Merge branch 'branch-24.12' into nccl-clique-init

rapidsai · Dec 5, 2024 · 7f1961b · 7f1961b
2 parents 33a4cd1 + 4ecd925
commit 7f1961b
Show file tree

Hide file tree

Showing 52 changed files with 2,046 additions and 1,762 deletions.
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
@@ -80,7 +80,6 @@ jobs:
     with:
       build_type: pull-request
       enable_check_symbols: true
-      symbol_exclusions: raft_cutlass
   conda-python-build:
     needs: conda-cpp-build
     secrets: inherit

diff --git a/.github/workflows/test.yaml b/.github/workflows/test.yaml
@@ -23,7 +23,6 @@ jobs:
       date: ${{ inputs.date }}
       sha: ${{ inputs.sha }}
       enable_check_symbols: true
-      symbol_exclusions: raft_cutlass
   conda-cpp-tests:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]

diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -93,7 +93,10 @@ repos:
               - id: codespell
                 additional_dependencies: [tomli]
                 args: ["--toml", "pyproject.toml"]
-                exclude: (?x)^(^CHANGELOG.md$)
+                exclude: |
+                  (?x)
+                    ^CHANGELOG[.]md$|
+                    ^cpp/cmake/patches/cutlass/build-export[.]patch$
       - repo: https://github.com/pre-commit/pre-commit-hooks
         rev: v4.5.0
         hooks:

diff --git a/ci/release/update-version.sh b/ci/release/update-version.sh
@@ -35,7 +35,6 @@ function sed_runner() {
     sed -i.bak ''"$1"'' $2 && rm -f ${2}.bak
 }
 
-sed_runner "s/set(RAPIDS_VERSION .*)/set(RAPIDS_VERSION \"${NEXT_SHORT_TAG}\")/g" cpp/template/cmake/thirdparty/fetch_rapids.cmake
 sed_runner 's/'"find_and_configure_ucxx(VERSION .*"'/'"find_and_configure_ucxx(VERSION  ${NEXT_UCXX_SHORT_TAG_PEP440}"'/g' python/raft-dask/cmake/thirdparty/get_ucxx.cmake
 sed_runner 's/'"branch-.*"'/'"branch-${NEXT_UCXX_SHORT_TAG_PEP440}"'/g' python/raft-dask/cmake/thirdparty/get_ucxx.cmake
 

diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
@@ -268,6 +268,10 @@ if(RAFT_COMPILE_LIBRARY)
     src/raft_runtime/random/rmat_rectangular_generator_int64_float.cu
     src/raft_runtime/random/rmat_rectangular_generator_int_double.cu
     src/raft_runtime/random/rmat_rectangular_generator_int_float.cu
+    src/raft_runtime/solver/lanczos_solver_int64_double.cu
+    src/raft_runtime/solver/lanczos_solver_int64_float.cu
+    src/raft_runtime/solver/lanczos_solver_int_double.cu
+    src/raft_runtime/solver/lanczos_solver_int_float.cu
   )
   set_target_properties(
     raft_objs

diff --git a/cpp/cmake/patches/cutlass/build-export.patch b/cpp/cmake/patches/cutlass/build-export.patch
@@ -0,0 +1,27 @@
+From e0a9597946257a01ae8444200f836ee51d5597ba Mon Sep 17 00:00:00 2001
+From: Kyle Edwards <[email protected]>
+Date: Wed, 20 Nov 2024 16:37:38 -0500
+Subject: [PATCH] Remove erroneous include directories
+
+These directories are left over from when CuTe was a separate
+CMake project. Remove them.
+---
+ CMakeLists.txt | 2 --
+ 1 file changed, 2 deletions(-)
+
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index 7419bdf5e..545384d82 100755
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -665,8 +665,6 @@ target_include_directories(
+   $<INSTALL_INTERFACE:include>
+   $<BUILD_INTERFACE:${CUTLASS_INCLUDE_DIR}>
+   $<BUILD_INTERFACE:${CMAKE_CURRENT_BINARY_DIR}/include>
+-  $<BUILD_INTERFACE:${cute_SOURCE_DIR}/include>
+-  $<BUILD_INTERFACE:${cute_SOURCE_DIR}/examples>
+   )
+
+ # Mark CTK headers as system to supress warnings from them
+-- 
+2.34.1
+
diff --git a/cpp/cmake/patches/cutlass_override.json b/cpp/cmake/patches/cutlass_override.json
@@ -0,0 +1,16 @@
+{
+  "packages" : {
+    "cutlass" : {
+      "version": "3.5.1",
+      "git_url": "https://github.com/NVIDIA/cutlass.git",
+      "git_tag": "v${version}",
+      "patches" : [
+        {
+          "file" : "${current_json_dir}/cutlass/build-export.patch",
+          "issue" : "Fix build directory export",
+          "fixed_in" : ""
+        }
+      ]
+    }
+  }
+}
diff --git a/cpp/cmake/thirdparty/get_cutlass.cmake b/cpp/cmake/thirdparty/get_cutlass.cmake
@@ -13,7 +13,9 @@
 # =============================================================================
 
 function(find_and_configure_cutlass)
-  set(oneValueArgs VERSION REPOSITORY PINNED_TAG)
+  set(options)
+  set(oneValueArgs)
+  set(multiValueArgs)
   cmake_parse_arguments(PKG "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
 
