=11.11.3.6
 - libcurand-dev=10.3.0.86

diff --git a/conda/environments/bench_ann_cuda-128_arch-aarch64.yaml b/conda/environments/bench_ann_cuda-128_arch-aarch64.yaml
@@ -9,8 +9,8 @@ channels:
 dependencies:
 - benchmark>=1.8.2
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - click
 - cmake>=3.30.4
 - cuda-cudart-dev
@@ -27,7 +27,7 @@ dependencies:
 - gcc_linux-aarch64=13.*
 - glog>=0.6.0
 - h5py>=3.8.0
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev
 - libcurand-dev
 - libcusolver-dev

diff --git a/conda/environments/bench_ann_cuda-128_arch-x86_64.yaml b/conda/environments/bench_ann_cuda-128_arch-x86_64.yaml
@@ -9,8 +9,8 @@ channels:
 dependencies:
 - benchmark>=1.8.2
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - click
 - cmake>=3.30.4
 - cuda-cudart-dev
@@ -29,7 +29,7 @@ dependencies:
 - h5py>=3.8.0
 - libaio
 - libboost-devel=1.87
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev
 - libcurand-dev
 - libcusolver-dev

diff --git a/conda/environments/go_cuda-118_arch-aarch64.yaml b/conda/environments/go_cuda-118_arch-aarch64.yaml
@@ -8,8 +8,8 @@ channels:
 - nvidia
 dependencies:
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-nvtx=11.8
 - cuda-profiler-api=11.8.86
@@ -19,7 +19,7 @@ dependencies:
 - dlpack>=0.8,<1.0
 - gcc_linux-aarch64=11.*
 - go
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev=11.11.3.6
 - libcublas=11.11.3.6
 - libcurand-dev=10.3.0.86

diff --git a/conda/environments/go_cuda-118_arch-x86_64.yaml b/conda/environments/go_cuda-118_arch-x86_64.yaml
@@ -8,8 +8,8 @@ channels:
 - nvidia
 dependencies:
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-nvtx=11.8
 - cuda-profiler-api=11.8.86
@@ -19,7 +19,7 @@ dependencies:
 - dlpack>=0.8,<1.0
 - gcc_linux-64=11.*
 - go
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev=11.11.3.6
 - libcublas=11.11.3.6
 - libcurand-dev=10.3.0.86

diff --git a/conda/environments/go_cuda-128_arch-aarch64.yaml b/conda/environments/go_cuda-128_arch-aarch64.yaml
@@ -8,8 +8,8 @@ channels:
 - nvidia
 dependencies:
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-cudart-dev
 - cuda-nvcc
@@ -20,7 +20,7 @@ dependencies:
 - dlpack>=0.8,<1.0
 - gcc_linux-aarch64=13.*
 - go
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev
 - libcurand-dev
 - libcusolver-dev

diff --git a/conda/environments/go_cuda-128_arch-x86_64.yaml b/conda/environments/go_cuda-128_arch-x86_64.yaml
@@ -8,8 +8,8 @@ channels:
 - nvidia
 dependencies:
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-cudart-dev
 - cuda-nvcc
@@ -20,7 +20,7 @@ dependencies:
 - dlpack>=0.8,<1.0
 - gcc_linux-64=13.*
 - go
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev
 - libcurand-dev
 - libcusolver-dev

