Merge branch 'branch-23.10' of https://github.com/rapidsai/raft into …

…faiss-ivf

CHANGELOG.md

@@ -1,3 +1,97 @@
+# raft 23.08.00 (9 Aug 2023)
+
+## 🚨 Breaking Changes
+
+- Separate CAGRA index type from internal idx type ([#1664](https://github.com/rapidsai/raft/pull/1664)) [@tfeher](https://github.com/tfeher)
+- Stop using setup.py in build.sh ([#1645](https://github.com/rapidsai/raft/pull/1645)) [@vyasr](https://github.com/vyasr)
+- CAGRA max_queries auto configuration ([#1613](https://github.com/rapidsai/raft/pull/1613)) [@enp1s0](https://github.com/enp1s0)
+- Rename the CAGRA prune function to optimize ([#1588](https://github.com/rapidsai/raft/pull/1588)) [@enp1s0](https://github.com/enp1s0)
+- CAGRA pad dataset for 128bit vectorized load ([#1505](https://github.com/rapidsai/raft/pull/1505)) [@tfeher](https://github.com/tfeher)
+- Sparse Pairwise Distances API Updates ([#1502](https://github.com/rapidsai/raft/pull/1502)) [@divyegala](https://github.com/divyegala)
+- Cagra index construction without copying device mdarrays ([#1494](https://github.com/rapidsai/raft/pull/1494)) [@tfeher](https://github.com/tfeher)
+- [FEA] Masked NN for connect_components ([#1445](https://github.com/rapidsai/raft/pull/1445)) [@tarang-jain](https://github.com/tarang-jain)
+- Limiting workspace memory resource ([#1356](https://github.com/rapidsai/raft/pull/1356)) [@achirkin](https://github.com/achirkin)
+
+## 🐛 Bug Fixes
+
+- Remove push condition on docs-build ([#1693](https://github.com/rapidsai/raft/pull/1693)) [@raydouglass](https://github.com/raydouglass)
+- IVF-PQ: Fix illegal memory access with large max_samples ([#1685](https://github.com/rapidsai/raft/pull/1685)) [@achirkin](https://github.com/achirkin)
+- Fix missing parameter for select_k ([#1682](https://github.com/rapidsai/raft/pull/1682)) [@ucassjy](https://github.com/ucassjy)
+- Separate CAGRA index type from internal idx type ([#1664](https://github.com/rapidsai/raft/pull/1664)) [@tfeher](https://github.com/tfeher)
+- Add rmm to pylibraft run dependencies, since it is used by Cython. ([#1656](https://github.com/rapidsai/raft/pull/1656)) [@bdice](https://github.com/bdice)
+- Hotfix: wrong constant in IVF-PQ fp_8bit2half ([#1654](https://github.com/rapidsai/raft/pull/1654)) [@achirkin](https://github.com/achirkin)
+- Fix sparse KNN for large batches ([#1640](https://github.com/rapidsai/raft/pull/1640)) [@viclafargue](https://github.com/viclafargue)
+- Fix uploading of RAFT nightly packages ([#1638](https://github.com/rapidsai/raft/pull/1638)) [@dantegd](https://github.com/dantegd)
+- Fix cagra multi CTA bug ([#1628](https://github.com/rapidsai/raft/pull/1628)) [@enp1s0](https://github.com/enp1s0)
+- pass correct stream to cutlass kernel launch of L2/cosine pairwise distance kernels ([#1597](https://github.com/rapidsai/raft/pull/1597)) [@mdoijade](https://github.com/mdoijade)
+- Fix launchconfig y-gridsize too large in epilogue kernel ([#1586](https://github.com/rapidsai/raft/pull/1586)) [@mfoerste4](https://github.com/mfoerste4)
+- Fix update version and pinnings for 23.08. ([#1556](https://github.com/rapidsai/raft/pull/1556)) [@bdice](https://github.com/bdice)
+- Fix for function exposing KNN merge ([#1418](https://github.com/rapidsai/raft/pull/1418)) [@viclafargue](https://github.com/viclafargue)
+
+## 📖 Documentation
+
+- Critical doc fixes and updates for 23.08 ([#1705](https://github.com/rapidsai/raft/pull/1705)) [@cjnolet](https://github.com/cjnolet)
+- Fix the documentation about changing the logging level ([#1596](https://github.com/rapidsai/raft/pull/1596)) [@enp1s0](https://github.com/enp1s0)
+- Fix raft::bitonic_sort small usage example ([#1580](https://github.com/rapidsai/raft/pull/1580)) [@enp1s0](https://github.com/enp1s0)
