Merge branch 'branch-23.12' into fea/add_ci_checks_for_kernel_visibility

rapidsai · Nov 2, 2023 · 2e73846 · 2e73846
2 parents ced48f8 + 32a6026
commit 2e73846
Show file tree

Hide file tree

Showing 53 changed files with 922 additions and 248 deletions.
diff --git a/VERSION b/VERSION
@@ -0,0 +1 @@
+23.12.00
diff --git a/ci/build_cpp.sh b/ci/build_cpp.sh
@@ -9,8 +9,10 @@ export CMAKE_GENERATOR=Ninja
 
 rapids-print-env
 
+version=$(rapids-generate-version)
+
 rapids-logger "Begin cpp build"
 
-rapids-conda-retry mambabuild conda/recipes/libraft
+RAPIDS_PACKAGE_VERSION=${version} rapids-conda-retry mambabuild conda/recipes/libraft
 
 rapids-upload-conda-to-s3 cpp
diff --git a/ci/build_python.sh b/ci/build_python.sh
@@ -13,6 +13,17 @@ rapids-logger "Begin py build"
 
 CPP_CHANNEL=$(rapids-download-conda-from-s3 cpp)
 
+version=$(rapids-generate-version)
+git_commit=$(git rev-parse HEAD)
+export RAPIDS_PACKAGE_VERSION=${version} 
+echo "${version}" > VERSION
+
+package_dir="python"
+for package_name in pylibraft raft-dask; do 
+  underscore_package_name=$(echo "${package_name}" | tr "-" "_")
+  sed -i "/^__git_commit__/ s/= .*/= \"${git_commit}\"/g" "${package_dir}/${package_name}/${underscore_package_name}/_version.py"
+done
+
 # TODO: Remove `--no-test` flags once importing on a CPU
 # node works correctly
 rapids-conda-retry mambabuild \

diff --git a/ci/build_wheel.sh b/ci/build_wheel.sh
@@ -10,9 +10,8 @@ underscore_package_name=$(echo "${package_name}" | tr "-" "_")
 source rapids-configure-sccache
 source rapids-date-string
 
-# Use gha-tools rapids-pip-wheel-version to generate wheel version then
-# update the necessary files
-version_override="$(rapids-pip-wheel-version ${RAPIDS_DATE_STRING})"
+version=$(rapids-generate-version)
+git_commit=$(git rev-parse HEAD)
 
 RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
 
@@ -22,9 +21,11 @@ PACKAGE_CUDA_SUFFIX="-${RAPIDS_PY_CUDA_SUFFIX}"
 
 # Patch project metadata files to include the CUDA version suffix and version override.
 pyproject_file="${package_dir}/pyproject.toml"
+version_file="${package_dir}/${underscore_package_name}/_version.py"
 
-sed -i "s/^version = .*/version = \"${version_override}\"/g" ${pyproject_file}
 sed -i "s/name = \"${package_name}\"/name = \"${package_name}${PACKAGE_CUDA_SUFFIX}\"/g" ${pyproject_file}
+echo "${version}" > VERSION
+sed -i "/^__git_commit__ / s/= .*/= \"${git_commit}\"/g" ${version_file}
 
 # For nightlies we want to ensure that we're pulling in alphas as well. The
 # easiest way to do so is to augment the spec with a constraint containing a

diff --git a/ci/release/update-version.sh b/ci/release/update-version.sh
@@ -43,9 +43,8 @@ sed_runner 's/'"pylibraft_version .*)"'/'"pylibraft_version ${NEXT_FULL_TAG})"'/
 sed_runner 's/'"raft_dask_version .*)"'/'"raft_dask_version ${NEXT_FULL_TAG})"'/g' python/raft-dask/CMakeLists.txt
 sed_runner 's/'"branch-.*\/RAPIDS.cmake"'/'"branch-${NEXT_SHORT_TAG}\/RAPIDS.cmake"'/g' fetch_rapids.cmake
 
-# Python __init__.py updates
-sed_runner "s/__version__ = .*/__version__ = \"${NEXT_FULL_TAG}\"/g" python/pylibraft/pylibraft/__init__.py
-sed_runner "s/__version__ = .*/__version__ = \"${NEXT_FULL_TAG}\"/g" python/raft-dask/raft_dask/__init__.py
+# Centralized version file update
+echo "${NEXT_FULL_TAG}" > VERSION
 
