[ENH] [FINAL] Header structure: combine all PRs into one (#1469)

This is a rebase of all the commits in PRs: - #1437 - #1438 - #1439 - #1440 - #1441 The original PRs have not been rebased to preserve review comments. This PR is up to date with branch 23.06. Closes #1416 Authors: - Allard Hendriksen (https://github.com/ahendriksen) Approvers: - Divye Gala (https://github.com/divyegala) - Corey J. Nolet (https://github.com/cjnolet) URL: #1469
rapidsai · Apr 28, 2023 · fbce1a4 · fbce1a4
1 parent 082be6e
commit fbce1a4
Show file tree

Hide file tree

Showing 431 changed files with 18,575 additions and 12,748 deletions.
diff --git a/README.md b/README.md
@@ -203,7 +203,7 @@ RAFT itself can be installed through conda, [CMake Package Manager (CPM)](https:
 
 The easiest way to install RAFT is through conda and several packages are provided.
 - `libraft-headers` RAFT headers
-- `libraft` (optional) shared library of pre-compiled template specializations and runtime APIs.
+- `libraft` (optional) shared library of pre-compiled template instantiations and runtime APIs.
 - `pylibraft` (optional) Python wrappers around RAFT algorithms and primitives.
 - `raft-dask` (optional) enables deployment of multi-node multi-GPU algorithms that use RAFT `raft::comms` in Dask clusters.
 
@@ -236,11 +236,11 @@ You can find an [example RAFT](cpp/template/README.md) project template in the `
 
 Additional CMake targets can be made available by adding components in the table below to the `RAFT_COMPONENTS` list above, separated by spaces. The `raft::raft` target will always be available. RAFT headers require, at a minimum, the CUDA toolkit libraries and RMM dependencies.
 
-| Component   | Target              | Description                                               | Base Dependencies                     |
-|-------------|---------------------|-----------------------------------------------------------|---------------------------------------|
-| n/a         | `raft::raft`        | Full RAFT header library                                  | CUDA toolkit, RMM, NVTX, CCCL, CUTLASS |
-| compiled    | `raft::compiled`    | Pre-compiled template specializations and runtime library | raft::raft                            |
-| distributed | `raft::distributed` | Dependencies for `raft::comms` APIs                       | raft::raft, UCX, NCCL                 |
+| Component   | Target              | Description                                              | Base Dependencies                      |
+|-------------|---------------------|----------------------------------------------------------|----------------------------------------|
+| n/a         | `raft::raft`        | Full RAFT header library                                 | CUDA toolkit, RMM, NVTX, CCCL, CUTLASS |
+| compiled    | `raft::compiled`    | Pre-compiled template instantiations and runtime library | raft::raft                             |
+| distributed | `raft::distributed` | Dependencies for `raft::comms` APIs                      | raft::raft, UCX, NCCL                  |
 
 ### Source
 
@@ -287,7 +287,7 @@ The folder structure mirrors other RAPIDS repos, with the following folders:
     - `util`: Various reusable tools and utilities for accelerated algorithm development
   - `internal`: A private header-only component that hosts the code shared between benchmarks and tests.
   - `scripts`: Helpful scripts for development
-  - `src`: Compiled APIs and template specializations for the shared libraries
+  - `src`: Compiled APIs and template instantiations for the shared libraries
   - `template`: A skeleton template containing the bare-bones file structure and cmake configuration for writing applications with RAFT.
   - `test`: Googletests source code
 - `docs`: Source code and scripts for building library documentation (Uses breath, doxygen, & pydocs)

diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
diff --git a/cpp/bench/ann/src/raft/raft_benchmark.cu b/cpp/bench/ann/src/raft/raft_benchmark.cu
@@ -22,10 +22,6 @@
 #include <type_traits>
 #include <utility>
 
-#ifdef RAFT_COMPILED
-#include <raft/neighbors/specializations.cuh>
-#endif
-
 #include "../common/ann_types.hpp"
 #include "../common/benchmark_util.hpp"
 #undef WARP_SIZE

diff --git a/cpp/bench/ann/src/raft/raft_ivf_flat.cu b/cpp/bench/ann/src/raft/raft_ivf_flat.cu
@@ -15,12 +15,8 @@
  */
 #include "raft_ivf_flat_wrapper.h"
 