   # if(RAFT_ENABLE_DIST_DEPENDENCIES OR RAFT_COMPILE_LIBRARIES)
@@ -34,13 +36,22 @@ function(find_and_configure_cutlass)
     set(CUDART_LIBRARY "${CUDA_cudart_static_LIBRARY}" CACHE FILEPATH "fixing cutlass cmake code" FORCE)
   endif()
 
+  include("${rapids-cmake-dir}/cpm/package_override.cmake")
+  rapids_cpm_package_override("${CMAKE_CURRENT_FUNCTION_LIST_DIR}/../patches/cutlass_override.json")
+
+  include("${rapids-cmake-dir}/cpm/detail/package_details.cmake")
+  rapids_cpm_package_details(cutlass version repository tag shallow exclude)
+
+  include("${rapids-cmake-dir}/cpm/detail/generate_patch_command.cmake")
+  rapids_cpm_generate_patch_command(cutlass ${version} patch_command)
+
   rapids_cpm_find(
-    NvidiaCutlass ${PKG_VERSION}
+    NvidiaCutlass ${version}
     GLOBAL_TARGETS nvidia::cutlass::cutlass
     CPM_ARGS
-    GIT_REPOSITORY ${PKG_REPOSITORY}
-    GIT_TAG ${PKG_PINNED_TAG}
-    GIT_SHALLOW TRUE
+    GIT_REPOSITORY ${repository}
+    GIT_TAG ${tag}
+    GIT_SHALLOW ${shallow} ${patch_command}
     OPTIONS "CUDAToolkit_ROOT ${CUDAToolkit_LIBRARY_DIR}"
   )
 
@@ -79,14 +90,4 @@ function(find_and_configure_cutlass)
   )
 endfunction()
 
-if(NOT RAFT_CUTLASS_GIT_TAG)
-  set(RAFT_CUTLASS_GIT_TAG v2.10.0)
-endif()
-
-if(NOT RAFT_CUTLASS_GIT_REPOSITORY)
-  set(RAFT_CUTLASS_GIT_REPOSITORY https://github.com/NVIDIA/cutlass.git)
-endif()
-
-find_and_configure_cutlass(
-  VERSION 2.10.0 REPOSITORY ${RAFT_CUTLASS_GIT_REPOSITORY} PINNED_TAG ${RAFT_CUTLASS_GIT_TAG}
-)
+find_and_configure_cutlass()
diff --git a/cpp/include/raft/core/device_mdspan.hpp b/cpp/include/raft/core/device_mdspan.hpp
@@ -210,7 +210,7 @@ auto constexpr make_device_strided_matrix_view(ElementType* ptr,
   constexpr auto is_row_major = std::is_same_v<LayoutPolicy, layout_c_contiguous>;
   constexpr auto is_col_major = std::is_same_v<LayoutPolicy, layout_f_contiguous>;
 
-  assert(is_row_major || is_col_major);
+  static_assert(is_row_major || is_col_major, "Unsupported layout policy for strided matrix view");
 
   IndexType stride0 = is_row_major ? (stride > 0 ? stride : n_cols) : 1;
   IndexType stride1 = is_row_major ? 1 : (stride > 0 ? stride : n_rows);

diff --git a/cpp/include/raft/distance/detail/pairwise_distance_epilogue_elementwise.h b/cpp/include/raft/distance/detail/pairwise_distance_epilogue_elementwise.h
@@ -61,6 +61,7 @@ class PairwiseDistanceEpilogueElementwise {
   using ElementT                      = ElementT_;
   static int const kElementsPerAccess = ElementsPerAccess;
   static int const kCount             = kElementsPerAccess;
+  static bool const kIsSingleSource   = true;
 
   using DistanceOp = DistanceOp_;
   using FinalOp    = FinalOp_;

diff --git a/cpp/include/raft/distance/fused_distance_nn-ext.cuh b/cpp/include/raft/distance/fused_distance_nn-ext.cuh
diff --git a/cpp/include/raft/distance/fused_distance_nn.cuh b/cpp/include/raft/distance/fused_distance_nn.cuh
@@ -15,10 +15,4 @@
  */
 #pragma once
 
-#ifndef RAFT_EXPLICIT_INSTANTIATE_ONLY
-#include "fused_distance_nn-inl.cuh"
-#endif
-
-#ifdef RAFT_COMPILED
-#include "fused_distance_nn-ext.cuh"
-#endif
+#include "fused_distance_nn-inl.cuh"
diff --git a/cpp/include/raft/linalg/detail/coalesced_reduction-ext.cuh b/cpp/include/raft/linalg/detail/coalesced_reduction-ext.cuh
diff --git a/cpp/include/raft/linalg/detail/coalesced_reduction.cuh b/cpp/include/raft/linalg/detail/coalesced_reduction.cuh
@@ -16,11 +16,4 @@
 
 #pragma once
 
-// Always include inline definitions of coalesced reduction, because we do not
-// force explicit instantion.
 #include "coalesced_reduction-inl.cuh"
-
-// Do include the extern template instantiations when possible.
-#ifdef RAFT_COMPILED
-#include "coalesced_reduction-ext.cuh"
-#endif