Merge branch 'branch-0.20' into lists_concatenate_list_elements

Dockerfile

@@ -19,7 +19,6 @@ RUN apt update -y --fix-missing && \
       git \
       gcc-${CC} \
       g++-${CXX} \
-      libboost-all-dev \
       tzdata && \
     apt-get autoremove -y && \
     apt-get clean && \
@@ -68,8 +67,8 @@ RUN if [ -f /cudf/docker/package_versions.sh ]; \
          conda env create --name cudf --file /cudf/conda/environments/cudf_dev_cuda${CUDA_SHORT_VERSION}.yml ; \
     fi
 
-ENV CC=/usr/bin/gcc-${CC}
-ENV CXX=/usr/bin/g++-${CXX}
+ENV CC=/opts/conda/envs/rapids/bin/gcc-${CC}
+ENV CXX=/opts/conda/envs/rapids/bin/g++-${CXX}
 
 # libcudf & cudf build/install
 RUN source activate cudf && \

conda/environments/cudf_dev_cuda11.0.yml

@@ -47,7 +47,6 @@ dependencies:
   - dlpack
   - arrow-cpp=1.0.1
   - arrow-cpp-proc * cuda
-  - boost-cpp>=1.72.0
   - double-conversion
   - rapidjson
   - flatbuffers

conda/environments/cudf_dev_cuda11.1.yml

@@ -47,7 +47,6 @@ dependencies:
   - dlpack
   - arrow-cpp=1.0.1
   - arrow-cpp-proc * cuda
-  - boost-cpp>=1.72.0
   - double-conversion
   - rapidjson
   - flatbuffers

conda/environments/cudf_dev_cuda11.2.yml

@@ -47,7 +47,6 @@ dependencies:
   - dlpack
   - arrow-cpp=1.0.1
   - arrow-cpp-proc * cuda
-  - boost-cpp>=1.72.0
   - double-conversion
   - rapidjson
   - flatbuffers

conda/recipes/libcudf/meta.yaml

@@ -39,12 +39,10 @@ requirements:
     - cudatoolkit {{ cuda_version }}.*
     - arrow-cpp 1.0.1
     - arrow-cpp-proc * cuda
-    - boost-cpp 1.72.0
     - dlpack
   run:
     - {{ pin_compatible('cudatoolkit', max_pin='x.x') }}
     - arrow-cpp-proc * cuda
-    - {{ pin_compatible('boost-cpp', max_pin='x.x.x') }}
     - {{ pin_compatible('dlpack', max_pin='x.x') }}
 
 test:
@@ -55,7 +53,7 @@ test:
     - test -f $PREFIX/include/cudf/ast/transform.hpp
     - test -f $PREFIX/include/cudf/ast/detail/linearizer.hpp
     - test -f $PREFIX/include/cudf/ast/detail/operators.hpp
-    - test -f $PREFIX/include/cudf/ast/linearizer.hpp
+    - test -f $PREFIX/include/cudf/ast/nodes.hpp
     - test -f $PREFIX/include/cudf/ast/operators.hpp
     - test -f $PREFIX/include/cudf/binaryop.hpp
     - test -f $PREFIX/include/cudf/labeling/label_bins.hpp

cpp/CMakeLists.txt

@@ -121,8 +121,6 @@ find_package(ZLIB REQUIRED)
 find_package(Threads REQUIRED)
 # add third party dependencies using CPM
 include(cmake/thirdparty/CUDF_GetCPM.cmake)
-# find boost
-include(cmake/thirdparty/CUDF_FindBoost.cmake)
 # find jitify
 include(cmake/thirdparty/CUDF_GetJitify.cmake)
 # find thrust/cub
@@ -477,7 +475,6 @@ add_dependencies(cudf jitify_preprocess_run)
 # Specify the target module library dependencies
 target_link_libraries(cudf
            PUBLIC ZLIB::ZLIB
-                  Boost::filesystem
                   ${ARROW_LIBRARIES}
                   cudf::Thrust
                   rmm::rmm)

cpp/cmake/cudf-build-config.cmake.in

@@ -43,8 +43,6 @@ find_dependency(ZLIB)
 
 # add third party dependencies using CPM
 include(@CUDF_SOURCE_DIR@/cmake/thirdparty/CUDF_GetCPM.cmake)
-# find boost
-include(@CUDF_SOURCE_DIR@/cmake/thirdparty/CUDF_FindBoost.cmake)
 # find jitify
 include(@CUDF_SOURCE_DIR@/cmake/thirdparty/CUDF_GetJitify.cmake)
 # find thrust/cub

cpp/cmake/cudf-config.cmake.in

@@ -71,11 +71,6 @@ find_dependency(CUDAToolkit)
 find_dependency(Threads)
 find_dependency(ZLIB)
 
