Adding base header-only conda package without cuda math libs (#1386)

cc @MatthiasKohl  @bdice 

Making sure CI agrees w/ this change. @MatthiasKohl, if CI succeeds here let's try to plug the resulting conda packages into a cugraph-ops PR to make sure cugraph-ops CI is happy as well.

Authors:
  - Corey J. Nolet (https://github.com/cjnolet)
  - Robert Maynard (https://github.com/robertmaynard)

Approvers:
  - AJ Schmidt (https://github.com/ajschmidt8)
  - Divye Gala (https://github.com/divyegala)

URL: #1386

build.sh

@@ -522,7 +522,7 @@ fi
 # Initiate build for example RAFT application template (if needed)
 
 if hasArg template; then
-    pushd cpp/template
+    pushd ${REPODIR}/cpp/template
     ./build.sh
     popd
 fi

conda/recipes/libraft/meta.yaml

@@ -16,7 +16,7 @@ source:
   git_url: ../../..
 
 outputs:
-  - name: libraft-headers
+  - name: libraft-headers-only
     version: {{ version }}
     script: build_libraft_headers.sh
     build:
@@ -50,20 +50,26 @@ outputs:
         - ninja
         - sysroot_{{ target_platform }} {{ sysroot_version }}
       host:
-        - cuda-profiler-api {{ cuda_profiler_api_host_version }}
-        - cudatoolkit ={{ cuda_version }}
-        - libcublas {{ libcublas_host_version }}
-        - libcublas-dev {{ libcublas_host_version }}
-        - libcurand {{ libcurand_host_version }}
-        - libcurand-dev {{ libcurand_host_version }}
-        - libcusolver {{ libcusolver_host_version }}
-        - libcusolver-dev {{ libcusolver_host_version }}
-        - libcusparse {{ libcusparse_host_version }}
-        - libcusparse-dev {{ libcusparse_host_version }}
         - librmm ={{ minor_version }}
+        - cudatoolkit {{ cuda_version }}
+    about:
+      home: https://rapids.ai/
+      license: Apache-2.0
+      summary: libraft-headers-only library
+  - name: libraft-headers
+    version: {{ version }}
+    build:
+      number: {{ GIT_DESCRIBE_NUMBER }}
+      string: cuda{{ cuda_major }}_{{ date_string }}_{{ GIT_DESCRIBE_HASH }}_{{ GIT_DESCRIBE_NUMBER }}
+      ignore_run_exports_from:
+        - {{ compiler('cuda') }}
+        - librmm
+    requirements:
       run:
-        - {{ pin_compatible('cudatoolkit', max_pin='x', min_pin='x') }}
+        - {{ pin_subpackage('libraft-headers-only', exact=True) }}
         - cuda-profiler-api {{ cuda_profiler_api_run_version }}
+        - cudatoolkit {{ cuda_version }}
+        - librmm ={{ minor_version }}
         - libcublas {{ libcublas_run_version }}
         - libcublas-dev {{ libcublas_run_version }}
         - libcurand {{ libcurand_run_version }}
@@ -72,7 +78,6 @@ outputs:
         - libcusolver-dev {{ libcusolver_run_version }}
         - libcusparse {{ libcusparse_run_version }}
         - libcusparse-dev {{ libcusparse_run_version }}
-        - librmm ={{ minor_version }}
     about:
       home: https://rapids.ai/
       license: Apache-2.0
@@ -130,7 +135,6 @@ outputs:
         - sysroot_{{ target_platform }} {{ sysroot_version }}
       host:
         - {{ pin_subpackage('libraft', exact=True) }}
-        - {{ pin_subpackage('libraft-headers', exact=True) }}
         - cuda-profiler-api {{ cuda_profiler_api_host_version }}
         - gmock {{ gtest_version }}
         - gtest {{ gtest_version }}
@@ -144,7 +148,6 @@ outputs:
         - libcusparse-dev {{ libcusparse_host_version }}
       run:
         - {{ pin_subpackage('libraft', exact=True) }}
-        - {{ pin_subpackage('libraft-headers', exact=True) }}
         - gmock {{ gtest_version }}
         - gtest {{ gtest_version }}
     about:
@@ -170,19 +173,10 @@ outputs:
         - sysroot_{{ target_platform }} {{ sysroot_version }}
       host:
         - {{ pin_subpackage('libraft', exact=True) }}
-        - {{ pin_subpackage('libraft-headers', exact=True) }}
-        - cuda-profiler-api {{ cuda_profiler_api_host_version }}
         - libcublas {{ libcublas_host_version }}
         - libcublas-dev {{ libcublas_host_version }}
-        - libcurand {{ libcurand_host_version }}
-        - libcurand-dev {{ libcurand_host_version }}
-        - libcusolver {{ libcusolver_host_version }}
-        - libcusolver-dev {{ libcusolver_host_version }}
-        - libcusparse {{ libcusparse_host_version }}
-        - libcusparse-dev {{ libcusparse_host_version }}
       run:
         - {{ pin_subpackage('libraft', exact=True) }}
-        - {{ pin_subpackage('libraft-headers', exact=True) }}
     about:
       home: https://rapids.ai/
       license: Apache-2.0
@@ -206,23 +200,14 @@ outputs:
         - sysroot_{{ target_platform }} {{ sysroot_version }}
       host:
         - {{ pin_subpackage('libraft', exact=True) }}
-        - {{ pin_subpackage('libraft-headers', exact=True) }}
-        - cuda-profiler-api {{ cuda_profiler_api_host_version }}
         - libcublas {{ libcublas_host_version }}
         - libcublas-dev {{ libcublas_host_version }}
-        - libcurand {{ libcurand_host_version }}
-        - libcurand-dev {{ libcurand_host_version }}
-        - libcusolver {{ libcusolver_host_version }}
-        - libcusolver-dev {{ libcusolver_host_version }}
-        - libcusparse {{ libcusparse_host_version }}
-        - libcusparse-dev {{ libcusparse_host_version }}
         - glog {{ glog_version }}
         - nlohmann_json {{ nlohmann_json_version }}
         - libfaiss>=1.7.1
         - faiss-proc=*=cuda
       run:
         - {{ pin_subpackage('libraft', exact=True) }}
-        - {{ pin_subpackage('libraft-headers', exact=True) }}
         - glog {{ glog_version }}
         - faiss-proc=*=cuda
         - libfaiss {{ faiss_version }}

