Making cuco, thrust, and mdspan optional dependencies. (#585)

In addition to the cuco dependency, the following changes are included: 1. ability to turn off thrust and mdspan dependencies (rmm is still required) 2. compiling libraries now defaults to the same setting of `BUILD_TESTS` (tests are still enabled) 3. cuco dependency is disabled by default (unless distance component is enabled) 4. the headers which are safe to expose in public APIs are moved over to `core/` directory. Authors: - Corey J. Nolet (https://github.com/cjnolet) Approvers: - Dante Gama Dessavre (https://github.com/dantegd) - Robert Maynard (https://github.com/robertmaynard) - Mark Sadang (https://github.com/msadang) URL: #585
rapidsai · Apr 19, 2022 · d0601af · d0601af
1 parent 2f6d247
commit d0601af
Show file tree

Hide file tree

Showing 43 changed files with 3,767 additions and 3,642 deletions.
diff --git a/BUILD.md b/BUILD.md
@@ -26,12 +26,12 @@
 In addition to the libraries included with cudatoolkit 11.0+, there are some other dependencies below for building RAFT from source. Many of the dependencies are optional and depend only on the primitives being used. All of these can be installed with cmake or [rapids-cpm](https://github.com/rapidsai/rapids-cmake#cpm) and many of them can be installed with [conda](https://anaconda.org).
 
 #### Required
-- [Thrust](https://github.com/NVIDIA/thrust) v1.15 / [CUB](https://github.com/NVIDIA/cub)
 - [RMM](https://github.com/rapidsai/rmm) corresponding to RAFT version.
-- [mdspan](https://github.com/rapidsai/mdspan)
 
