More cmake improvements and compilation fix to handle cudf nightly update

CMakeLists.txt

@@ -25,8 +25,8 @@ project(
   DESCRIPTION "GPU memory management and data representation library")
 
 # Options
-option(BUILD_TESTS "Build the test suite" ON)
-option(BUILD_BENCHMARKS "Build the benchmark suite" ON)
+option(CUCASCADE_BUILD_TESTS "Build the test suite" ON)
+option(CUCASCADE_BUILD_BENCHMARKS "Build the benchmark suite" ON)
 option(CUCASCADE_BUILD_SHARED_LIBS "Build shared library" ON)
 option(CUCASCADE_BUILD_STATIC_LIBS "Build static library" ON)
 
@@ -60,7 +60,7 @@ endif()
 # =============================================================================
 # Compiler warnings
 # =============================================================================
-option(WARNINGS_AS_ERRORS "Treat compiler warnings as errors" ON)
+option(CUCASCADE_WARNINGS_AS_ERRORS "Treat compiler warnings as errors" ON)
 
 # Warning flags for C/C++
 set(CUCASCADE_CXX_WARNING_FLAGS
@@ -118,16 +118,14 @@ target_include_directories(cucascade_objects
 # Link dependencies to object library
 target_link_libraries(cucascade_objects PUBLIC ${CUCASCADE_PUBLIC_LINK_LIBS})
 
-# Language standards set via properties (not compile features) to avoid
-# propagating cuda_std_* requirements to consumers without CUDA enabled.
-set_target_properties(cucascade_objects PROPERTIES CXX_STANDARD 20
-                                                   CXX_STANDARD_REQUIRED ON
-                                                   CXX_EXTENSIONS OFF
-                                                   CUDA_STANDARD 20
-                                                   CUDA_STANDARD_REQUIRED ON)
+# cxx_std_20 is PUBLIC so consumers know they need C++20 to use our headers.
+# cuda_std_20 is kept separate and applied PRIVATE to cucascade_objects to avoid
+# propagating it to consumers that have no CUDA compiler requirement.
+target_compile_features(cucascade_objects PUBLIC cxx_std_20)
+target_compile_features(cucascade_objects PRIVATE cuda_std_20)
 
-# Position independent code (required for shared library)
-set_target_properties(cucascade_objects PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(cucascade_objects PROPERTIES CXX_EXTENSIONS OFF
+                                                   POSITION_INDEPENDENT_CODE ON)
 
 # =============================================================================
 # Static library
@@ -139,8 +137,7 @@ if(CUCASCADE_BUILD_STATIC_LIBS)
   target_include_directories(cucascade_static
                              PUBLIC ${CUCASCADE_PUBLIC_INCLUDE_DIRS})
   target_link_libraries(cucascade_static PUBLIC ${CUCASCADE_PUBLIC_LINK_LIBS})
-  target_compile_features(cucascade_static
-                          PUBLIC ${CUCASCADE_PUBLIC_COMPILE_FEATURES})
+  target_compile_features(cucascade_static PUBLIC cxx_std_20)
 
   set_target_properties(
     cucascade_static PROPERTIES OUTPUT_NAME cucascade EXPORT_NAME
@@ -157,8 +154,7 @@ if(CUCASCADE_BUILD_SHARED_LIBS)
   target_include_directories(cucascade_shared
                              PUBLIC ${CUCASCADE_PUBLIC_INCLUDE_DIRS})
   target_link_libraries(cucascade_shared PUBLIC ${CUCASCADE_PUBLIC_LINK_LIBS})
-  target_compile_features(cucascade_shared
-                          PUBLIC ${CUCASCADE_PUBLIC_COMPILE_FEATURES})
+  target_compile_features(cucascade_shared PUBLIC cxx_std_20)
 
   set_target_properties(
     cucascade_shared
@@ -239,14 +235,14 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/cuCascadeConfig.cmake
 # =============================================================================
 # Tests
 # =============================================================================
-if(BUILD_TESTS)
+if(CUCASCADE_BUILD_TESTS)
   enable_testing()
   add_subdirectory(test)
 endif()
 
 # =============================================================================
 # Benchmarks
 # =============================================================================
-if(BUILD_BENCHMARKS)
+if(CUCASCADE_BUILD_BENCHMARKS)
   add_subdirectory(benchmark)
 endif()

benchmark/CMakeLists.txt

@@ -45,8 +45,6 @@ set(BENCHMARK_SOURCES
 
 # Create benchmark executable
 add_executable(cucascade_benchmarks ${BENCHMARK_SOURCES})
-set_target_properties(cucascade_benchmarks PROPERTIES CXX_STANDARD 20
-                                                      CXX_STANDARD_REQUIRED ON)
 