 # Wheel testing script
 sed_runner "s/branch-.*/branch-${NEXT_SHORT_TAG}/g" ci/test_wheel_raft_dask.sh

diff --git a/conda/environments/all_cuda-118_arch-aarch64.yaml b/conda/environments/all_cuda-118_arch-aarch64.yaml
@@ -12,6 +12,7 @@ dependencies:
 - clang-tools=16.0.6
 - clang==16.0.6
 - cmake>=3.26.4
+- cuda-nvtx=11.8
 - cuda-profiler-api=11.8.86
 - cuda-python>=11.7.1,<12.0a0
 - cuda-version=11.8

diff --git a/conda/environments/all_cuda-118_arch-x86_64.yaml b/conda/environments/all_cuda-118_arch-x86_64.yaml
@@ -12,6 +12,7 @@ dependencies:
 - clang-tools=16.0.6
 - clang==16.0.6
 - cmake>=3.26.4
+- cuda-nvtx=11.8
 - cuda-profiler-api=11.8.86
 - cuda-python>=11.7.1,<12.0a0
 - cuda-version=11.8

diff --git a/conda/environments/all_cuda-120_arch-aarch64.yaml b/conda/environments/all_cuda-120_arch-aarch64.yaml
@@ -14,6 +14,7 @@ dependencies:
 - cmake>=3.26.4
 - cuda-cudart-dev
 - cuda-nvcc
+- cuda-nvtx-dev
 - cuda-profiler-api
 - cuda-python>=12.0,<13.0a0
 - cuda-version=12.0

diff --git a/conda/environments/all_cuda-120_arch-x86_64.yaml b/conda/environments/all_cuda-120_arch-x86_64.yaml
@@ -14,6 +14,7 @@ dependencies:
 - cmake>=3.26.4
 - cuda-cudart-dev
 - cuda-nvcc
+- cuda-nvtx-dev
 - cuda-profiler-api
 - cuda-python>=12.0,<13.0a0
 - cuda-version=12.0

diff --git a/conda/environments/bench_ann_cuda-118_arch-aarch64.yaml b/conda/environments/bench_ann_cuda-118_arch-aarch64.yaml
@@ -12,6 +12,7 @@ dependencies:
 - clang-tools=16.0.6
 - clang==16.0.6
 - cmake>=3.26.4
+- cuda-nvtx=11.8
 - cuda-profiler-api=11.8.86
 - cuda-version=11.8
 - cudatoolkit

diff --git a/conda/environments/bench_ann_cuda-118_arch-x86_64.yaml b/conda/environments/bench_ann_cuda-118_arch-x86_64.yaml
@@ -12,6 +12,7 @@ dependencies:
 - clang-tools=16.0.6
 - clang==16.0.6
 - cmake>=3.26.4
+- cuda-nvtx=11.8
 - cuda-profiler-api=11.8.86
 - cuda-version=11.8
 - cudatoolkit

diff --git a/conda/environments/bench_ann_cuda-120_arch-aarch64.yaml b/conda/environments/bench_ann_cuda-120_arch-aarch64.yaml
@@ -0,0 +1,40 @@
+# This file is generated by `rapids-dependency-file-generator`.
+# To make changes, edit ../../dependencies.yaml and run `rapids-dependency-file-generator`.
+channels:
+- rapidsai
+- rapidsai-nightly
+- dask/label/dev
+- conda-forge
+- nvidia
+dependencies:
+- benchmark>=1.8.2
+- c-compiler
+- clang-tools=16.0.6
+- clang==16.0.6
+- cmake>=3.26.4
+- cuda-cudart-dev
+- cuda-nvcc
+- cuda-nvtx-dev
+- cuda-profiler-api
+- cuda-version=12.0
+- cxx-compiler
+- cython>=3.0.0
+- gcc_linux-aarch64=11.*
+- glog>=0.6.0
+- h5py>=3.8.0
+- hnswlib=0.7.0
+- libcublas-dev
+- libcurand-dev
+- libcusolver-dev
+- libcusparse-dev
+- matplotlib
+- nccl>=2.9.9
+- ninja
+- nlohmann_json>=3.11.2
+- openblas
+- pandas
+- pyyaml
+- rmm==23.12.*
+- scikit-build>=0.13.1
+- sysroot_linux-aarch64==2.17
+name: bench_ann_cuda-120_arch-aarch64
diff --git a/conda/environments/bench_ann_cuda-120_arch-x86_64.yaml b/conda/environments/bench_ann_cuda-120_arch-x86_64.yaml
@@ -0,0 +1,40 @@
+# This file is generated by `rapids-dependency-file-generator`.
+# To make changes, edit ../../dependencies.yaml and run `rapids-dependency-file-generator`.
+channels:
+- rapidsai
+- rapidsai-nightly
+- dask/label/dev
+- conda-forge
+- nvidia
+dependencies:
+- benchmark>=1.8.2
+- c-compiler
+- clang-tools=16.0.6
+- clang==16.0.6
+- cmake>=3.26.4
+- cuda-cudart-dev
+- cuda-nvcc
+- cuda-nvtx-dev
+- cuda-profiler-api
+- cuda-version=12.0
+- cxx-compiler
+- cython>=3.0.0
+- gcc_linux-64=11.*
+- glog>=0.6.0
+- h5py>=3.8.0
+- hnswlib=0.7.0
+- libcublas-dev
+- libcurand-dev
+- libcusolver-dev
+- libcusparse-dev
+- matplotlib
+- nccl>=2.9.9
+- ninja
+- nlohmann_json>=3.11.2
+- openblas
+- pandas
+- pyyaml
+- rmm==23.12.*
+- scikit-build>=0.13.1
+- sysroot_linux-64==2.17
+name: bench_ann_cuda-120_arch-x86_64
diff --git a/conda/recipes/libraft/meta.yaml b/conda/recipes/libraft/meta.yaml
@@ -2,7 +2,7 @@
 