 #### Optional
-- [cuCollections](https://github.com/NVIDIA/cuCollections) - Used in `raft::sparse::distance` API
+- [mdspan](https://github.com/rapidsai/mdspan) - On by default but can be disabled. 
+- [Thrust](https://github.com/NVIDIA/thrust) v1.15 / [CUB](https://github.com/NVIDIA/cub) - On by default but can be disabled.
+- [cuCollections](https://github.com/NVIDIA/cuCollections) - Used in `raft::sparse::distance` API.
 - [Libcu++](https://github.com/NVIDIA/libcudacxx) v1.7.0
 - [FAISS](https://github.com/facebookresearch/faiss) v1.7.0 - Used in `raft::spatial::knn` API and needed to build tests.
 - [NCCL](https://github.com/NVIDIA/nccl) - Used in `raft::comms` API and needed to build `Pyraft`
@@ -53,6 +53,11 @@ The following example will download the needed dependencies and install the RAFT
 ./build.sh libraft --install
 ```
 
+The `--minimal-deps` flag can be used to install the headers with minimal dependencies:
+```bash
+./build.sh libraft --install --minimal-deps
+```
+
 ### <a id="shared_cpp_libs"></a>C++ Shared Libraries (optional)
 
 For larger projects which make heavy use of the pairwise distances or nearest neighbors APIs, shared libraries can be built to speed up compile times. These shared libraries can also significantly improve re-compile times both while developing RAFT and developing against the APIs. Build all of the available shared libraries by passing `--compile-libs` flag to `build.sh`:
@@ -69,7 +74,14 @@ Add the `--install` flag to the above example to also install the shared librari
 
 ### <a id="gtests"></a>Tests
 
-Compile the tests using the `tests` target in `build.sh`. By default, the shared libraries are assumed to be already built and on the library path. Add `--compile-libs` to also compile them.
+Compile the tests using the `tests` target in `build.sh`.
+
+```bash
+./build.sh libraft tests
+```
+
+Test compile times can be improved significantly by using the optional shared libraries. If installed, they will be used automatically when building the tests but `--compile-libs` can be used to add additional compilation units and compile them with the tests.
+
 ```bash
 ./build.sh libraft tests --compile-libs
 ```
@@ -110,11 +122,13 @@ RAFT's cmake has the following configurable flags available:.
 | --- | --- | --- | --- |
 | BUILD_TESTS | ON, OFF | ON | Compile Googletests |
 | BUILD_BENCH | ON, OFF | ON | Compile benchmarks |
+| raft_FIND_COMPONENTS | nn distance | | Configures the optional components as a space-separated list |
 | RAFT_COMPILE_LIBRARIES | ON, OFF | OFF | Compiles all `libraft` shared libraries (these are required for Googletests) |
-| RAFT_COMPILE_NN_LIBRARY | ON, OFF | ON | Compiles the `libraft-nn` shared library |
-| RAFT_COMPILE_DIST_LIBRARY | ON, OFF | ON | Compiles the `libraft-distance` shared library |
+| RAFT_COMPILE_NN_LIBRARY | ON, OFF | OFF | Compiles the `libraft-nn` shared library |
+| RAFT_COMPILE_DIST_LIBRARY | ON, OFF | OFF | Compiles the `libraft-distance` shared library |
 | RAFT_ENABLE_NN_DEPENDENCIES | ON, OFF | OFF | Searches for dependencies of nearest neighbors API, such as FAISS, and compiles them if not found. Needed for `raft::spatial::knn` |
-| RAFT_ENABLE_cuco_DEPENDENCY | ON, OFF | ON | Enables the cuCollections dependency used by `raft::sparse::distance` |
+| RAFT_ENABLE_thrust_DEPENDENCY | ON, OFF | ON | Enables the Thrust dependency. This can be disabled when using many simple utilities or to override with a different Thrust version. |
+| RAFT_ENABLE_mdspan_DEPENDENCY | ON, OFF | ON | Enables the std::mdspan dependency. This can be disabled when using many simple utilities. |
 | RAFT_ENABLE_nccl_DEPENDENCY | ON, OFF | OFF | Enables NCCL dependency used by `raft::comms` and needed to build `pyraft` |
 | RAFT_ENABLE_ucx_DEPENDENCY | ON, OFF | OFF | Enables UCX dependency used by `raft::comms` and needed to build `pyraft` |
 | RAFT_USE_FAISS_STATIC | ON, OFF | OFF | Statically link FAISS into `libraft-nn` | 
@@ -212,7 +226,8 @@ set(RAFT_PINNED_TAG "branch-${RAFT_VERSION}")
 function(find_and_configure_raft)
   set(oneValueArgs VERSION FORK PINNED_TAG USE_FAISS_STATIC
           COMPILE_LIBRARIES ENABLE_NN_DEPENDENCIES CLONE_ON_PIN
-          USE_NN_LIBRARY USE_DISTANCE_LIBRARY)
+          USE_NN_LIBRARY USE_DISTANCE_LIBRARY 
+          ENABLE_thrust_DEPENDENCY ENABLE_mdspan_DEPENDENCY)
   cmake_parse_arguments(PKG "${options}" "${oneValueArgs}"
                             "${multiValueArgs}" ${ARGN} )
 
@@ -256,6 +271,8 @@ function(find_and_configure_raft)
           "RAFT_ENABLE_NN_DEPENDENCIES ${PKG_ENABLE_NN_DEPENDENCIES}"
           "RAFT_USE_FAISS_STATIC ${PKG_USE_FAISS_STATIC}"
           "RAFT_COMPILE_LIBRARIES ${PKG_COMPILE_LIBRARIES}"
+          "RAFT_ENABLE_thrust_DEPENDENCY ${PKG_ENABLE_thrust_DEPENDENCY}"
+          "RAFT_ENABLE_mdspan_DEPENDENCY ${PKG_ENABLE_mdspan_DEPENDENCY}"
   )
 
 endfunction()
@@ -272,11 +289,13 @@ find_and_configure_raft(VERSION    ${RAFT_VERSION}.00
         # even if it's already installed.
         CLONE_ON_PIN     ON
 
-        COMPILE_LIBRARIES      NO
-        USE_NN_LIBRARY         NO
-        USE_DISTANCE_LIBRARY   NO
-        ENABLE_NN_DEPENDENCIES NO  # This builds FAISS if not installed
-        USE_FAISS_STATIC       NO
+        COMPILE_LIBRARIES        NO
+        USE_NN_LIBRARY           NO
+        USE_DISTANCE_LIBRARY     NO
+        ENABLE_NN_DEPENDENCIES   NO  # This builds FAISS if not installed
+        USE_FAISS_STATIC         NO
+        ENABLE_thrust_DEPENDENCY YES
+        ENABLE_mdspan_DEPENDENCY YES
 )
 ```
 

diff --git a/README.md b/README.md
@@ -1,6 +1,6 @@
 # <div align="left"><img src="https://rapids.ai/assets/images/rapids_logo.png" width="90px"/>&nbsp;RAFT: Reusable Accelerated Functions and Tools</div>
 
