Start to implement C bindings for CUDASTF (this will later be used in conjunction with cython to generate python support)

Author: caugonnet

c/CMakeLists.txt

@@ -1 +1,2 @@
 add_subdirectory(parallel)
+add_subdirectory(experimental/stf/)

c/experimental/stf/CMakeLists.txt

@@ -0,0 +1,60 @@
+cmake_minimum_required(VERSION 3.21)
+
+project(CCCL_C_EXPERIMENTAL_STF LANGUAGES CUDA CXX C)
+
+option(CCCL_C_EXPERIMENTAL_STF_ENABLE_TESTING "Build cccl.experimental.c.stf tests." OFF)
+
+# FIXME Ideally this would be handled by presets and install rules, but for now
+# consumers may override this to control the target location of cccl.c.experimental.stf.
+set(CCCL_C_EXPERIMENTAL_STF_LIBRARY_OUTPUT_DIRECTORY "" CACHE PATH "Override output directory for the cccl.c.experimental.stf library")
+mark_as_advanced(CCCL_C_PARALLEL_LIBRARY_OUTPUT_DIRECTORY)
+
+file(GLOB_RECURSE srcs
+  RELATIVE "${CMAKE_CURRENT_LIST_DIR}"
+  CONFIGURE_DEPENDS
+  "src/*.cu" "src/*.cuh"
+)
+
+add_library(cccl.c.experimental.stf SHARED ${srcs})
+set_property(TARGET cccl.c.experimental.stf PROPERTY POSITION_INDEPENDENT_CODE ON)
+cccl_configure_target(cccl.c.experimental.stf DIALECT 17)
+
+# Override the properties set by cccl_configure_target:
+if (CCCL_C_EXPERIMENTAL_STF_LIBRARY_OUTPUT_DIRECTORY)
+  set_target_properties(cccl.c.parallel PROPERTIES
+    LIBRARY_OUTPUT_DIRECTORY "${CCCL_C_EXPERIMENTAL_STF_LIBRARY_OUTPUT_DIRECTORY}"
+    ARCHIVE_OUTPUT_DIRECTORY "${CCCL_C_EXPERIMENTAL_STF_LIBRARY_OUTPUT_DIRECTORY}"
+  )
+endif()
+
+find_package(CUDAToolkit REQUIRED)
+set_target_properties(cccl.c.experimental.stf PROPERTIES CUDA_RUNTIME_LIBRARY STATIC)
+target_link_libraries(cccl.c.experimental.stf PRIVATE
+  CUDA::cudart_static
+  CUDA::nvrtc
+  CUDA::nvJitLink
+  CUDA::cuda_driver
+  cccl.compiler_interface_cpp20
+  cccl.c.parallel.jit_template
+  CUB::CUB
+  Thrust::Thrust
+  nlohmann_json::nlohmann_json
+)
+# target_compile_definitions(cccl.c.experimental.stf PUBLIC CCCL_C_EXPERIMENTAL=1)
+# target_compile_definitions(cccl.c.experimental.stf PRIVATE
+#   NVRTC_GET_TYPE_NAME=1
+#   CUB_DISABLE_CDP=1
+#   CUB_DEFINE_RUNTIME_POLICIES
+# )
+target_compile_options(cccl.c.experimental.stf PRIVATE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--extended-lambda>)
+
+target_include_directories(cccl.c.experimental.stf PUBLIC "include")
+target_include_directories(cccl.c.experimental.stf PRIVATE "src")
+
+if (CCCL_C_Parallel_ENABLE_TESTING)
+  add_subdirectory(test)
+endif()
+
+if (CCCL_C_Parallel_ENABLE_HEADER_TESTING)
+  include(cmake/CParallelHeaderTesting.cmake)
+endif()

c/experimental/stf/include/cccl/c/experimental/stf/stf.h

@@ -0,0 +1,22 @@
+// TODO use CCCL_C_EXTERN_C_BEGIN/CCCL_C_EXTERN_C_END
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+typedef struct stf_ctx_handle stf_ctx_handle;
+
+void stf_ctx_create(stf_ctx_handle *handle);
+void stf_ctx_finalize(stf_ctx_handle *handle);
+
+struct stf_task_handle {
+    void *handle;
+};
+
+struct stf_logical_data_handle {
+    void *handle;
+};
+
+#ifdef __cplusplus
+}
+#endif

c/experimental/stf/src/stf.cu

@@ -0,0 +1,26 @@
+#include <cccl/c/experimental/stf/stf.h>
+//#include <cccl/c/parallel/include/cccl/c/extern_c.h>
+#include <cuda/experimental/stf.cuh>
+
+using namespace cuda::experimental::stf;
+
+extern "C"
+{
+
+struct stf_ctx_handle {
+    context *ctx;
+};
+
+void stf_ctx_create(stf_ctx_handle *handle)
+{
+    return new context{};
+}
+
+void stf_ctx_finalize(stf_ctx_handle *handle)
+{
+    if (handle) {
+        handle->finalize();
+    }
+}
+
+}