 # Usage:
 #   conda build . -c conda-forge -c nvidia -c rapidsai
-{% set version = environ.get('GIT_DESCRIBE_TAG', '0.0.0.dev').lstrip('v') %}
+{% set version = environ['RAPIDS_PACKAGE_VERSION'].lstrip('v') + environ.get('VERSION_SUFFIX', '') %}
 {% set minor_version = version.split('.')[0] + '.' + version.split('.')[1] %}
 {% set cuda_version = '.'.join(environ['RAPIDS_CUDA_VERSION'].split('.')[:2]) %}
 {% set cuda_major = cuda_version.split('.')[0] %}
@@ -13,7 +13,7 @@ package:
   name: libraft-split
 
 source:
-  git_url: ../../..
+  path: ../../..
 
 outputs:
   - name: libraft-headers-only

diff --git a/conda/recipes/pylibraft/meta.yaml b/conda/recipes/pylibraft/meta.yaml
@@ -2,7 +2,7 @@
 
 # Usage:
 #   conda build . -c conda-forge -c numba -c rapidsai -c pytorch
-{% set version = environ.get('GIT_DESCRIBE_TAG', '0.0.0.dev').lstrip('v') %}
+{% set version = environ['RAPIDS_PACKAGE_VERSION'].lstrip('v') + environ.get('VERSION_SUFFIX', '') %}
 {% set minor_version = version.split('.')[0] + '.' + version.split('.')[1] %}
 {% set py_version = environ['CONDA_PY'] %}
 {% set cuda_version = '.'.join(environ['RAPIDS_CUDA_VERSION'].split('.')[:2]) %}
@@ -14,7 +14,7 @@ package:
   version: {{ version }}
 
 source:
-  git_url: ../../..
+  path: ../../..
 
 build:
   number: {{ GIT_DESCRIBE_NUMBER }}

diff --git a/conda/recipes/raft-ann-bench-cpu/meta.yaml b/conda/recipes/raft-ann-bench-cpu/meta.yaml
@@ -2,7 +2,7 @@
 
 # Usage:
 #   conda build . -c conda-forge -c nvidia -c rapidsai
-{% set version = environ.get('GIT_DESCRIBE_TAG', '0.0.0.dev').lstrip('v') %}
+{% set version = environ['RAPIDS_PACKAGE_VERSION'].lstrip('v') + environ.get('VERSION_SUFFIX', '') %}
 {% set minor_version = version.split('.')[0] + '.' + version.split('.')[1] %}
 {% set py_version = environ['CONDA_PY'] %}
 {% set cuda_version = '.'.join(environ['RAPIDS_CUDA_VERSION'].split('.')[:2]) %}
@@ -14,7 +14,7 @@ package:
   script: build.sh
 
 source:
-  git_url: ../../..
+  path: ../../..
 
 build:
   script_env:

diff --git a/conda/recipes/raft-ann-bench/meta.yaml b/conda/recipes/raft-ann-bench/meta.yaml
@@ -2,7 +2,7 @@
 