 # Set include directories
 target_include_directories(

include/cucascade/memory/small_pinned_host_memory_resource.hpp

@@ -22,6 +22,7 @@
 #include <rmm/cuda_stream_view.hpp>
 #include <rmm/detail/cccl_adaptors.hpp>
 #include <rmm/mr/device_memory_resource.hpp>
+#include <rmm/version_config.hpp>
 
 #include <array>
 #include <cstddef>
@@ -137,9 +138,17 @@ class small_pinned_host_memory_resource : public rmm::mr::device_memory_resource
   std::vector<fixed_multiple_blocks_allocation> owned_allocations_;
 };
 
+// rmm::detail::polyfill::async_resource_with was removed in RMM 26.x;
+// use cuda::mr::resource_with directly for newer versions.
+#if RMM_VERSION_MAJOR >= 26
+static_assert(cuda::mr::resource_with<small_pinned_host_memory_resource,
+                                      cuda::mr::device_accessible,
+                                      cuda::mr::host_accessible>);
+#else
 static_assert(rmm::detail::polyfill::async_resource_with<small_pinned_host_memory_resource,
                                                          cuda::mr::device_accessible,
                                                          cuda::mr::host_accessible>);
+#endif
 
 }  // namespace memory
 }  // namespace cucascade

src/data/representation_converter.cpp

@@ -508,12 +508,12 @@ static memory::column_metadata plan_column_copy(const cudf::column_view& col,
     memory::column_metadata offsets_meta{};
     offsets_meta.type_id       = cudf::type_id::INT32;
     offsets_meta.num_rows      = 1;  // one offset value (0) for 0 strings
-    offsets_meta.null_count   = 0;
+    offsets_meta.null_count    = 0;
     offsets_meta.has_null_mask = false;
-    offsets_meta.has_data     = true;
-    offsets_meta.data_size    = sizeof(int32_t);
-    current_offset            = align_up_fast(current_offset, 8u);
-    offsets_meta.data_offset  = current_offset;
+    offsets_meta.has_data      = true;
+    offsets_meta.data_size     = sizeof(int32_t);
+    current_offset             = align_up_fast(current_offset, 8u);
+    offsets_meta.data_offset   = current_offset;
     current_offset += offsets_meta.data_size;
     offsets_meta.is_synthetic_empty_offsets = true;
     meta.children.push_back(std::move(offsets_meta));
@@ -558,11 +558,12 @@ static void collect_d2h_ops(const void* src,
 }
 
 /**
- * @brief Zero a region in the host allocation (used for synthetic STRING offsets with no device source).
+ * @brief Zero a region in the host allocation (used for synthetic STRING offsets with no device
+ * source).
  */
 static void zero_region(memory::fixed_multiple_blocks_allocation& alloc,
-                       std::size_t alloc_offset,
-                       std::size_t size)
+                        std::size_t alloc_offset,
+                        std::size_t size)
 {
   if (size == 0 || !alloc || alloc->size() == 0) { return; }
   const std::size_t block_size = alloc->block_size();
@@ -787,9 +788,7 @@ std::unique_ptr<idata_representation> convert_host_fast_to_gpu(
 {
   auto& fast_source      = source.cast<host_data_representation>();
   const auto& fast_table = fast_source.get_host_table();
-  if (!fast_table) {
-    throw std::runtime_error("convert_host_fast_to_gpu: host table is null");
-  }
+  if (!fast_table) { throw std::runtime_error("convert_host_fast_to_gpu: host table is null"); }
   if (!fast_table->allocation) {
     throw std::runtime_error("convert_host_fast_to_gpu: host table allocation is null");
   }

test/CMakeLists.txt

@@ -47,12 +47,8 @@ set(TEST_SOURCES
 
 # Create test executable
 add_executable(cucascade_tests ${TEST_SOURCES})
-set_target_properties(cucascade_tests PROPERTIES CXX_STANDARD 20
-                                                 CXX_STANDARD_REQUIRED ON
-                                                 CUDA_STANDARD 20
-                                                 CUDA_STANDARD_REQUIRED ON
-                                                 CUDA_ARCHITECTURES
-                                                 "${CMAKE_CUDA_ARCHITECTURES}")
+set_target_properties(cucascade_tests PROPERTIES CUDA_STANDARD 20
+                                                 CUDA_STANDARD_REQUIRED ON)
 
 # Set include directories
 target_include_directories(