ANN_BENCH (#130)

Porting the ANN benchmarks from RAFT.
- [x] Make it build

Sanity check that benchmarks work (runs and gives reasonable recall for Deep-1M dataset)
- [x] cuVS brute force kNN 
- [x] cuVS IVF-Flat
- [x] cuVS IVF-PQ (+ refinement)
- [x] cuVS CAGRA
- [x] cuVS CAGRA-Q (+refinement)
- [x] Faiss GPU/CPU IVF-Flat & IVF-PQ
- [x] HNSW
- [x] CAGRA + HNSW 
- [x] GGNN

NB: the indices built using the old ANN_BENCH in raft tend to crash in cuvs search benchmarks during index deserialization - don't forget to build the indexes anew when testing.

Authors:
  - Artem M. Chirkin (https://github.com/achirkin)
  - Malte Förster (https://github.com/mfoerste4)
  - Tamas Bela Feher (https://github.com/tfeher)
  - Micka (https://github.com/lowener)
  - Corey J. Nolet (https://github.com/cjnolet)

Approvers:
  - Tamas Bela Feher (https://github.com/tfeher)
  - James Lamb (https://github.com/jameslamb)
  - Corey J. Nolet (https://github.com/cjnolet)

URL: #130

build.sh

@@ -18,8 +18,8 @@ ARGS=$*
 # scripts, and that this script resides in the repo dir!
 REPODIR=$(cd $(dirname $0); pwd)
 
-VALIDARGS="clean libcuvs python rust docs tests examples --uninstall  -v -g -n --compile-static-lib --allgpuarch --no-nvtx --show_depr_warn --incl-cache-stats --time -h"
-HELP="$0 [<target> ...] [<flag> ...] [--cmake-args=\"<args>\"] [--cache-tool=<tool>] [--limit-tests=<targets>] [--build-metrics=<filename>]
+VALIDARGS="clean libcuvs python rust docs tests bench-ann examples --uninstall  -v -g -n --compile-static-lib --allgpuarch --no-nvtx --show_depr_warn --incl-cache-stats --time -h"
+HELP="$0 [<target> ...] [<flag> ...] [--cmake-args=\"<args>\"] [--cache-tool=<tool>] [--limit-tests=<targets>] [--limit-bench-ann=<targets>] [--build-metrics=<filename>]
  where <target> is:
    clean            - remove all existing build artifacts and configuration (start over)
    libcuvs          - build the cuvs C++ code only. Also builds the C-wrapper library
@@ -28,6 +28,7 @@ HELP="$0 [<target> ...] [<flag> ...] [--cmake-args=\"<args>\"] [--cache-tool=<to
    rust             - build the cuvs Rust bindings
    docs             - build the documentation
    tests            - build the tests
+   bench-ann        - build end-to-end ann benchmarks
    examples         - build the examples
 
  and <flag> is:
@@ -37,6 +38,7 @@ HELP="$0 [<target> ...] [<flag> ...] [--cmake-args=\"<args>\"] [--cache-tool=<to
    --uninstall                 - uninstall files for specified targets which were built and installed prior
    --compile-static-lib        - compile static library for all components
    --limit-tests               - semicolon-separated list of test executables to compile (e.g. NEIGHBORS_TEST;CLUSTER_TEST)
+   --limit-bench-ann           - semicolon-separated list of ann benchmark executables to compute (e.g. HNSWLIB_ANN_BENCH;RAFT_IVF_PQ_ANN_BENCH)
    --allgpuarch                - build for all supported GPU architectures
    --no-nvtx                   - disable nvtx (profiling markers), but allow enabling it in downstream projects
    --show_depr_warn            - show cmake deprecation warnings
@@ -70,6 +72,7 @@ BUILD_REPORT_METRICS=""
 BUILD_REPORT_INCL_CACHE_STATS=OFF
 
 TEST_TARGETS="NEIGHBORS_ANN_CAGRA_TEST"
+ANN_BENCH_TARGETS="CUVS_ANN_BENCH_ALL"
 
 CACHE_ARGS=""
 NVTX=ON
@@ -150,6 +153,21 @@ function limitTests {
     fi
 }
 
+function limitAnnBench {
+    # Check for option to limit the set of test binaries to build
+    if [[ -n $(echo $ARGS | { grep -E "\-\-limit\-bench-ann" || true; } ) ]]; then
+        # There are possible weird edge cases that may cause this regex filter to output nothing and fail silently
+        # the true pipe will catch any weird edge cases that may happen and will cause the program to fall back
+        # on the invalid option error
+        LIMIT_ANN_BENCH_TARGETS=$(echo $ARGS | sed -e 's/.*--limit-bench-ann=//' -e 's/ .*//')
+        if [[ -n ${LIMIT_ANN_BENCH_TARGETS} ]]; then
+            # Remove the full LIMIT_TEST_TARGETS argument from list of args so that it passes validArgs function
+            ARGS=${ARGS//--limit-bench-ann=$LIMIT_ANN_BENCH_TARGETS/}
+            ANN_BENCH_TARGETS=${LIMIT_ANN_BENCH_TARGETS}
+        fi
+    fi
+}
+
 function buildMetrics {
     # Check for multiple build-metrics options
     if [[ $(echo $ARGS | { grep -Eo "\-\-build\-metrics" || true; } | wc -l ) -gt 1 ]]; then
@@ -179,6 +197,7 @@ if (( ${NUMARGS} != 0 )); then
     cmakeArgs
     cacheTool
     limitTests
+    limitAnnBench
     buildMetrics
     for a in ${ARGS}; do
         if ! (echo " ${VALIDARGS} " | grep -q " ${a} "); then
@@ -255,6 +274,11 @@ if hasArg tests || (( ${NUMARGS} == 0 )); then
     fi
 fi
 