-RAFT contains fundamental widely-used algorithms and primitives for data science, graph and machine learning. The algorithms are CUDA-accelerated and form building-blocks for rapidly composing analytics.
+RAFT contains fundamental widely-used algorithms and primitives for data science and machine learning. The algorithms are CUDA-accelerated and form building-blocks for rapidly composing analytics.
 
 By taking a primitives-based approach to algorithm development, RAFT 
 - accelerates algorithm construction time

diff --git a/build.sh b/build.sh
@@ -18,7 +18,7 @@ ARGS=$*
 # script, and that this script resides in the repo dir!
 REPODIR=$(cd $(dirname $0); pwd)
 
-VALIDARGS="clean libraft pyraft pylibraft docs tests bench clean -v -g --install --compile-libs --compile-nn --compile-dist --allgpuarch --nvtx --show_depr_warn -h --buildfaiss"
+VALIDARGS="clean libraft pyraft pylibraft docs tests bench clean -v -g --install --compile-libs --compile-nn --compile-dist --allgpuarch --nvtx --show_depr_warn -h --buildfaiss --minimal-deps"
 HELP="$0 [<target> ...] [<flag> ...]
  where <target> is:
    clean            - remove all existing build artifacts and configuration (start over)
@@ -36,6 +36,8 @@ HELP="$0 [<target> ...] [<flag> ...]
    --compile-libs   - compile shared libraries for all components
    --compile-nn     - compile shared library for nn component
    --compile-dist   - compile shared library for distance component
+   --minimal-deps   - disables dependencies like thrust so they can be overridden.
+                      can be useful for a pure header-only install
    --allgpuarch     - build for all supported GPU architectures
    --buildfaiss     - build faiss statically into raft
    --install        - install cmake targets
@@ -62,6 +64,9 @@ COMPILE_LIBRARIES=OFF
 COMPILE_NN_LIBRARY=OFF
 COMPILE_DIST_LIBRARY=OFF
 ENABLE_NN_DEPENDENCIES=OFF
+
+ENABLE_thrust_DEPENDENCY=ON
+
 ENABLE_ucx_DEPENDENCY=OFF
 ENABLE_nccl_DEPENDENCY=OFF
 
@@ -105,6 +110,11 @@ fi
 if hasArg --install; then
   INSTALL_TARGET="install"
 fi
+
+if hasArg --minimal-deps; then
+  ENABLE_thrust_DEPENDENCY=OFF
+fi
+
 if hasArg -v; then
     VERBOSE_FLAG="-v"
     CMAKE_LOG_LEVEL="VERBOSE"
@@ -218,7 +228,8 @@ if (( ${NUMARGS} == 0 )) || hasArg libraft || hasArg docs || hasArg tests || has
           -DRAFT_COMPILE_DIST_LIBRARY=${COMPILE_DIST_LIBRARY} \
           -DRAFT_USE_FAISS_STATIC=${BUILD_STATIC_FAISS} \
           -DRAFT_ENABLE_nccl_DEPENDENCY=${ENABLE_nccl_DEPENDENCY} \
-          -DRAFT_ENABLE_ucx_DEPENDENCY=${ENABLE_ucx_DEPENDENCY}
+          -DRAFT_ENABLE_ucx_DEPENDENCY=${ENABLE_ucx_DEPENDENCY} \
+          -DRAFT_ENABLE_thrust_DEPENDENCY=${ENABLE_thrust_DEPENDENCY}
 
   if [[ ${CMAKE_TARGET} != "" ]]; then
       echo "-- Compiling targets: ${CMAKE_TARGET}, verbose=${VERBOSE_FLAG}"