+
+## 🚀 New Features
+
+- Use rapids-cmake new parallel testing feature ([#1623](https://github.com/rapidsai/raft/pull/1623)) [@robertmaynard](https://github.com/robertmaynard)
+- Add support for row-major slice ([#1591](https://github.com/rapidsai/raft/pull/1591)) [@lowener](https://github.com/lowener)
+- IVF-PQ tutorial notebook ([#1544](https://github.com/rapidsai/raft/pull/1544)) [@achirkin](https://github.com/achirkin)
+- [FEA] Masked NN for connect_components ([#1445](https://github.com/rapidsai/raft/pull/1445)) [@tarang-jain](https://github.com/tarang-jain)
+- raft: Build CUDA 12 packages ([#1388](https://github.com/rapidsai/raft/pull/1388)) [@vyasr](https://github.com/vyasr)
+- Limiting workspace memory resource ([#1356](https://github.com/rapidsai/raft/pull/1356)) [@achirkin](https://github.com/achirkin)
+
+## 🛠️ Improvements
+
+- Pin `dask` and `distributed` for `23.08` release ([#1711](https://github.com/rapidsai/raft/pull/1711)) [@galipremsagar](https://github.com/galipremsagar)
+- Add algo parameter for CAGRA ANN bench ([#1687](https://github.com/rapidsai/raft/pull/1687)) [@tfeher](https://github.com/tfeher)
+- ANN benchmarks python wrapper for splitting billion-scale dataset groundtruth ([#1679](https://github.com/rapidsai/raft/pull/1679)) [@divyegala](https://github.com/divyegala)
+- Rename CAGRA parameter num_parents to search_width ([#1676](https://github.com/rapidsai/raft/pull/1676)) [@tfeher](https://github.com/tfeher)
+- Renaming namespaces to promote CAGRA from experimental ([#1666](https://github.com/rapidsai/raft/pull/1666)) [@cjnolet](https://github.com/cjnolet)
+- CAGRA Python wrappers ([#1665](https://github.com/rapidsai/raft/pull/1665)) [@dantegd](https://github.com/dantegd)
+- Add notebook for Vector Search - Question Retrieval ([#1662](https://github.com/rapidsai/raft/pull/1662)) [@lowener](https://github.com/lowener)
+- Fix CMake CUDA support for pylibraft when raft is found. ([#1659](https://github.com/rapidsai/raft/pull/1659)) [@bdice](https://github.com/bdice)
+- Cagra ANN benchmark improvements ([#1658](https://github.com/rapidsai/raft/pull/1658)) [@tfeher](https://github.com/tfeher)
+- ANN-benchmarks: avoid using the dataset during search when possible ([#1657](https://github.com/rapidsai/raft/pull/1657)) [@achirkin](https://github.com/achirkin)
+- Revert CUDA 12.0 CI workflows to branch-23.08. ([#1652](https://github.com/rapidsai/raft/pull/1652)) [@bdice](https://github.com/bdice)
+- ANN: Optimize host-side refine ([#1651](https://github.com/rapidsai/raft/pull/1651)) [@achirkin](https://github.com/achirkin)
+- Cagra template instantiations ([#1650](https://github.com/rapidsai/raft/pull/1650)) [@tfeher](https://github.com/tfeher)
+- Modify comm_split to avoid ucp ([#1649](https://github.com/rapidsai/raft/pull/1649)) [@ChuckHastings](https://github.com/ChuckHastings)
+- Stop using setup.py in build.sh ([#1645](https://github.com/rapidsai/raft/pull/1645)) [@vyasr](https://github.com/vyasr)
+- IVF-PQ: Add a (faster) direct conversion fp8->half ([#1644](https://github.com/rapidsai/raft/pull/1644)) [@achirkin](https://github.com/achirkin)
+- Simplify `bench/ann` scripts to Python based module ([#1642](https://github.com/rapidsai/raft/pull/1642)) [@divyegala](https://github.com/divyegala)
+- Further removal of uses-setup-env-vars ([#1639](https://github.com/rapidsai/raft/pull/1639)) [@dantegd](https://github.com/dantegd)
+- Drop blank line in `raft-dask/meta.yaml` ([#1637](https://github.com/rapidsai/raft/pull/1637)) [@jakirkham](https://github.com/jakirkham)
+- Enable conservative memory allocations for RAFT IVF-Flat benchmarks. ([#1634](https://github.com/rapidsai/raft/pull/1634)) [@tfeher](https://github.com/tfeher)
