Fix CCCL C headers to be compileable by C compiler (#3885)

* Closes gh-3882 This change ensures that cccl/c/parallel headers are compilable by C compiler. 1. Corrected typedef struct and typedef enum so that C types have the same names as C++. That is `typdef struct name {...} otherName;` was changed to `typedef struct name {...} name;` 2. `noexcept` qualifier was removed from function declarations. 3. Implementation was changed to move `noexcept` implementations into a namespace, and implement publicly declared functions as thin shims calling these noexcept functions from a namespace. 4. Adds c/parallel/test/test_header.c This is a C file which includes all public headers, and contains a trivial `int main(void)` function. The file is compiled using C compiler to produce an object file. Hence it does not result in a dedicate test to run, but any non C-compliant changes to these headers would break test compilation step. N.B.: Every new header file must be included in `test_headers.c` file for it to be tested. 5. Piggy-backing on this PR, added runtime error condition checking in test_main.cpp to verify that `cudaSetDevice` call did not report an error. 6. Modified cccl/c/parallel CMake project list of languages to include C. * Revert using noexcept implementations and call them from C declared function Instead simply remove use of noexcept in C++ definitions of these functions per Bernhard's suggestion
NVIDIA · Feb 21, 2025 · fd0cd24 · fd0cd24
1 parent ca004f5
commit fd0cd24
Show file tree

Hide file tree

Showing 15 changed files with 146 additions and 121 deletions.
diff --git a/c/parallel/CMakeLists.txt b/c/parallel/CMakeLists.txt
@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.21)
 
-project(CCCL_C_Parallel LANGUAGES CUDA CXX)
+project(CCCL_C_Parallel LANGUAGES CUDA CXX C)
 
 option(CCCL_C_Parallel_ENABLE_TESTING "Build CUDA Experimental's tests." OFF)
 option(CCCL_C_Parallel_ENABLE_HEADER_TESTING "Build CUDA Experimental's standalone headers." OFF)

diff --git a/c/parallel/include/cccl/c/for.h b/c/parallel/include/cccl/c/for.h
@@ -21,14 +21,14 @@
 
 CCCL_C_EXTERN_C_BEGIN
 
-struct cccl_device_for_build_result_t
+typedef struct cccl_device_for_build_result_t
 {
   int cc;
   void* cubin;
   size_t cubin_size;
   CUlibrary library;
   CUkernel static_kernel;
-};
+} cccl_device_for_build_result_t;
 
 CCCL_C_API CUresult cccl_device_for_build(
   cccl_device_for_build_result_t* build,
@@ -39,14 +39,10 @@ CCCL_C_API CUresult cccl_device_for_build(
   const char* cub_path,
   const char* thrust_path,
   const char* libcudacxx_path,
-  const char* ctk_path) noexcept;
+  const char* ctk_path);
 
 CCCL_C_API CUresult cccl_device_for(
-  cccl_device_for_build_result_t build,
-  cccl_iterator_t d_data,
-  int64_t num_items,
-  cccl_op_t op,
-  CUstream stream) noexcept;
+  cccl_device_for_build_result_t build, cccl_iterator_t d_data, int64_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);
 

diff --git a/c/parallel/include/cccl/c/merge_sort.h b/c/parallel/include/cccl/c/merge_sort.h
@@ -21,7 +21,7 @@
 
 CCCL_C_EXTERN_C_BEGIN
 
-struct cccl_device_merge_sort_build_result_t
+typedef struct cccl_device_merge_sort_build_result_t
 {
   int cc;
   void* cubin;
@@ -30,7 +30,7 @@ struct cccl_device_merge_sort_build_result_t
   CUkernel block_sort_kernel;
   CUkernel partition_kernel;
   CUkernel merge_kernel;
-};
+} cccl_device_merge_sort_build_result_t;
 
 CCCL_C_API CUresult cccl_device_merge_sort_build(
   cccl_device_merge_sort_build_result_t* build,
@@ -44,7 +44,7 @@ CCCL_C_API CUresult cccl_device_merge_sort_build(
   const char* cub_path,
   const char* thrust_path,
   const char* libcudacxx_path,
-  const char* ctk_path) noexcept;
+  const char* ctk_path);
 
 CCCL_C_API CUresult cccl_device_merge_sort(
   cccl_device_merge_sort_build_result_t build,
@@ -56,8 +56,8 @@ CCCL_C_API CUresult cccl_device_merge_sort(
   cccl_iterator_t d_out_items,
   unsigned long long num_items,
   cccl_op_t op,
-  CUstream stream) noexcept;
+  CUstream stream);
 