 # Usage:
 #   conda build . -c conda-forge -c nvidia -c rapidsai
-{% set version = environ.get('GIT_DESCRIBE_TAG', '0.0.0.dev').lstrip('v') %}
+{% set version = environ['RAPIDS_PACKAGE_VERSION'].lstrip('v') + environ.get('VERSION_SUFFIX', '') %}
 {% set minor_version = version.split('.')[0] + '.' + version.split('.')[1] %}
 {% set py_version = environ['CONDA_PY'] %}
 {% set cuda_version = '.'.join(environ['RAPIDS_CUDA_VERSION'].split('.')[:2]) %}
@@ -16,7 +16,7 @@ package:
   script: build.sh
 
 source:
-  git_url: ../../..
+  path: ../../..
 
 build:
   script_env:

diff --git a/conda/recipes/raft-dask/meta.yaml b/conda/recipes/raft-dask/meta.yaml
@@ -2,7 +2,7 @@
 
 # Usage:
 #   conda build . -c conda-forge -c numba -c rapidsai -c pytorch
-{% set version = environ.get('GIT_DESCRIBE_TAG', '0.0.0.dev').lstrip('v') %}
+{% set version = environ['RAPIDS_PACKAGE_VERSION'].lstrip('v') + environ.get('VERSION_SUFFIX', '') %}
 {% set minor_version = version.split('.')[0] + '.' + version.split('.')[1] %}
 {% set py_version = environ['CONDA_PY'] %}
 {% set cuda_version = '.'.join(environ['RAPIDS_CUDA_VERSION'].split('.')[:2]) %}
@@ -14,7 +14,7 @@ package:
   version: {{ version }}
 
 source:
-  git_url: ../../..
+  path: ../../..
 
 build:
   number: {{ GIT_DESCRIBE_NUMBER }}

diff --git a/cpp/bench/ann/src/common/ann_types.hpp b/cpp/bench/ann/src/common/ann_types.hpp
@@ -120,7 +120,7 @@ class ANN : public AnnBase {
   // The advantage of this way is that index has smaller size
   // and many indices can share one dataset.
   //
-  // AlgoProperty::need_dataset_when_search of such algorithm should be true,
+  // SearchParam::needs_dataset() of such algorithm should be true,
   // 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.

diff --git a/cpp/bench/ann/src/common/benchmark.hpp b/cpp/bench/ann/src/common/benchmark.hpp
@@ -215,11 +215,13 @@ void bench_search(::benchmark::State& state,
       search_param->metric_objective = metric_objective;
     } catch (const std::exception& e) {
       state.SkipWithError("Failed to create an algo: " + std::string(e.what()));
+      return;
     }
-    algo->set_search_param(*search_param);
+
     auto algo_property = parse_algo_property(algo->get_preference(), sp_json);
     current_algo_props = std::make_shared<AlgoProperty>(algo_property.dataset_memory_type,
                                                         algo_property.query_memory_type);
+
     if (search_param->needs_dataset()) {
       try {
         algo->set_search_dataset(dataset->base_set(current_algo_props->dataset_memory_type),
@@ -231,6 +233,14 @@ void bench_search(::benchmark::State& state,
         return;
       }
     }
+
+    try {
+      algo->set_search_param(*search_param);
+
+    } catch (const std::exception& ex) {
+      state.SkipWithError("An error occurred setting search parameters: " + std::string(ex.what()));
+      return;
+    }
   }
 
   const auto algo_property = *current_algo_props;

diff --git a/cpp/bench/ann/src/faiss/faiss_cpu_wrapper.h b/cpp/bench/ann/src/faiss/faiss_cpu_wrapper.h
@@ -152,6 +152,7 @@ void FaissCpu<T>::build(const T* dataset, size_t nrow, cudaStream_t stream)
   index_->train(nrow, dataset);  // faiss::IndexFlat::train() will do nothing
   assert(index_->is_trained);
   index_->add(nrow, dataset);
+  index_refine_ = std::make_unique<faiss::IndexRefineFlat>(this->index_.get(), dataset);
 }
 
 template <typename T>
@@ -163,7 +164,6 @@ void FaissCpu<T>::set_search_param(const AnnSearchParam& param)
   dynamic_cast<faiss::IndexIVF*>(index_.get())->nprobe = nprobe;
 
   if (search_param.refine_ratio > 1.0) {
-    this->index_refine_ = std::make_unique<faiss::IndexRefineFlat>(this->index_.get());
     this->index_refine_.get()->k_factor = search_param.refine_ratio;
   }