cpp/CMakeLists.txt

@@ -179,24 +179,23 @@ target_include_directories(
 )
 
 # Keep RAFT as lightweight as possible. Only CUDA libs and rmm should be used in global target.
-target_link_libraries(
-  raft
-  INTERFACE rmm::rmm
-            cuco::cuco
-            nvidia::cutlass::cutlass
-            CUDA::cublas${_ctk_static_suffix}
-            CUDA::curand${_ctk_static_suffix}
-            CUDA::cusolver${_ctk_static_suffix}
-            CUDA::cusparse${_ctk_static_suffix}
-            raft::Thrust
-)
+target_link_libraries(raft INTERFACE rmm::rmm cuco::cuco nvidia::cutlass::cutlass raft::Thrust)
 
 target_compile_features(raft INTERFACE cxx_std_17 $<BUILD_INTERFACE:cuda_std_17>)
 target_compile_options(
   raft INTERFACE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--expt-extended-lambda
                  --expt-relaxed-constexpr>
 )
 
+set(RAFT_CUSOLVER_DEPENDENCY CUDA::cusolver${_ctk_static_suffix})
+set(RAFT_CUBLAS_DEPENDENCY CUDA::cublas${_ctk_static_suffix})
+set(RAFT_CURAND_DEPENDENCY CUDA::curand${_ctk_static_suffix})
+set(RAFT_CUSPARSE_DEPENDENCY CUDA::cusparse${_ctk_static_suffix})
+
+set(RAFT_CTK_MATH_DEPENDENCIES ${RAFT_CUBLAS_DEPENDENCY} ${RAFT_CUSOLVER_DEPENDENCY}
+                               ${RAFT_CUSPARSE_DEPENDENCY} ${RAFT_CURAND_DEPENDENCY}
+)
+
 # Endian detection
 include(TestBigEndian)
 test_big_endian(BIG_ENDIAN)
@@ -454,7 +453,13 @@ if(RAFT_COMPILE_LIBRARY)
                INTERFACE_POSITION_INDEPENDENT_CODE ON
   )
 
-  target_link_libraries(raft_lib PUBLIC raft::raft $<TARGET_NAME_IF_EXISTS:OpenMP::OpenMP_CXX>)
+  target_link_libraries(
+    raft_lib
+    PUBLIC raft::raft
+           ${RAFT_CTK_MATH_DEPENDENCIES} # TODO: Once `raft::resources` is used everywhere, this
+                                         # will just be cublas
+           $<TARGET_NAME_IF_EXISTS:OpenMP::OpenMP_CXX>
+  )
   target_compile_options(
     raft_lib PRIVATE "$<$<COMPILE_LANGUAGE:CXX>:${RAFT_CXX_FLAGS}>"
                      "$<$<COMPILE_LANGUAGE:CUDA>:${RAFT_CUDA_FLAGS}>"
@@ -606,13 +611,39 @@ if(TARGET raft_lib)
   list(APPEND raft_export_sets raft-compiled-lib-exports)
 endif()
 