-#ifdef RAFT_COMPILED
-#include <raft/neighbors/specializations.cuh>
-#endif
-
 namespace raft::bench::ann {
 template class RaftIvfFlatGpu<float, int64_t>;
 template class RaftIvfFlatGpu<uint8_t, int64_t>;
 template class RaftIvfFlatGpu<int8_t, int64_t>;
-}  // namespace raft::bench::ann
+}  // namespace raft::bench::ann
diff --git a/cpp/bench/ann/src/raft/raft_ivf_flat_wrapper.h b/cpp/bench/ann/src/raft/raft_ivf_flat_wrapper.h
@@ -29,6 +29,7 @@
 #include <raft/neighbors/ivf_flat_types.hpp>
 #include <raft/util/cudart_utils.hpp>
 #include <rmm/device_uvector.hpp>
+#include <rmm/mr/device/pool_memory_resource.hpp>
 #include <stdexcept>
 #include <string>
 #include <type_traits>

diff --git a/cpp/bench/ann/src/raft/raft_ivf_pq.cu b/cpp/bench/ann/src/raft/raft_ivf_pq.cu
@@ -15,10 +15,6 @@
  */
 #include "raft_ivf_pq_wrapper.h"
 
-#ifdef RAFT_COMPILED
-#include <raft/neighbors/specializations.cuh>
-#endif
-
 namespace raft::bench::ann {
 template class RaftIvfPQ<float, int64_t>;
 template class RaftIvfPQ<uint8_t, int64_t>;

diff --git a/cpp/bench/prims/CMakeLists.txt b/cpp/bench/prims/CMakeLists.txt
@@ -17,7 +17,7 @@
 
 function(ConfigureBench)
 
-  set(options OPTIONAL LIB)
+  set(options OPTIONAL LIB EXPLICIT_INSTANTIATE_ONLY)
   set(oneValueArgs NAME)
   set(multiValueArgs PATH TARGETS CONFIGURATIONS)
 
@@ -55,6 +55,10 @@ function(ConfigureBench)
                           "$<$<COMPILE_LANGUAGE:CUDA>:${RAFT_CUDA_FLAGS}>"
   )
 
+  if(ConfigureTest_EXPLICIT_INSTANTIATE_ONLY)
+    target_compile_definitions(${BENCH_NAME} PRIVATE "RAFT_EXPLICIT_INSTANTIATE_ONLY")
+  endif()
+
   target_include_directories(
     ${BENCH_NAME} PUBLIC "$<BUILD_INTERFACE:${RAFT_SOURCE_DIR}/bench/prims>"
   )
@@ -71,7 +75,7 @@ endfunction()
 if(BUILD_PRIMS_BENCH)
   ConfigureBench(
     NAME CLUSTER_BENCH PATH bench/prims/cluster/kmeans_balanced.cu bench/prims/cluster/kmeans.cu
-    bench/prims/main.cpp OPTIONAL LIB
+    bench/prims/main.cpp OPTIONAL LIB EXPLICIT_INSTANTIATE_ONLY
   )
 
   ConfigureBench(
@@ -93,6 +97,7 @@ if(BUILD_PRIMS_BENCH)
     bench/prims/main.cpp
     OPTIONAL
     LIB
+    EXPLICIT_INSTANTIATE_ONLY
   )
 
   ConfigureBench(
@@ -112,7 +117,7 @@ if(BUILD_PRIMS_BENCH)
 
   ConfigureBench(
     NAME MATRIX_BENCH PATH bench/prims/matrix/argmin.cu bench/prims/matrix/gather.cu
-    bench/prims/matrix/select_k.cu bench/prims/main.cpp OPTIONAL LIB
+    bench/prims/matrix/select_k.cu bench/prims/main.cpp OPTIONAL LIB EXPLICIT_INSTANTIATE_ONLY
   )
 
   ConfigureBench(
@@ -139,5 +144,6 @@ if(BUILD_PRIMS_BENCH)
     bench/prims/main.cpp
     OPTIONAL
     LIB
+    EXPLICIT_INSTANTIATE_ONLY
   )
 endif()
diff --git a/cpp/bench/prims/cluster/kmeans.cu b/cpp/bench/prims/cluster/kmeans.cu
@@ -18,10 +18,6 @@
 #include <raft/cluster/kmeans.cuh>
 #include <raft/cluster/kmeans_types.hpp>
 