diff --git a/...ironments/rust_cuda-128_arch-aarch64.yaml → ...ironments/rust_cuda-129_arch-aarch64.yaml b/...ironments/rust_cuda-128_arch-aarch64.yaml → ...ironments/rust_cuda-129_arch-aarch64.yaml
@@ -8,17 +8,17 @@ channels:
 - nvidia
 dependencies:
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-cudart-dev
 - cuda-nvcc
 - cuda-nvtx-dev
 - cuda-profiler-api
-- cuda-version=12.8
+- cuda-version=12.9
 - cxx-compiler
 - gcc_linux-aarch64=13.*
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev
 - libcurand-dev
 - libcusolver-dev
@@ -30,4 +30,4 @@ dependencies:
 - ninja
 - rust
 - sysroot_linux-aarch64==2.28
-name: rust_cuda-128_arch-aarch64
+name: rust_cuda-129_arch-aarch64
diff --git a/...vironments/rust_cuda-128_arch-x86_64.yaml → ...vironments/rust_cuda-129_arch-x86_64.yaml b/...vironments/rust_cuda-128_arch-x86_64.yaml → ...vironments/rust_cuda-129_arch-x86_64.yaml
@@ -8,17 +8,17 @@ channels:
 - nvidia
 dependencies:
 - c-compiler
-- clang-tools==16.0.6
-- clang==16.0.6
+- clang-tools==20.1.4
+- clang==20.1.4
 - cmake>=3.30.4
 - cuda-cudart-dev
 - cuda-nvcc
 - cuda-nvtx-dev
 - cuda-profiler-api
-- cuda-version=12.8
+- cuda-version=12.9
 - cxx-compiler
 - gcc_linux-64=13.*
-- libclang==16.0.6
+- libclang==20.1.4
 - libcublas-dev
 - libcurand-dev
 - libcusolver-dev
@@ -30,4 +30,4 @@ dependencies:
 - ninja
 - rust
 - sysroot_linux-64==2.28
-name: rust_cuda-128_arch-x86_64
+name: rust_cuda-129_arch-x86_64
diff --git a/cpp/bench/ann/CMakeLists.txt b/cpp/bench/ann/CMakeLists.txt
@@ -21,11 +21,15 @@ option(CUVS_ANN_BENCH_USE_FAISS_GPU_FLAT "Include faiss' brute-force knn algorit
 option(CUVS_ANN_BENCH_USE_FAISS_GPU_IVF_FLAT "Include faiss' ivf flat algorithm in benchmark" ON)
 option(CUVS_ANN_BENCH_USE_FAISS_GPU_IVF_PQ "Include faiss' ivf pq algorithm in benchmark" ON)
 option(CUVS_ANN_BENCH_USE_FAISS_GPU_CAGRA "Include faiss' cagra algorithm in benchmark" ON)
+option(CUVS_ANN_BENCH_USE_FAISS_GPU_CAGRA_HNSW
+       "Include faiss' cagra algorithm for build and hnsw for search in benchmark" ON
+)
 option(CUVS_ANN_BENCH_USE_FAISS_CPU_FLAT "Include faiss' cpu brute-force algorithm in benchmark" ON)
 option(CUVS_ANN_BENCH_USE_FAISS_CPU_IVF_FLAT "Include faiss' cpu ivf flat algorithm in benchmark"
        ON
 )
 option(CUVS_ANN_BENCH_USE_FAISS_CPU_IVF_PQ "Include faiss' cpu ivf pq algorithm in benchmark" ON)
+option(CUVS_ANN_BENCH_USE_FAISS_CPU_HNSW_FLAT "Include faiss' hnsw algorithm in benchmark" ON)
 option(CUVS_ANN_BENCH_USE_CUVS_IVF_FLAT "Include cuVS ivf flat algorithm in benchmark" ON)
 option(CUVS_ANN_BENCH_USE_CUVS_IVF_PQ "Include cuVS ivf pq algorithm in benchmark" ON)
 option(CUVS_ANN_BENCH_USE_CUVS_CAGRA "Include cuVS CAGRA in benchmark" ON)
@@ -276,6 +280,12 @@ if(CUVS_ANN_BENCH_USE_FAISS_CPU_IVF_PQ)
   )
 endif()
 
+if(CUVS_ANN_BENCH_USE_FAISS_CPU_HNSW_FLAT)
+  ConfigureAnnBench(
+    NAME FAISS_CPU_HNSW_FLAT PATH src/faiss/faiss_cpu_benchmark.cpp LINKS ${CUVS_FAISS_TARGETS}
+  )
+endif()
+
 if(CUVS_ANN_BENCH_USE_FAISS_GPU_IVF_FLAT AND CUVS_FAISS_ENABLE_GPU)
   ConfigureAnnBench(
     NAME FAISS_GPU_IVF_FLAT PATH src/faiss/faiss_gpu_benchmark.cu LINKS ${CUVS_FAISS_TARGETS}
@@ -304,6 +314,13 @@ if(CUVS_ANN_BENCH_USE_FAISS_GPU_CAGRA AND CUVS_FAISS_ENABLE_GPU)
   )
 endif()
 
