CCCL C v2

.gitignore

@@ -1,6 +1,9 @@
 .idea/
 build*/
 .cache
+# Shared caches for the cu12/cu13 Python wheel builds (ccache + CPM source).
+.ccache/
+.cpm-cache/
 .aws
 .config
 _deps/catch2-src/

CMakeLists.txt

@@ -53,6 +53,11 @@ option(CCCL_ENABLE_THRUST "Enable the Thrust developer build." OFF)
 option(CCCL_ENABLE_TESTING "Enable CUDA C++ Core Library tests." OFF)
 option(CCCL_ENABLE_EXAMPLES "Enable CUDA C++ Core Library examples." OFF)
 option(CCCL_ENABLE_C_PARALLEL "Enable CUDA C Parallel Library." OFF)
+option(
+  CCCL_ENABLE_C_PARALLEL_V2
+  "Enable CUDA C Parallel Library v2 (HostJIT-based)."
+  OFF
+)
 option(CCCL_ENABLE_C_EXPERIMENTAL_STF "Enable CUDA C CUDASTF Library." OFF)
 option(CCCL_ENABLE_NVBENCH_HELPER "Enable the NVBench Helper Dev Build." OFF)
 
@@ -122,7 +127,11 @@ if (CCCL_ENABLE_UNSTABLE)
   add_subdirectory(cudax)
 endif()
 
-if (CCCL_ENABLE_C_PARALLEL OR CCCL_ENABLE_C_EXPERIMENTAL_STF)
+if (
+  CCCL_ENABLE_C_PARALLEL
+  OR CCCL_ENABLE_C_PARALLEL_V2
+  OR CCCL_ENABLE_C_EXPERIMENTAL_STF
+)
   add_subdirectory(c)
 endif()
 

CMakePresets.json

@@ -407,11 +407,12 @@
       }
     },
     {
-      "name": "cccl-c-parallel-hostjit",
-      "displayName": "CCCL C Parallel Library (HostJIT)",
-      "inherits": "cccl-c-parallel",
+      "name": "cccl-c-parallel-v2",
+      "displayName": "CCCL C Parallel Library v2 (HostJIT)",
+      "inherits": "base",
       "cacheVariables": {
-        "CCCL_C_Parallel_ENABLE_HOSTJIT": true
+        "CCCL_ENABLE_C_PARALLEL_V2": true,
+        "CCCL_C_Parallel_V2_ENABLE_TESTING": true
       }
     },
     {
@@ -647,8 +648,8 @@
       "configurePreset": "cccl-c-parallel"
     },
     {
-      "name": "cccl-c-parallel-hostjit",
-      "configurePreset": "cccl-c-parallel-hostjit"
+      "name": "cccl-c-parallel-v2",
+      "configurePreset": "cccl-c-parallel-v2"
     },
     {
       "name": "cccl-c-stf",
@@ -930,8 +931,8 @@
       "inherits": "base"
     },
     {
-      "name": "cccl-c-parallel-hostjit",
-      "configurePreset": "cccl-c-parallel-hostjit",
+      "name": "cccl-c-parallel-v2",
+      "configurePreset": "cccl-c-parallel-v2",
       "inherits": "base"
     },
     {

c/CMakeLists.txt

@@ -1,7 +1,19 @@
+if (CCCL_ENABLE_C_PARALLEL AND CCCL_ENABLE_C_PARALLEL_V2)
+  message(
+    FATAL_ERROR
+    "CCCL_ENABLE_C_PARALLEL and CCCL_ENABLE_C_PARALLEL_V2 are mutually exclusive. "
+    "v2 is the HostJIT-based successor of v1; pick one."
+  )
+endif()
+
 if (CCCL_ENABLE_C_PARALLEL)
   add_subdirectory(parallel)
 endif()
 
+if (CCCL_ENABLE_C_PARALLEL_V2)
+  add_subdirectory(parallel.v2)
+endif()
+
 if (CCCL_ENABLE_C_EXPERIMENTAL_STF)
   add_subdirectory(experimental/stf)
 endif()