-#if defined RAFT_COMPILED
-#include <raft/cluster/specializations.cuh>
-#endif
-
 namespace raft::bench::cluster {
 
 struct KMeansBenchParams {

diff --git a/cpp/bench/prims/cluster/kmeans_balanced.cu b/cpp/bench/prims/cluster/kmeans_balanced.cu
@@ -18,10 +18,6 @@
 #include <raft/cluster/kmeans_balanced.cuh>
 #include <raft/random/rng.cuh>
 
-#if defined RAFT_COMPILED
-#include <raft/cluster/specializations.cuh>
-#endif
-
 namespace raft::bench::cluster {
 
 struct KMeansBalancedBenchParams {

diff --git a/cpp/bench/prims/distance/distance_common.cuh b/cpp/bench/prims/distance/distance_common.cuh
@@ -17,9 +17,6 @@
 #include <common/benchmark.hpp>
 #include <raft/distance/distance.cuh>
 #include <raft/util/cudart_utils.hpp>
-#if defined RAFT_COMPILED
-#include <raft/distance/specializations.cuh>
-#endif
 #include <rmm/device_uvector.hpp>
 
 namespace raft::bench::distance {

diff --git a/cpp/bench/prims/distance/fused_l2_nn.cu b/cpp/bench/prims/distance/fused_l2_nn.cu
@@ -18,9 +18,6 @@
 #include <raft/distance/fused_l2_nn.cuh>
 #include <raft/linalg/norm.cuh>
 #include <raft/util/cudart_utils.hpp>
-#if defined RAFT_COMPILED
-#include <raft/distance/specializations.cuh>
-#endif
 #include <rmm/device_uvector.hpp>
 
 namespace raft::bench::distance {

diff --git a/cpp/bench/prims/distance/kernels.cu b/cpp/bench/prims/distance/kernels.cu
@@ -13,10 +13,6 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#if defined RAFT_COMPILED
-#include <raft/distance/specializations.cuh>
-#endif
-
 #include <common/benchmark.hpp>
 #include <memory>
 #include <raft/core/device_resources.hpp>

diff --git a/cpp/bench/prims/distance/masked_nn.cu b/cpp/bench/prims/distance/masked_nn.cu
@@ -30,10 +30,6 @@
 #include <raft/random/rng.cuh>
 #include <raft/util/cudart_utils.hpp>
 
-#ifdef RAFT_COMPILED
-#include <raft/distance/specializations.cuh>
-#endif
-
 namespace raft::bench::distance::masked_nn {
 
 // Introduce various sparsity patterns

diff --git a/cpp/bench/prims/matrix/select_k.cu b/cpp/bench/prims/matrix/select_k.cu
@@ -23,10 +23,6 @@
 #include <raft/sparse/detail/utils.h>
 #include <raft/util/cudart_utils.hpp>
 
-#if defined RAFT_COMPILED
-#include <raft/matrix/specializations.cuh>
-#endif
-
 #include <raft/matrix/detail/select_radix.cuh>
 #include <raft/matrix/detail/select_warpsort.cuh>
 #include <raft/matrix/select_k.cuh>

diff --git a/cpp/bench/prims/neighbors/knn.cuh b/cpp/bench/prims/neighbors/knn.cuh
@@ -24,10 +24,6 @@
 #include <raft/neighbors/ivf_pq.cuh>
 #include <raft/spatial/knn/knn.cuh>
 
-#if defined RAFT_COMPILED
-#include <raft/neighbors/specializations.cuh>
-#endif
-
 #include <rmm/mr/device/managed_memory_resource.hpp>
 #include <rmm/mr/device/per_device_resource.hpp>
 

diff --git a/cpp/bench/prims/neighbors/refine_float_int64_t.cu b/cpp/bench/prims/neighbors/refine_float_int64_t.cu
@@ -17,11 +17,6 @@
 #include "refine.cuh"
 #include <common/benchmark.hpp>
 