+string(
+  APPEND
+  code_string
+  [=[
+ option(RAFT_ENABLE_CUSOLVER_DEPENDENCY "Enable cusolver dependency" ON)
+ option(RAFT_ENABLE_CUBLAS_DEPENDENCY "Enable cublas dependency" ON)
+ option(RAFT_ENABLE_CURAND_DEPENDENCY "Enable curand dependency" ON)
+ option(RAFT_ENABLE_CUSPARSE_DEPENDENCY "Enable cusparse dependency" ON)
+
+mark_as_advanced(RAFT_ENABLE_CUSOLVER_DEPENDENCY)
+mark_as_advanced(RAFT_ENABLE_CUBLAS_DEPENDENCY)
+mark_as_advanced(RAFT_ENABLE_CURAND_DEPENDENCY)
+mark_as_advanced(RAFT_ENABLE_CUSPARSE_DEPENDENCY)
+
+target_link_libraries(raft::raft INTERFACE
+  $<$<BOOL:${RAFT_ENABLE_CUSOLVER_DEPENDENCY}>:${RAFT_CUSOLVER_DEPENDENCY}>
+  $<$<BOOL:${RAFT_ENABLE_CUBLAS_DEPENDENCY}>:${RAFT_CUBLAS_DEPENDENCY}>
+  $<$<BOOL:${RAFT_ENABLE_CUSPARSE_DEPENDENCY}>:${RAFT_CUSPARSE_DEPENDENCY}>
+  $<$<BOOL:${RAFT_ENABLE_CURAND_DEPENDENCY}>:${RAFT_CURAND_DEPENDENCY}>
+)
+]=]
+)
+
+# Use `rapids_export` for 22.04 as it will have COMPONENT support
 rapids_export(
   INSTALL raft
   EXPORT_SET raft-exports
   COMPONENTS ${raft_components}
   COMPONENTS_EXPORT_SET ${raft_export_sets}
   GLOBAL_TARGETS raft compiled distributed
-  NAMESPACE raft:: DOCUMENTATION doc_string FINAL_CODE_BLOCK code_string
+  NAMESPACE raft::
+  DOCUMENTATION doc_string
+  FINAL_CODE_BLOCK code_string
 )
 
 # ##################################################################################################
@@ -622,8 +653,10 @@ rapids_export(
   EXPORT_SET raft-exports
   COMPONENTS ${raft_components}
   COMPONENTS_EXPORT_SET ${raft_export_sets}
-  GLOBAL_TARGETS raft
-  compiled distributed DOCUMENTATION doc_string NAMESPACE raft:: FINAL_CODE_BLOCK code_string
+  GLOBAL_TARGETS raft compiled distributed
+  DOCUMENTATION doc_string
+  NAMESPACE raft::
+  FINAL_CODE_BLOCK code_string
 )
 
 # ##################################################################################################

cpp/bench/ann/CMakeLists.txt

@@ -80,6 +80,7 @@ function(ConfigureAnnBench)
             $<$<BOOL:${RAFT_ANN_BENCH_USE_MULTIGPU}>:NCCL::NCCL>
             ${ConfigureAnnBench_LINKS}
             Threads::Threads
+            ${RAFT_CTK_MATH_DEPENDENCIES}
             $<TARGET_NAME_IF_EXISTS:OpenMP::OpenMP_CXX>
             $<TARGET_NAME_IF_EXISTS:conda_env>
   )

cpp/include/raft/linalg/detail/cublas_wrappers.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022, NVIDIA CORPORATION.
+ * Copyright (c) 2022-2023, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -965,7 +965,8 @@ inline cublasStatus_t cublasdot(cublasHandle_t handle,
                                 cudaStream_t stream)
 {
   RAFT_CUBLAS_TRY(cublasSetStream(handle, stream));
-  return cublasSdot(handle, n, x, incx, y, incy, result);
+  return cublasDotEx(
+    handle, n, x, CUDA_R_32F, incx, y, CUDA_R_32F, incy, result, CUDA_R_32F, CUDA_R_32F);
 }
 
 template <>
@@ -979,7 +980,8 @@ inline cublasStatus_t cublasdot(cublasHandle_t handle,
                                 cudaStream_t stream)
 {
   RAFT_CUBLAS_TRY(cublasSetStream(handle, stream));
-  return cublasDdot(handle, n, x, incx, y, incy, result);
+  return cublasDotEx(
+    handle, n, x, CUDA_R_64F, incx, y, CUDA_R_64F, incy, result, CUDA_R_64F, CUDA_R_64F);
 }
 /** @} */
 

cpp/include/raft/spectral/detail/modularity_maximization.hpp

@@ -32,24 +32,6 @@
 #include <raft/spectral/eigen_solvers.cuh>
 #include <raft/spectral/matrix_wrappers.hpp>
 