+- [FEA] Codepacking for IVF-flat ([#1632](https://github.com/rapidsai/raft/pull/1632)) [@tarang-jain](https://github.com/tarang-jain)
+- Fixing ann bench cmake (and docs) ([#1630](https://github.com/rapidsai/raft/pull/1630)) [@cjnolet](https://github.com/cjnolet)
+- [WIP] Test  CI issues ([#1626](https://github.com/rapidsai/raft/pull/1626)) [@VibhuJawa](https://github.com/VibhuJawa)
+- Set pool memory resource for raft IVF ANN benchmarks ([#1625](https://github.com/rapidsai/raft/pull/1625)) [@tfeher](https://github.com/tfeher)
+- Adding sort option to matrix::select_k api ([#1615](https://github.com/rapidsai/raft/pull/1615)) [@cjnolet](https://github.com/cjnolet)
+- CAGRA max_queries auto configuration ([#1613](https://github.com/rapidsai/raft/pull/1613)) [@enp1s0](https://github.com/enp1s0)
+- Use exceptions instead of `exit(-1)` ([#1594](https://github.com/rapidsai/raft/pull/1594)) [@benfred](https://github.com/benfred)
+- [REVIEW] Add scheduler_file argument to support MNMG setup ([#1593](https://github.com/rapidsai/raft/pull/1593)) [@VibhuJawa](https://github.com/VibhuJawa)
+- Rename the CAGRA prune function to optimize ([#1588](https://github.com/rapidsai/raft/pull/1588)) [@enp1s0](https://github.com/enp1s0)
+- This PR adds support to __half and nb_bfloat16 to myAtomicReduce ([#1585](https://github.com/rapidsai/raft/pull/1585)) [@Kh4ster](https://github.com/Kh4ster)
+- [IMP] move core CUDA RT macros to cuda_rt_essentials.hpp ([#1584](https://github.com/rapidsai/raft/pull/1584)) [@MatthiasKohl](https://github.com/MatthiasKohl)
+- preprocessor syntax fix ([#1582](https://github.com/rapidsai/raft/pull/1582)) [@AyodeAwe](https://github.com/AyodeAwe)
+- use rapids-upload-docs script ([#1578](https://github.com/rapidsai/raft/pull/1578)) [@AyodeAwe](https://github.com/AyodeAwe)
+- Unpin `dask` and `distributed` for development and fix `merge_labels` test ([#1574](https://github.com/rapidsai/raft/pull/1574)) [@galipremsagar](https://github.com/galipremsagar)
+- Remove documentation build scripts for Jenkins ([#1570](https://github.com/rapidsai/raft/pull/1570)) [@ajschmidt8](https://github.com/ajschmidt8)
+- Add support to __half and nv_bfloat16 to most math functions ([#1554](https://github.com/rapidsai/raft/pull/1554)) [@Kh4ster](https://github.com/Kh4ster)
+- Add RAFT ANN benchmark for CAGRA ([#1552](https://github.com/rapidsai/raft/pull/1552)) [@enp1s0](https://github.com/enp1s0)
+- Update CAGRA knn_graph_sort to use Raft::bitonic_sort ([#1550](https://github.com/rapidsai/raft/pull/1550)) [@enp1s0](https://github.com/enp1s0)
+- Add identity matrix function ([#1548](https://github.com/rapidsai/raft/pull/1548)) [@lowener](https://github.com/lowener)
+- Unpin scikit-build upper bound ([#1547](https://github.com/rapidsai/raft/pull/1547)) [@vyasr](https://github.com/vyasr)
+- Migrate wheel workflow scripts locally ([#1546](https://github.com/rapidsai/raft/pull/1546)) [@divyegala](https://github.com/divyegala)
+- Add sample filtering for ivf_flat. Filtering code refactoring and cleanup ([#1541](https://github.com/rapidsai/raft/pull/1541)) [@alexanderguzhva](https://github.com/alexanderguzhva)
+- CAGRA pad dataset for 128bit vectorized load ([#1505](https://github.com/rapidsai/raft/pull/1505)) [@tfeher](https://github.com/tfeher)
+- Sparse Pairwise Distances API Updates ([#1502](https://github.com/rapidsai/raft/pull/1502)) [@divyegala](https://github.com/divyegala)
+- Add CAGRA gbench ([#1496](https://github.com/rapidsai/raft/pull/1496)) [@tfeher](https://github.com/tfeher)
+- Cagra index construction without copying device mdarrays ([#1494](https://github.com/rapidsai/raft/pull/1494)) [@tfeher](https://github.com/tfeher)
+
 # raft 23.06.00 (7 Jun 2023)
 