+if(CUVS_ANN_BENCH_USE_FAISS_GPU_CAGRA_HNSW AND CUVS_FAISS_ENABLE_GPU)
+  ConfigureAnnBench(
+    NAME FAISS_GPU_CAGRA_HNSW PATH src/faiss/faiss_gpu_benchmark.cu LINKS ${CUVS_FAISS_TARGETS}
+    raft::raft
+  )
+endif()
+
 if(CUVS_ANN_BENCH_USE_GGNN)
   include(cmake/thirdparty/get_glog)
   ConfigureAnnBench(

diff --git a/cpp/bench/ann/src/common/ann_types.hpp b/cpp/bench/ann/src/common/ann_types.hpp
@@ -168,7 +168,7 @@ class algo : public algo_base {
   // and set_search_dataset() should save the passed-in pointer somewhere.
   // The client code should call set_search_dataset() before searching,
   // and should not release dataset before searching is finished.
-  virtual void set_search_dataset(const T* /*dataset*/, size_t /*nrow*/){};
+  virtual void set_search_dataset(const T* /*dataset*/, size_t /*nrow*/) {};
 
   /**
    * Make a shallow copy of the algo wrapper that shares the resources and ensures thread-safe
@@ -181,6 +181,6 @@ class algo : public algo_base {
 #define REGISTER_ALGO_INSTANCE(DataT)                                                              \
   template auto cuvs::bench::create_algo<DataT>(                                                   \
     const std::string&, const std::string&, int, const nlohmann::json&)                            \
-    ->std::unique_ptr<cuvs::bench::algo<DataT>>;                                                   \
+    -> std::unique_ptr<cuvs::bench::algo<DataT>>;                                                  \
   template auto cuvs::bench::create_search_param<DataT>(const std::string&, const nlohmann::json&) \
-    ->std::unique_ptr<typename cuvs::bench::algo<DataT>::search_param>;
+    -> std::unique_ptr<typename cuvs::bench::algo<DataT>::search_param>;
diff --git a/cpp/bench/ann/src/common/util.hpp b/cpp/bench/ann/src/common/util.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2023-2024, NVIDIA CORPORATION.
+ * Copyright (c) 2023-2025, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -388,20 +388,22 @@ inline auto host_info()
   uint64_t cpu_freq_max    = 0;
   int host_processors_used = 0;
   int host_cores_used      = 0;
-  std::set<int> host_cores_selected{};
+  std::set<std::pair<int, int>> host_cores_selected{};  // pairs of (socket_id, core_id)
   for (int cpu_id = 0; cpu_id < host_processors_configured; cpu_id++) {
     if (CPU_ISSET_S(cpu_id, affinity_mask_buf.size(), affinity_mask) == 0) { continue; }
     host_processors_used++;
     std::string cpu_fpath = "/sys/devices/system/cpu/cpu" + std::to_string(cpu_id);
     if (!std::filesystem::exists(cpu_fpath)) { continue; }
 
     int this_cpu_core          = 0;
+    int this_cpu_package       = 0;
     uint64_t this_cpu_freq_min = 0;
     uint64_t this_cpu_freq_max = 0;
     std::ifstream(cpu_fpath + "/topology/core_id") >> this_cpu_core;
+    std::ifstream(cpu_fpath + "/topology/physical_package_id") >> this_cpu_package;
     std::ifstream(cpu_fpath + "/cpufreq/scaling_min_freq") >> this_cpu_freq_min;
     std::ifstream(cpu_fpath + "/cpufreq/scaling_max_freq") >> this_cpu_freq_max;
-    host_cores_selected.insert(this_cpu_core);
+    host_cores_selected.insert(std::make_pair(this_cpu_package, this_cpu_core));
     cpu_freq_min = cpu_freq_min == 0
                      ? (this_cpu_freq_min * 1000ull)
                      : std::min<uint64_t>(this_cpu_freq_min * 1000ull, cpu_freq_min);