-#if defined RAFT_COMPILED
-#include <raft/neighbors/specializations/refine.cuh>
-#include <raft/spatial/knn/specializations.cuh>
-#endif
-
 using namespace raft::neighbors;
 
 namespace raft::bench::neighbors {

diff --git a/cpp/bench/prims/neighbors/refine_uint8_t_int64_t.cu b/cpp/bench/prims/neighbors/refine_uint8_t_int64_t.cu
@@ -17,10 +17,6 @@
 #include "refine.cuh"
 #include <common/benchmark.hpp>
 
-#if defined RAFT_COMPILED
-#include <raft/neighbors/specializations.cuh>
-#endif
-
 using namespace raft::neighbors;
 
 namespace raft::bench::neighbors {

diff --git a/cpp/doxygen/Doxyfile b/cpp/doxygen/Doxyfile
@@ -918,6 +918,7 @@ EXCLUDE_SYMLINKS       = NO
 # Note that the wildcards are matched against the file with absolute path, so to
 # exclude all test directories for example use the pattern */test/*
 
+# TODO: remove specializations from exclude patterns when headers have been removed.
 EXCLUDE_PATTERNS       = */detail/* \
                          */specializations/* \
                          */thirdparty/*

diff --git a/cpp/include/raft/cluster/detail/kmeans_common.cuh b/cpp/include/raft/cluster/detail/kmeans_common.cuh
@@ -38,6 +38,7 @@
 #include <raft/distance/distance.cuh>
 #include <raft/distance/distance_types.hpp>
 #include <raft/distance/fused_l2_nn.cuh>
+#include <raft/linalg/norm.cuh>
 #include <raft/linalg/reduce_rows_by_key.cuh>
 #include <raft/linalg/unary_op.cuh>
 #include <raft/matrix/gather.cuh>

diff --git a/cpp/include/raft/cluster/specializations.cuh b/cpp/include/raft/cluster/specializations.cuh
@@ -13,12 +13,10 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#ifndef __CLUSTER_SPECIALIZATIONS_H
-#define __CLUSTER_SPECIALIZATIONS_H
-
 #pragma once
 
-#include <raft/distance/specializations.cuh>
-#include <raft/neighbors/specializations.cuh>
-
-#endif
+#pragma message(                                            \
+    __FILE__                                                \
+    " is deprecated and will be removed."                   \
+    " Including specializations is not necessary any more." \
+    " For more information, see: https://docs.rapids.ai/api/raft/nightly/using_libraft.html")
diff --git a/cpp/include/raft/core/detail/macros.hpp b/cpp/include/raft/core/detail/macros.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.
@@ -40,6 +40,40 @@
 #define RAFT_INLINE_FUNCTION _RAFT_HOST_DEVICE _RAFT_FORCEINLINE
 #endif
 
+// The RAFT_INLINE_CONDITIONAL is a conditional inline specifier that removes
+// the inline specification when RAFT_COMPILED is defined.
+//
+// When RAFT_COMPILED is not defined, functions may be defined in multiple
+// translation units and we do not want that to lead to linker errors.
+//
+// When RAFT_COMPILED is defined, this serves two purposes:
+//
+// 1. It triggers a multiple definition error message when memory_pool-inl.hpp
+// (for instance) is accidentally included in multiple translation units.
+//
+// 2. We function definitions to be non-inline, because non-inline functions
+// symbols are always exported in the object symbol table. For inline functions,
+// the compiler may elide the external symbol, which results in linker errors.
+#ifdef RAFT_COMPILED
+#define RAFT_INLINE_CONDITIONAL
+#else
+#define RAFT_INLINE_CONDITIONAL inline
+#endif  // RAFT_COMPILED
+
+// The RAFT_WEAK_FUNCTION specificies that:
+//
+// 1. A function may be defined in multiple translation units (like inline)
+//
+// 2. Must still emit an external symbol (unlike inline). This enables declaring
+// a function signature in an `-ext` header and defining it in a source file.
+//
+// From
+// https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#Common-Function-Attributes:
+//
+// "The weak attribute causes a declaration of an external symbol to be emitted
+// as a weak symbol rather than a global."
+#define RAFT_WEAK_FUNCTION __attribute__((weak))
+
 /**
  * Some macro magic to remove optional parentheses of a macro argument.
  * See https://stackoverflow.com/a/62984543