c/parallel.v2/CMakeLists.txt

@@ -0,0 +1,110 @@
+cmake_minimum_required(VERSION 3.21)
+
+project(CCCL_C_Parallel_V2 LANGUAGES CUDA CXX C)
+
+# Bootstrap CCCL cmake helpers when building c/parallel.v2 in isolation
+# (i.e. not as a subdirectory of the CCCL super-project).
+if (NOT COMMAND cccl_configure_target)
+  # Repo root is two levels up from this file (c/parallel.v2 -> c -> cccl)
+  get_filename_component(
+    _cccl_root
+    "${CMAKE_CURRENT_SOURCE_DIR}/../.."
+    ABSOLUTE
+  )
+  set(CCCL_SOURCE_DIR "${_cccl_root}" CACHE PATH "CCCL repo root" FORCE)
+  set(
+    CCCL_BINARY_DIR
+    "${CMAKE_CURRENT_BINARY_DIR}"
+    CACHE PATH
+    "CCCL binary root"
+    FORCE
+  )
+  include("${_cccl_root}/cmake/CCCLUtilities.cmake")
+  include("${_cccl_root}/cmake/CCCLConfigureTarget.cmake")
+  include("${_cccl_root}/cmake/CCCLGetDependencies.cmake")
+  if (NOT TARGET cccl.compiler_interface)
+    add_library(cccl.compiler_interface INTERFACE)
+  endif()
+endif()
+
+option(CCCL_C_Parallel_V2_ENABLE_TESTING "Build cccl.c.parallel.v2 tests." OFF)
+
+set(
+  CCCL_C_PARALLEL_V2_LIBRARY_OUTPUT_DIRECTORY
+  ""
+  CACHE PATH
+  "Override output directory for the cccl.c.parallel.v2 library"
+)
+mark_as_advanced(CCCL_C_PARALLEL_V2_LIBRARY_OUTPUT_DIRECTORY)
+
+file(
+  GLOB_RECURSE srcs
+  RELATIVE "${CMAKE_CURRENT_LIST_DIR}"
+  CONFIGURE_DEPENDS
+  "src/*.cu"
+  "src/*.cpp"
+)
+# hostjit sources are built as a separate library
+list(FILTER srcs EXCLUDE REGEX "^src/hostjit/")
+# Editor lock/temp files
+list(FILTER srcs EXCLUDE REGEX "/\\.#")
+
+add_library(cccl.c.parallel.v2 SHARED ${srcs})
+set_property(TARGET cccl.c.parallel.v2 PROPERTY POSITION_INDEPENDENT_CODE ON)
+cccl_configure_target(cccl.c.parallel.v2 DIALECT 20)
+
+if (CCCL_C_PARALLEL_V2_LIBRARY_OUTPUT_DIRECTORY)
+  set_target_properties(
+    cccl.c.parallel.v2
+    PROPERTIES
+      LIBRARY_OUTPUT_DIRECTORY "${CCCL_C_PARALLEL_V2_LIBRARY_OUTPUT_DIRECTORY}"
+      ARCHIVE_OUTPUT_DIRECTORY "${CCCL_C_PARALLEL_V2_LIBRARY_OUTPUT_DIRECTORY}"
+      RUNTIME_OUTPUT_DIRECTORY "${CCCL_C_PARALLEL_V2_LIBRARY_OUTPUT_DIRECTORY}"
+  )
+endif()
+
+cccl_get_cub()
+cccl_get_cudatoolkit()
+cccl_get_thrust()
+
+add_subdirectory(src/hostjit)
+
+set_target_properties(cccl.c.parallel.v2 PROPERTIES CUDA_RUNTIME_LIBRARY STATIC)
+target_link_libraries(
+  cccl.c.parallel.v2
+  PRIVATE
+    cccl.compiler_interface
+    CUDA::cudart_static
+    CUDA::nvrtc # for nvrtcGetTypeName in src/util/types.h
+    CUDA::cuda_driver
+    CUB::CUB
+    Thrust::Thrust
+    cccl.c.parallel.v2.hostjit_lib # transitively brings in nvJitLink, nvfatbin, nvptxcompiler
+)
+
+if (WIN32)
+  target_link_libraries(cccl.c.parallel.v2 PRIVATE Dbghelp)
+endif()
+
+target_compile_definitions(
+  cccl.c.parallel.v2
+  PUBLIC CCCL_C_EXPERIMENTAL=1
+  PRIVATE #
+    NVRTC_GET_TYPE_NAME=1
+    CUB_DISABLE_CDP=1
+    CUB_DEFINE_RUNTIME_POLICIES
+)
+target_compile_options(
+  cccl.c.parallel.v2
+  PRIVATE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--extended-lambda>
+)
+
+target_include_directories(
+  cccl.c.parallel.v2 #
+  PUBLIC "include"
+  PRIVATE "src" "src/hostjit/include"
+)
+
+if (CCCL_C_Parallel_V2_ENABLE_TESTING)
+  add_subdirectory(test)
+endif()