diff --git a/conda/recipes/libraft_distance/meta.yaml b/conda/recipes/libraft_distance/meta.yaml
@@ -44,7 +44,6 @@ requirements:
     - ucx-py {{ ucx_py_version }}
     - ucx-proc=*=gpu
     - gtest=1.10.0
-    - gmock
     - librmm {{ minor_version }}
   run:
     - libraft-headers {{ version }}

diff --git a/conda/recipes/libraft_headers/build.sh b/conda/recipes/libraft_headers/build.sh
@@ -1,4 +1,4 @@
 #!/usr/bin/env bash
 # Copyright (c) 2022, NVIDIA CORPORATION.
 
-./build.sh libraft --install -v --allgpuarch
+./build.sh libraft --install -v --allgpuarch
diff --git a/conda/recipes/libraft_headers/meta.yaml b/conda/recipes/libraft_headers/meta.yaml
@@ -43,7 +43,6 @@ requirements:
     - ucx-py {{ ucx_py_version }}
     - ucx-proc=*=gpu
     - gtest=1.10.0
-    - gmock
     - librmm {{ minor_version}}
     - libcusolver>=11.2.1
   run:

diff --git a/conda/recipes/libraft_nn/meta.yaml b/conda/recipes/libraft_nn/meta.yaml
@@ -44,7 +44,6 @@ requirements:
     - faiss-proc=*=cuda
     - libfaiss 1.7.0 *_cuda
     - gtest=1.10.0
-    - gmock
     - librmm {{ minor_version }}
   run:
     - {{ pin_compatible('cudatoolkit', max_pin='x', min_pin='x') }}

diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
@@ -13,6 +13,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #=============================================================================
+set(RAPIDS_VERSION "22.06")
+set(RAFT_VERSION "${RAPIDS_VERSION}.00")
 
 cmake_minimum_required(VERSION 3.20.1 FATAL_ERROR)
 file(DOWNLOAD https://raw.githubusercontent.com/rapidsai/rapids-cmake/branch-22.06/RAPIDS.cmake
@@ -26,7 +28,7 @@ include(rapids-find)
 
 rapids_cuda_init_architectures(RAFT)
 
-project(RAFT VERSION 22.06.00 LANGUAGES CXX CUDA)
+project(RAFT VERSION ${RAFT_VERSION} LANGUAGES CXX CUDA)
 
 # Needed because GoogleBenchmark changes the state of FindThreads.cmake, causing subsequent runs to
 # have different values for the `Threads::Threads` target. Setting this flag ensures
@@ -55,16 +57,22 @@ 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)
 option(CUDA_STATIC_RUNTIME "Statically link the CUDA runtime" OFF)
 option(DETECT_CONDA_ENV "Enable detection of conda environment for dependencies" ON)
-option(DISABLE_DEPRECATION_WARNINGS "Disable depreaction warnings " ON)
+option(DISABLE_DEPRECATION_WARNINGS "Disable deprecaction warnings " ON)
 option(DISABLE_OPENMP "Disable OpenMP" OFF)
 option(NVTX "Enable nvtx markers" OFF)
 
-option(RAFT_COMPILE_LIBRARIES "Enable building raft shared library instantiations" ON)
+option(RAFT_COMPILE_LIBRARIES "Enable building raft shared library instantiations" ${BUILD_TESTS})
 option(RAFT_COMPILE_NN_LIBRARY "Enable building raft nearest neighbors shared library instantiations" OFF)
 option(RAFT_COMPILE_DIST_LIBRARY "Enable building raft distant shared library instantiations" OFF)
 option(RAFT_ENABLE_NN_DEPENDENCIES "Search for raft::nn dependencies like faiss" ${RAFT_COMPILE_LIBRARIES})
 
-option(RAFT_ENABLE_cuco_DEPENDENCY "Enable cuCollections dependency" ON)
+option(RAFT_ENABLE_mdspan_DEPENDENCY "Enable mdspan dependency" ON)
+option(RAFT_ENABLE_thrust_DEPENDENCY "Enable Thrust dependency" ON)
+
+if(BUILD_TESTS AND NOT RAFT_ENABLE_thrust_DEPENDENCY)
+  message(VERBOSE "RAFT: BUILD_TESTS is enabled, overriding RAFT_ENABLE_thrust_DEPENDENCY")
+  set(RAFT_ENABLE_thrust_DEPENDENCY ON)
+endif()
 