 ## 🚨 Breaking Changes

README.md

@@ -318,6 +318,7 @@ The [build](docs/source/build.md) instructions contain more details on building
 
 The folder structure mirrors other RAPIDS repos, with the following folders:
 
+- `bench/ann`: Python scripts for running ANN benchmarks
 - `ci`: Scripts for running CI in PRs
 - `conda`: Conda recipes and development conda environments
 - `cpp`: Source code for C++ libraries.

scripts/ann-benchmarks/algos.yaml → bench/ann/algos.yaml

@@ -1,4 +1,4 @@
-faise_gpu_ivf_flat:
+faiss_gpu_ivf_flat:
   executable: FAISS_IVF_FLAT_ANN_BENCH
   disabled: false
 faiss_gpu_flat:

scripts/ann-benchmarks/data_export.py → bench/ann/data_export.py

@@ -19,18 +19,18 @@
 
 
 def export_results(output_filepath, recompute, groundtruth_filepath,
-                   result_filepaths):
+                   result_filepath):
     print(f"Writing output file to: {output_filepath}")
     ann_bench_scripts_dir = os.path.join(os.getenv("RAFT_HOME"),
                                          "cpp/bench/ann/scripts")
     ann_bench_scripts_path = os.path.join(ann_bench_scripts_dir,
                                           "eval.pl")
     if recompute:
         p = subprocess.Popen([ann_bench_scripts_path, "-f", "-o", output_filepath,
-                              groundtruth_filepath] + result_filepaths)
+                              groundtruth_filepath, result_filepath])
     else:
         p = subprocess.Popen([ann_bench_scripts_path, "-o", output_filepath,
-                              groundtruth_filepath] + result_filepaths)
+                              groundtruth_filepath, result_filepath])
     p.wait()
 
 
@@ -41,18 +41,23 @@ def main():
                         required=True)
     parser.add_argument("--recompute", action="store_true",
                         help="Recompute metrics")
-    parser.add_argument("--groundtruth",
-                        help="Path to groundtruth.neighbors.ibin file for a dataset",
-                        required=True)
-    args, result_filepaths = parser.parse_known_args()
-
-    # if nothing is provided
-    if len(result_filepaths) == 0:
-        raise ValueError("No filepaths to results were provided")
+    parser.add_argument("--dataset",
+                        help="Name of the dataset to export results for",
+                        default="glove-100-inner")
+    parser.add_argument(
+        "--dataset-path",
+        help="path to dataset folder",
+        default=os.path.join(os.getenv("RAFT_HOME"), 
+                             "bench", "ann", "data")
+    )
+
+    args = parser.parse_args()
 