-# Don't look for a Boost CMake configuration file because it adds the
-# `-DBOOST_ALL_NO_LIB` and `-DBOOST_FILESYSTEM_DYN_LINK` compile defs
-set(Boost_NO_BOOST_CMAKE ON)
-find_dependency(Boost @CUDF_MIN_VERSION_Boost@ COMPONENTS filesystem)
-
 find_dependency(Arrow @CUDF_VERSION_Arrow@)
 
 set(ArrowCUDA_DIR "${Arrow_DIR}")

cpp/cmake/thrust.patch

@@ -42,3 +42,25 @@ index 1ffeef0..5e80800 100644
            for (int ITEM = 1; ITEM < ITEMS_PER_THREAD; ++ITEM)
            {
              if (ITEMS_PER_THREAD * tid + ITEM < num_remaining)
+diff --git a/thrust/system/cuda/detail/scan_by_key.h b/thrust/system/cuda/detail/scan_by_key.h
+index fe4b321c..b3974c69 100644
+--- a/thrust/system/cuda/detail/scan_by_key.h
++++ b/thrust/system/cuda/detail/scan_by_key.h
+@@ -513,7 +513,7 @@ namespace __scan_by_key {
+             scan_op(scan_op_)
+       {
+         int  tile_idx      = blockIdx.x;
+-        Size tile_base     = ITEMS_PER_TILE * tile_idx;
++        Size tile_base     = ITEMS_PER_TILE * static_cast<Size>(tile_idx);
+         Size num_remaining = num_items - tile_base;
+
+         if (num_remaining > ITEMS_PER_TILE)
+@@ -734,7 +734,7 @@ namespace __scan_by_key {
+                              ScanOp                     scan_op,
+                              AddInitToScan              add_init_to_scan)
+   {
+-    int          num_items    = static_cast<int>(thrust::distance(keys_first, keys_last));
++    size_t       num_items    = static_cast<size_t>(thrust::distance(keys_first, keys_last));
+     size_t       storage_size = 0;
+     cudaStream_t stream       = cuda_cub::stream(policy);
+     bool         debug_sync   = THRUST_DEBUG_SYNC_FLAG;

cpp/include/cudf/ast/detail/linearizer.hpp

@@ -83,10 +83,7 @@ class linearizer;
  * This class is a part of a "visitor" pattern with the `linearizer` class.
  * Nodes inheriting from this class can accept visitors.
  */
-class node {
-  friend class detail::linearizer;
-
- private:
+struct node {
   virtual cudf::size_type accept(detail::linearizer& visitor) const = 0;
 };
 
@@ -102,10 +99,6 @@ class node {
  * resolved into intermediate data storage in shared memory.
  */
 class linearizer {
-  friend class literal;
-  friend class column_reference;
-  friend class expression;
-
  public:
   /**
    * @brief Construct a new linearizer object

cpp/include/cudf/ast/detail/operators.hpp

@@ -753,43 +753,6 @@ struct operator_functor<ast_operator::NOT> {
   }
 };
 
-#if 0
-/**
- * @brief Functor used to double-type-dispatch binary operators.
- *
- * This functor's `operator()` is templated to validate calls to its operators based on the input
- * type, as determined by the `is_valid_binary_op` trait.
- *
- * @tparam OperatorFunctor Binary operator functor.
- */
-template <typename OperatorFunctor>
-struct double_dispatch_binary_operator_types {
-  template <typename LHS,
-            typename RHS,
-            typename F,
-            typename... Ts,
-            std::enable_if_t<is_valid_binary_op<OperatorFunctor, LHS, RHS>>* = nullptr>
-  CUDA_HOST_DEVICE_CALLABLE void operator()(F&& f, Ts&&... args)
-  {
-    f.template operator()<OperatorFunctor, LHS, RHS>(std::forward<Ts>(args)...);
-  }
-
-  template <typename LHS,
-            typename RHS,
-            typename F,
-            typename... Ts,
-            std::enable_if_t<!is_valid_binary_op<OperatorFunctor, LHS, RHS>>* = nullptr>
-  CUDA_HOST_DEVICE_CALLABLE void operator()(F&& f, Ts&&... args)
-  {
-#ifndef __CUDA_ARCH__
-    CUDF_FAIL("Invalid binary operation.");
-#else
-    cudf_assert(false && "Invalid binary operation.");
-#endif
-  }
-};
-#endif
-
 /**
  * @brief Functor used to single-type-dispatch binary operators.
  *
@@ -856,16 +819,6 @@ struct type_dispatch_binary_op {
                                             F&& f,
                                             Ts&&... args)
   {
-#if 0
-    // Double dispatch
-    /*
-    double_type_dispatcher(lhs_type,
-                           rhs_type,
-                           detail::double_dispatch_binary_operator_types<operator_functor<op>>{},
-                           std::forward<F>(f),
-                           std::forward<Ts>(args)...);
-    */
-#endif
     // Single dispatch (assume lhs_type == rhs_type)
     type_dispatcher(lhs_type,
                     detail::single_dispatch_binary_operator_types<operator_functor<op>>{},