 # Currently, UCX and NCCL are only needed to build Pyraft and so a simple find_package() is sufficient
 option(RAFT_ENABLE_nccl_DEPENDENCY "Enable NCCL dependency" OFF)
@@ -75,6 +83,7 @@ option(RAFT_EXCLUDE_FAISS_FROM_ALL "Exclude FAISS targets from RAFT's 'all' targ
 include(CMakeDependentOption)
 cmake_dependent_option(RAFT_USE_FAISS_STATIC "Build and statically link the FAISS library for nearest neighbors search on GPU" ON RAFT_COMPILE_LIBRARIES OFF)
 
+message(VERBOSE "RAFT: Building optional components: ${raft_FIND_COMPONENTS}")
 message(VERBOSE "RAFT: Build RAFT unit-tests: ${BUILD_TESTS}")
 message(VERBOSE "RAFT: Building raft C++ benchmarks: ${BUILD_BENCH}")
 message(VERBOSE "RAFT: Enable detection of conda environment for dependencies: ${DETECT_CONDA_ENV}")
@@ -123,6 +132,10 @@ include(cmake/modules/ConfigureCUDA.cmake)
 ##############################################################################
 # - Requirements -------------------------------------------------------------
 
+if(distance IN_LIST raft_FIND_COMPONENTS OR RAFT_COMPILE_LIBRARIES OR RAFT_COMPILE_DIST_LIBRARY)
+  set(RAFT_ENABLE_cuco_DEPENDENCY ON)
+endif()
+
 # add third party dependencies using CPM
 rapids_cpm_init()
 
@@ -151,17 +164,20 @@ target_include_directories(raft INTERFACE
         "$<BUILD_INTERFACE:${RAFT_SOURCE_DIR}/include>"
         "$<INSTALL_INTERFACE:include>")
 
+# Keep RAFT as lightweight as possible.
+# Only CUDA libs, rmm, and mdspan should
+# be used in global target.
 target_link_libraries(raft INTERFACE
-        raft::Thrust
         $<$<BOOL:${NVTX}>:CUDA::nvToolsExt>
         CUDA::cublas
         CUDA::curand
         CUDA::cusolver
         CUDA::cudart
         CUDA::cusparse
         rmm::rmm
-        $<$<BOOL:${RAFT_ENABLE_cuco_DEPENDENCY}>:cuco::cuco>
-        std::mdspan)
+        $<$<BOOL:${RAFT_ENABLE_thrust_DEPENDENCY}>:raft::Thrust>
+        $<$<BOOL:${RAFT_ENABLE_mdspan_DEPENDENCY}>:std::mdspan>
+)
 
 target_compile_definitions(raft INTERFACE $<$<BOOL:${NVTX}>:NVTX_ENABLED>)
 target_compile_features(raft INTERFACE cxx_std_17 $<BUILD_INTERFACE:cuda_std_17>)
@@ -248,6 +264,7 @@ endif()
 
 target_link_libraries(raft_distance INTERFACE
     raft::raft
+    $<$<BOOL:${RAFT_ENABLE_cuco_DEPENDENCY}>:cuco::cuco>
     $<TARGET_NAME_IF_EXISTS:raft_distance_lib>
     $<TARGET_NAME_IF_EXISTS:raft::raft_distance_lib>
 )
@@ -301,6 +318,7 @@ endif()
 
 target_link_libraries(raft_nn INTERFACE
     raft::raft
+    $<$<BOOL:${RAFT_ENABLE_NN_DEPENDENCIES}>:faiss::faiss>
     $<TARGET_NAME_IF_EXISTS:raft_nn_lib>
     $<TARGET_NAME_IF_EXISTS:raft::raft_nn_lib>)
 
@@ -341,14 +359,17 @@ install(DIRECTORY include/raft
 install(FILES include/raft.hpp
         DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/raft)
 
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/include/raft/version_config.hpp
+        DESTINATION include/raft)
+
 ##############################################################################
 # - install export -----------------------------------------------------------
 set(doc_string
         [=[
 Provide targets for the RAFT: Reusable Accelerated Functions and Tools
 
 RAFT contains fundamental widely-used algorithms and primitives
-for data science, graph, and ml.
+for data science and machine learning.
 