-#ifdef COLLECT_TIME_STATISTICS
-#include <cuda_profiler_api.h>
-#include <stddef.h>
-#include <sys/resource.h>
-#include <sys/sysinfo.h>
-#include <sys/time.h>
-#endif
-
-#ifdef COLLECT_TIME_STATISTICS
-static double timer(void)
-{
-  struct timeval tv;
-  cudaDeviceSynchronize();
-  gettimeofday(&tv, NULL);
-  return (double)tv.tv_sec + (double)tv.tv_usec / 1000000.0;
-}
-#endif
-
 namespace raft {
 namespace spectral {
 namespace detail {

cpp/test/CMakeLists.txt

@@ -31,12 +31,13 @@ function(ConfigureTest)
 
   target_link_libraries(
     ${TEST_NAME}
-    PRIVATE raft::raft
+    PRIVATE raft
             raft_internal
             $<$<BOOL:${ConfigureTest_LIB}>:raft::compiled>
             GTest::gtest
             GTest::gtest_main
             Threads::Threads
+            ${RAFT_CTK_MATH_DEPENDENCIES}
             $<TARGET_NAME_IF_EXISTS:OpenMP::OpenMP_CXX>
             $<TARGET_NAME_IF_EXISTS:conda_env>
   )

docs/source/build.md

@@ -151,18 +151,22 @@ make -j<parallel_level> install
 
 RAFT's cmake has the following configurable flags available:.
 
-| Flag                      | Possible Values      | Default Value | Behavior |
-|---------------------------|----------------------| --- | --- |
-| BUILD_TESTS               | ON, OFF              | ON | Compile Googletests |
-| BUILD_PRIMS_BENCH               | ON, OFF              | OFF | Compile benchmarks |
+| Flag                            | Possible Values      | Default Value | Behavior                                                                     |
+|---------------------------------|----------------------| --- |------------------------------------------------------------------------------|
+| BUILD_TESTS                     | ON, OFF              | ON | Compile Googletests                                                          |
+| BUILD_PRIMS_BENCH                     | ON, OFF              | OFF | Compile benchmarks                                                           |
 | BUILD_ANN_BENCH               | ON, OFF              | OFF | Compile end-to-end ANN benchmarks |
-| raft_FIND_COMPONENTS      | compiled distributed | | Configures the optional components as a space-separated list |
 | RAFT_COMPILE_LIBRARY      | ON, OFF              | ON if either BUILD_TESTS or BUILD_PRIMS_BENCH is ON; otherwise OFF | Compiles all `libraft` shared libraries (these are required for Googletests) |
-| DETECT_CONDA_ENV          | ON, OFF              | ON | Enable detection of conda environment for dependencies |
-| RAFT_NVTX                 | ON, OFF              | OFF | Enable NVTX Markers |
-| CUDA_ENABLE_KERNELINFO    | ON, OFF              | OFF | Enables `kernelinfo` in nvcc. This is useful for `compute-sanitizer` |
-| CUDA_ENABLE_LINEINFO      | ON, OFF              | OFF | Enable the -lineinfo option for nvcc |
-| CUDA_STATIC_RUNTIME       | ON, OFF              | OFF | Statically link the CUDA runtime |
+| raft_FIND_COMPONENTS            | compiled distributed | | Configures the optional components as a space-separated list                 |
+| RAFT_ENABLE_CUBLAS_DEPENDENCY   | ON, OFF | ON | Link against cublas library in `raft::raft`                                  | 
+| RAFT_ENABLE_CUSOLVER_DEPENDENCY | ON, OFF | ON | Link against cusolver library in `raft::raft`                                | 
+| RAFT_ENABLE_CUSPARSE_DEPENDENCY | ON, OFF | ON | Link against cusparse library in `raft::raft`                                | 
+| RAFT_ENABLE_CUSOLVER_DEPENDENCY | ON, OFF | ON | Link against curand library in `raft::raft`                                  | 
+| DETECT_CONDA_ENV                | ON, OFF              | ON | Enable detection of conda environment for dependencies                       |
+| RAFT_NVTX                       | ON, OFF              | OFF | Enable NVTX Markers                                                          |
+| CUDA_ENABLE_KERNELINFO          | ON, OFF              | OFF | Enables `kernelinfo` in nvcc. This is useful for `compute-sanitizer`         |
+| CUDA_ENABLE_LINEINFO            | ON, OFF              | OFF | Enable the -lineinfo option for nvcc                                         |
+| CUDA_STATIC_RUNTIME             | ON, OFF              | OFF | Statically link the CUDA runtime                                             |
 
 Currently, shared libraries are provided for the `libraft-nn` and `libraft-distance` components.