c/parallel.v2/include/cccl/c/binary_search.h

@@ -0,0 +1,76 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of CUDA Experimental in CUDA C++ Core Libraries,
+// under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+// SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#pragma once
+
+#ifndef CCCL_C_EXPERIMENTAL
+#  error "C exposure is experimental and subject to change. Define CCCL_C_EXPERIMENTAL to acknowledge this notice."
+#endif // !CCCL_C_EXPERIMENTAL
+
+#include <cuda.h>
+#include <stdint.h>
+
+#include <cccl/c/extern_c.h>
+#include <cccl/c/types.h>
+
+CCCL_C_EXTERN_C_BEGIN
+
+typedef struct cccl_device_binary_search_build_result_t
+{
+  int cc;
+  void* cubin;
+  size_t cubin_size;
+  void* jit_compiler; // hostjit::JITCompiler*
+  void* binary_search_fn; // int(*)(void*, ull, void*, ull, void*, void*)
+} cccl_device_binary_search_build_result_t;
+
+CCCL_C_API CUresult cccl_device_binary_search_build(
+  cccl_device_binary_search_build_result_t* build,
+  cccl_binary_search_mode_t mode,
+  cccl_iterator_t d_data,
+  cccl_iterator_t d_values,
+  cccl_iterator_t d_out,
+  cccl_op_t op,
+  int cc_major,
+  int cc_minor,
+  const char* cub_path,
+  const char* thrust_path,
+  const char* libcudacxx_path,
+  const char* ctk_path);
+
+// Extended version with build configuration
+CCCL_C_API CUresult cccl_device_binary_search_build_ex(
+  cccl_device_binary_search_build_result_t* build,
+  cccl_binary_search_mode_t mode,
+  cccl_iterator_t d_data,
+  cccl_iterator_t d_values,
+  cccl_iterator_t d_out,
+  cccl_op_t op,
+  int cc_major,
+  int cc_minor,
+  const char* cub_path,
+  const char* thrust_path,
+  const char* libcudacxx_path,
+  const char* ctk_path,
+  cccl_build_config* config);
+
+CCCL_C_API CUresult cccl_device_binary_search(
+  cccl_device_binary_search_build_result_t build,
+  cccl_iterator_t d_data,
+  uint64_t num_items,
+  cccl_iterator_t d_values,
+  uint64_t num_values,
+  cccl_iterator_t d_out,
+  cccl_op_t op,
+  CUstream stream);
+
+CCCL_C_API CUresult cccl_device_binary_search_cleanup(cccl_device_binary_search_build_result_t* bld_ptr);
+
+CCCL_C_EXTERN_C_END

c/parallel.v2/include/cccl/c/extern_c.h

@@ -0,0 +1,23 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of CUDA Experimental in CUDA Core Compute Libraries,
+// under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+// SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#pragma once
+
+#ifdef __cplusplus
+
+#  define CCCL_C_EXTERN_C_BEGIN extern "C" {
+#  define CCCL_C_EXTERN_C_END   }
+
+#else
+
+#  define CCCL_C_EXTERN_C_BEGIN
+#  define CCCL_C_EXTERN_C_END
+
+#endif

c/parallel.v2/include/cccl/c/for.h

@@ -0,0 +1,63 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of CUDA Experimental in CUDA C++ Core Libraries,
+// under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+// SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#pragma once
+
+#ifndef CCCL_C_EXPERIMENTAL
+#  error "C exposure is experimental and subject to change. Define CCCL_C_EXPERIMENTAL to acknowledge this notice."
+#endif // !CCCL_C_EXPERIMENTAL
+
+#include <cuda.h>
+#include <stdint.h>
+
+#include <cccl/c/extern_c.h>
+#include <cccl/c/types.h>
+
+CCCL_C_EXTERN_C_BEGIN
+
+typedef struct cccl_device_for_build_result_t
+{
+  int cc;
+  void* cubin;
+  size_t cubin_size;
+  void* jit_compiler; // hostjit::JITCompiler*
+  void* for_fn; // int(*)(void*, unsigned long long, void*)
+} cccl_device_for_build_result_t;
+
+CCCL_C_API CUresult cccl_device_for_build(
+  cccl_device_for_build_result_t* build,
+  cccl_iterator_t d_data,
+  cccl_op_t op,
+  int cc_major,
+  int cc_minor,
+  const char* cub_path,
+  const char* thrust_path,
+  const char* libcudacxx_path,
+  const char* ctk_path);
+
+// Extended version with build configuration
+CCCL_C_API CUresult cccl_device_for_build_ex(
+  cccl_device_for_build_result_t* build,
+  cccl_iterator_t d_data,
+  cccl_op_t op,
+  int cc_major,
+  int cc_minor,
+  const char* cub_path,
+  const char* thrust_path,
+  const char* libcudacxx_path,
+  const char* ctk_path,
+  cccl_build_config* config);
+
+CCCL_C_API CUresult cccl_device_for(
+  cccl_device_for_build_result_t build, cccl_iterator_t d_data, uint64_t num_items, cccl_op_t op, CUstream stream);
+
+CCCL_C_API CUresult cccl_device_for_cleanup(cccl_device_for_build_result_t* bld_ptr);
+
+CCCL_C_EXTERN_C_END