-    groundtruth_filepath = args.groundtruth
+    result_filepath = os.path.join(args.dataset_path, args.dataset, "result")
+    groundtruth_filepath = os.path.join(args.dataset_path, args.dataset, 
+                                        "groundtruth.neighbors.ibin")
     export_results(args.output, args.recompute, groundtruth_filepath,
-                   result_filepaths)
+                   result_filepath)
 
 
 if __name__ == "__main__":

scripts/ann-benchmarks/get_dataset.py → bench/ann/get_dataset.py

@@ -76,16 +76,17 @@ def download(name, normalize, ann_bench_data_path):
 def main():
     parser = argparse.ArgumentParser(
         formatter_class=argparse.ArgumentDefaultsHelpFormatter)
-    parser.add_argument("--name", help="dataset to download",
+    parser.add_argument("--dataset", help="dataset to download",
                         default="glove-100-angular")
-    parser.add_argument("--path", help="path to download dataset",
-                        default=os.path.join(os.getcwd(), "data"))
+    parser.add_argument("--dataset-path", help="path to download dataset",
+                        default=os.path.join(os.getenv("RAFT_HOME"), 
+                                             "bench", "ann", "data"))
     parser.add_argument("--normalize",
                         help="normalize cosine distance to inner product",
                         action="store_true")
     args = parser.parse_args()
 
-    download(args.name, args.normalize, args.path)
+    download(args.dataset, args.normalize, args.dataset_path)
 
 
 if __name__ == "__main__":

scripts/ann-benchmarks/plot.py → bench/ann/plot.py

@@ -208,7 +208,7 @@ def load_all_results(result_filepath):
 def main():
     parser = argparse.ArgumentParser(
         formatter_class=argparse.ArgumentDefaultsHelpFormatter)
-    parser.add_argument("--result_csv", help="Path to CSV Results", required=True)
+    parser.add_argument("--result-csv", help="Path to CSV Results", required=True)
     parser.add_argument("--output", help="Path to the PNG output file",
                         default=f"{os.getcwd()}/out.png")
     parser.add_argument(

scripts/ann-benchmarks/run.py → bench/ann/run.py

@@ -86,7 +86,17 @@ def main():
     parser.add_argument(
         "--configuration",
         help="path to configuration file for a dataset",
-        required=True
+    )
+    parser.add_argument(
+        "--dataset",
+        help="dataset whose configuration file will be used",
+        default="glove-100-inner"
+    )
+    parser.add_argument(
+        "--dataset-path",
+        help="path to dataset folder",
+        default=os.path.join(os.getenv("RAFT_HOME"), 
+                             "bench", "ann", "data")
     )
     parser.add_argument(
         "--build",
@@ -121,15 +131,29 @@ def main():
         search = args.search
 
     # Read configuration file associated to dataset
-    conf_filepath = args.configuration
+    if args.configuration:
+        conf_filepath = args.configuration
+    else:
+        conf_filepath = os.path.join(scripts_path, "conf", f"{args.dataset}.json")
     conf_filename = conf_filepath.split("/")[-1]
     conf_filedir = "/".join(conf_filepath.split("/")[:-1])
+    dataset_name = conf_filename.replace(".json", "")
+    dataset_path = os.path.join(args.dataset_path, dataset_name)
     if not os.path.exists(conf_filepath):
         raise FileNotFoundError(conf_filename)
 
     with open(conf_filepath, "r") as f:
         conf_file = json.load(f)
 
+    # Replace base, query to dataset-path
+    replacement_base_filepath = \
+        os.path.normpath(conf_file["dataset"]["base_file"]).split(os.path.sep)[-1]
+    conf_file["dataset"]["base_file"] = \
+        os.path.join(dataset_path, replacement_base_filepath)
+    replacement_query_filepath = \
+        os.path.normpath(conf_file["dataset"]["query_file"]).split(os.path.sep)[-1]
+    conf_file["dataset"]["query_file"] = \
+        os.path.join(dataset_path, replacement_query_filepath)
     # Ensure base and query files exist for dataset
     if not os.path.exists(conf_file["dataset"]["base_file"]):
         raise FileNotFoundError(conf_file["dataset"]["base_file"])
@@ -175,6 +199,14 @@ def main():
                     executables_to_run[executable_path] = {"index": []}
                 executables_to_run[executable_path]["index"].append(index)
 