 Optional Components:
   - nn
@@ -361,13 +382,18 @@ Imported Targets:
 
 ]=])
 
-set(code_string
-[=[
-
-if(NOT TARGET raft::Thrust)
-  thrust_create_target(raft::Thrust FROM_OPTIONS)
+set(code_string )
+if(RAFT_ENABLE_thrust_DEPENDENCY)
+  string(APPEND code_string
+  [=[
+  if(NOT TARGET raft::Thrust)
+    thrust_create_target(raft::Thrust FROM_OPTIONS)
+  endif()
+  ]=])
 endif()
 
+string(APPEND code_string
+[=[
 if(distance IN_LIST raft_FIND_COMPONENTS)
   enable_language(CUDA)
 endif()
@@ -381,8 +407,7 @@ if(nn IN_LIST raft_FIND_COMPONENTS)
       add_library(faiss ALIAS faiss::faiss)
   endif()
 endif()
-]=]
-        )
+]=])
 
 # Use `rapids_export` for 22.04 as it will have COMPONENT support
 include(cmake/modules/raft_export.cmake)

diff --git a/cpp/cmake/thirdparty/get_cuco.cmake b/cpp/cmake/thirdparty/get_cuco.cmake
@@ -16,21 +16,20 @@
 
 function(find_and_configure_cuco VERSION)
 
-    if(RAFT_ENABLE_cuco_DEPENDENCY)
-        rapids_cpm_find(cuco ${VERSION}
-          GLOBAL_TARGETS      cuco::cuco
-          BUILD_EXPORT_SET    raft-exports
-          INSTALL_EXPORT_SET  raft-exports
-          CPM_ARGS
-            GIT_REPOSITORY https://github.com/NVIDIA/cuCollections.git
-            GIT_TAG        fb58a38701f1c24ecfe07d8f1f208bbe80930da5
-            OPTIONS        "BUILD_TESTS OFF"
-                           "BUILD_BENCHMARKS OFF"
-                           "BUILD_EXAMPLES OFF"
-        )
-    endif()
-
+    rapids_cpm_find(cuco ${VERSION}
+      GLOBAL_TARGETS      cuco::cuco
+      BUILD_EXPORT_SET    raft-distance-exports
+      INSTALL_EXPORT_SET  raft-distance-exports
+      CPM_ARGS
+        GIT_REPOSITORY https://github.com/NVIDIA/cuCollections.git
+        GIT_TAG        6ec8b6dcdeceea07ab4456d32461a05c18864411
+        OPTIONS        "BUILD_TESTS OFF"
+                       "BUILD_BENCHMARKS OFF"
+                       "BUILD_EXAMPLES OFF"
+    )
 endfunction()
 
-# cuCollections doesn't have a version yet
-find_and_configure_cuco(0.0.1)
+if(RAFT_ENABLE_cuco_DEPENDENCY)
+    # cuCollections doesn't have a version yet
+    find_and_configure_cuco(0.0.1)
+endif()
diff --git a/cpp/cmake/thirdparty/get_libcudacxx.cmake b/cpp/cmake/thirdparty/get_libcudacxx.cmake
@@ -14,11 +14,13 @@
 
 # This function finds libcudacxx and sets any additional necessary environment variables.
 function(find_and_configure_libcudacxx)
+
   include(${rapids-cmake-dir}/cpm/libcudacxx.cmake)
 
   rapids_cpm_libcudacxx(BUILD_EXPORT_SET raft-exports
                         INSTALL_EXPORT_SET raft-exports)
-
 endfunction()
 
-find_and_configure_libcudacxx()
+if(RAFT_ENABLE_cuco_DEPENDENCY)
+  find_and_configure_libcudacxx()
+endif()