+if hasArg bench-ann || (( ${NUMARGS} == 0 )); then
+    BUILD_ANN_BENCH=ON
+    CMAKE_TARGET="${CMAKE_TARGET};${ANN_BENCH_TARGETS}"
+fi
+
 if hasArg --no-nvtx; then
     NVTX=OFF
 fi
@@ -327,6 +351,7 @@ if (( ${NUMARGS} == 0 )) || hasArg libcuvs || hasArg docs || hasArg tests || has
           -DDISABLE_DEPRECATION_WARNINGS=${DISABLE_DEPRECATION_WARNINGS} \
           -DBUILD_TESTS=${BUILD_TESTS} \
           -DBUILD_C_TESTS=${BUILD_TESTS} \
+          -DBUILD_ANN_BENCH=${BUILD_ANN_BENCH} \
           -DBUILD_CPU_ONLY=${BUILD_CPU_ONLY} \
           -DCMAKE_MESSAGE_LOG_LEVEL=${CMAKE_LOG_LEVEL} \
           ${CACHE_ARGS} \

conda/environments/all_cuda-118_arch-aarch64.yaml

@@ -41,6 +41,7 @@ dependencies:
 - numpy>=1.23,<2.0a0
 - numpydoc
 - nvcc_linux-aarch64=11.8
+- openblas
 - pre-commit
 - pydata-sphinx-theme
 - pylibraft==24.8.*,>=0.0.0a0

conda/environments/all_cuda-118_arch-x86_64.yaml

@@ -41,6 +41,7 @@ dependencies:
 - numpy>=1.23,<2.0a0
 - numpydoc
 - nvcc_linux-64=11.8
+- openblas
 - pre-commit
 - pydata-sphinx-theme
 - pylibraft==24.8.*,>=0.0.0a0

conda/environments/all_cuda-122_arch-aarch64.yaml

@@ -37,6 +37,7 @@ dependencies:
 - ninja
 - numpy>=1.23,<2.0a0
 - numpydoc
+- openblas
 - pre-commit
 - pydata-sphinx-theme
 - pylibraft==24.8.*,>=0.0.0a0

conda/environments/all_cuda-122_arch-x86_64.yaml

@@ -37,6 +37,7 @@ dependencies:
 - ninja
 - numpy>=1.23,<2.0a0
 - numpydoc
+- openblas
 - pre-commit
 - pydata-sphinx-theme
 - pylibraft==24.8.*,>=0.0.0a0

conda/recipes/libcuvs/build_libcuvs_tests.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
 # Copyright (c) 2022-2024, NVIDIA CORPORATION.
 
-./build.sh tests --allgpuarch --no-nvtx --build-metrics=tests_bench --incl-cache-stats
+./build.sh tests bench-ann --allgpuarch --no-nvtx --build-metrics=tests_bench --incl-cache-stats
 cmake --install cpp/build --component testing

conda/recipes/libcuvs/meta.yaml

@@ -198,6 +198,7 @@ outputs:
         - libraft ={{ minor_version }}
         - {{ pin_subpackage('libcuvs', exact=True) }}
         - cuda-version ={{ cuda_version }}
+        - openblas # required by some CPU algos in benchmarks
         {% if cuda_major == "11" %}
         - cuda-profiler-api {{ cuda11_cuda_profiler_api_run_version }}
         - libcublas {{ cuda11_libcublas_host_version }}

cpp/CMakeLists.txt

@@ -55,6 +55,7 @@ option(BUILD_SHARED_LIBS "Build cuvs shared libraries" ON)
 option(BUILD_TESTS "Build cuvs unit-tests" ON)
 option(BUILD_C_LIBRARY "Build raft C API library" OFF)
 option(BUILD_C_TESTS "Build raft C API tests" OFF)
+option(BUILD_ANN_BENCH "Build cuVS ann benchmarks" OFF)
 option(CUDA_ENABLE_KERNELINFO "Enable kernel resource usage info" OFF)
 option(CUDA_ENABLE_LINEINFO
        "Enable the -lineinfo option for nvcc (useful for cuda-memcheck / profiler)" OFF
@@ -92,6 +93,7 @@ include(CMakeDependentOption)
 
 message(VERBOSE "cuVS: Build cuVS unit-tests: ${BUILD_TESTS}")
 message(VERBOSE "cuVS: Build CPU only components: ${BUILD_CPU_ONLY}")
+message(VERBOSE "cuVS: Build ANN benchmarks: ${BUILD_ANN_BENCH}")
 message(VERBOSE "cuVS: Enable detection of conda environment for dependencies: ${DETECT_CONDA_ENV}")
 message(VERBOSE "cuVS: Disable depreaction warnings " ${DISABLE_DEPRECATION_WARNINGS})
 message(VERBOSE "cuVS: Disable OpenMP: ${DISABLE_OPENMP}")
@@ -184,6 +186,11 @@ endif()
 
 include(cmake/thirdparty/get_cutlass.cmake)
 
+if(BUILD_ANN_BENCH)
+  include(${rapids-cmake-dir}/cpm/gbench.cmake)
+  rapids_cpm_gbench(BUILD_STATIC)
+endif()
+
 # ##################################################################################################
 # * cuvs ---------------------------------------------------------------------
 
@@ -663,3 +670,10 @@ if(BUILD_TESTS OR BUILD_C_TESTS)
   include(internal/CMakeLists.txt)
   include(test/CMakeLists.txt)
 endif()
+
+# ##################################################################################################
+# * build ann benchmark executable -----------------------------------------------
+
+if(BUILD_ANN_BENCH)
+  include(bench/ann/CMakeLists.txt)
+endif()