+    # Replace build, search to dataset path
+    for executable_path in executables_to_run:
+        for pos, index in enumerate(executables_to_run[executable_path]["index"]):
+            index["file"] = os.path.join(dataset_path, "index", index["name"])
+            index["search_result_file"] = \
+                os.path.join(dataset_path, "result", index["name"])
+            executables_to_run[executable_path]["index"][pos] = index
+
     run_build_and_search(conf_filename, conf_file, executables_to_run,
                          args.force, conf_filedir, build, search)
 

build.sh

@@ -76,7 +76,7 @@ INSTALL_TARGET=install
 BUILD_REPORT_METRICS=""
 BUILD_REPORT_INCL_CACHE_STATS=OFF
 
-TEST_TARGETS="CLUSTER_TEST;CORE_TEST;DISTANCE_TEST;LABEL_TEST;LINALG_TEST;MATRIX_TEST;RANDOM_TEST;SOLVERS_TEST;SPARSE_TEST;SPARSE_DIST_TEST;SPARSE_NEIGHBORS_TEST;NEIGHBORS_TEST;STATS_TEST;UTILS_TEST"
+TEST_TARGETS="CLUSTER_TEST;CORE_TEST;DISTANCE_TEST;LABEL_TEST;LINALG_TEST;MATRIX_TEST;NEIGHBORS_TEST;NEIGHBORS_ANN_CAGRA_TEST;RANDOM_TEST;SOLVERS_TEST;SPARSE_TEST;SPARSE_DIST_TEST;SPARSE_NEIGHBORS_TEST;STATS_TEST;UTILS_TEST"
 BENCH_TARGETS="CLUSTER_BENCH;NEIGHBORS_BENCH;DISTANCE_BENCH;LINALG_BENCH;MATRIX_BENCH;SPARSE_BENCH;RANDOM_BENCH"
 
 CACHE_ARGS=""
@@ -315,10 +315,11 @@ if hasArg tests || (( ${NUMARGS} == 0 )); then
     # Force compile library when needed test targets are specified
     if [[ $CMAKE_TARGET == *"CLUSTER_TEST"* || \
           $CMAKE_TARGET == *"DISTANCE_TEST"* || \
+          $CMAKE_TARGET == *"MATRIX_TEST"* || \
+          $CMAKE_TARGET == *"NEIGHBORS_ANN_CAGRA_TEST"* || \
+          $CMAKE_TARGET == *"NEIGHBORS_TEST"* || \
           $CMAKE_TARGET == *"SPARSE_DIST_TEST" || \
           $CMAKE_TARGET == *"SPARSE_NEIGHBORS_TEST"* || \
-          $CMAKE_TARGET == *"MATRIX_TEST"* || \
-          $CMAKE_TARGET == *"NEIGHBORS_TEST" || \
           $CMAKE_TARGET == *"STATS_TEST"* ]]; then
       echo "-- Enabling compiled lib for gtests"
       COMPILE_LIBRARY=ON

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

@@ -19,10 +19,10 @@ dependencies:
 - cupy>=12.0.0
 - cxx-compiler
 - cython>=3.0.0
-- dask-core>=2023.5.1
+- dask-core>=2023.7.1
 - dask-cuda==23.10.*
-- dask>=2023.5.1
-- distributed>=2023.5.1
+- dask>=2023.7.1
+- distributed>=2023.7.1
 - doxygen>=1.8.20
 - gcc_linux-64=11.*
 - gmock>=1.13.0

conda/environments/all_cuda-120_arch-x86_64.yaml

@@ -19,10 +19,10 @@ dependencies:
 - cupy>=12.0.0
 - cxx-compiler
 - cython>=3.0.0
-- dask-core>=2023.5.1
+- dask-core>=2023.7.1
 - dask-cuda==23.10.*
-- dask>=2023.5.1
-- distributed>=2023.5.1
+- dask>=2023.7.1
+- distributed>=2023.7.1
 - doxygen>=1.8.20
 - gcc_linux-64=11.*
 - gmock>=1.13.0