-CCCL_C_API CUresult cccl_device_merge_sort_cleanup(cccl_device_merge_sort_build_result_t* bld_ptr) noexcept;
+CCCL_C_API CUresult cccl_device_merge_sort_cleanup(cccl_device_merge_sort_build_result_t* bld_ptr);
 
 CCCL_C_EXTERN_C_END
diff --git a/c/parallel/include/cccl/c/reduce.h b/c/parallel/include/cccl/c/reduce.h
@@ -21,7 +21,7 @@
 
 CCCL_C_EXTERN_C_BEGIN
 
-struct cccl_device_reduce_build_result_t
+typedef struct cccl_device_reduce_build_result_t
 {
   int cc;
   void* cubin;
@@ -31,7 +31,7 @@ struct cccl_device_reduce_build_result_t
   CUkernel single_tile_kernel;
   CUkernel single_tile_second_kernel;
   CUkernel reduction_kernel;
-};
+} cccl_device_reduce_build_result_t;
 
 // TODO return a union of nvtx/cuda/nvrtc errors or a string?
 CCCL_C_API CUresult cccl_device_reduce_build(
@@ -45,7 +45,7 @@ CCCL_C_API CUresult cccl_device_reduce_build(
   const char* cub_path,
   const char* thrust_path,
   const char* libcudacxx_path,
-  const char* ctk_path) noexcept;
+  const char* ctk_path);
 
 CCCL_C_API CUresult cccl_device_reduce(
   cccl_device_reduce_build_result_t build,
@@ -56,8 +56,8 @@ CCCL_C_API CUresult cccl_device_reduce(
   unsigned long long num_items,
   cccl_op_t op,
   cccl_value_t init,
-  CUstream stream) noexcept;
+  CUstream stream);
 
-CCCL_C_API CUresult cccl_device_reduce_cleanup(cccl_device_reduce_build_result_t* bld_ptr) noexcept;
+CCCL_C_API CUresult cccl_device_reduce_cleanup(cccl_device_reduce_build_result_t* bld_ptr);
 
 CCCL_C_EXTERN_C_END
diff --git a/c/parallel/include/cccl/c/scan.h b/c/parallel/include/cccl/c/scan.h
@@ -21,7 +21,7 @@
 
 CCCL_C_EXTERN_C_BEGIN
 
-struct cccl_device_scan_build_result_t
+typedef struct cccl_device_scan_build_result_t
 {
   int cc;
   void* cubin;
@@ -32,10 +32,10 @@ struct cccl_device_scan_build_result_t
   CUkernel scan_kernel;
   size_t description_bytes_per_tile;
   size_t payload_bytes_per_tile;
-};
+} cccl_device_scan_build_result_t;
 
 CCCL_C_API CUresult cccl_device_scan_build(
-  cccl_device_scan_build_result_t* build,
+  cccl_device_scan_build_result_t* build_ptr,
   cccl_iterator_t d_in,
   cccl_iterator_t d_out,
   cccl_op_t op,
@@ -45,7 +45,7 @@ CCCL_C_API CUresult cccl_device_scan_build(
   const char* cub_path,
   const char* thrust_path,
   const char* libcudacxx_path,
-  const char* ctk_path) noexcept;
+  const char* ctk_path);
 
 CCCL_C_API CUresult cccl_device_scan(
   cccl_device_scan_build_result_t build,
@@ -56,8 +56,8 @@ CCCL_C_API CUresult cccl_device_scan(
   unsigned long long num_items,
   cccl_op_t op,
   cccl_value_t init,
-  CUstream stream) noexcept;
+  CUstream stream);
 
-CCCL_C_API CUresult cccl_device_scan_cleanup(cccl_device_scan_build_result_t* bld_ptr) noexcept;
+CCCL_C_API CUresult cccl_device_scan_cleanup(cccl_device_scan_build_result_t* bld_ptr);
 
 CCCL_C_EXTERN_C_END
diff --git a/c/parallel/include/cccl/c/segmented_reduce.h b/c/parallel/include/cccl/c/segmented_reduce.h
@@ -21,32 +21,31 @@
 
 CCCL_C_EXTERN_C_BEGIN
 
-struct cccl_device_segmented_reduce_build_result_t
+typedef struct cccl_device_segmented_reduce_build_result_t
 {
   int cc;
   void* cubin;
   size_t cubin_size;
   CUlibrary library;
   unsigned long long accumulator_size;
-  unsigned long long offset_size;
   CUkernel segmented_reduce_kernel;
-};
+} cccl_device_segmented_reduce_build_result_t;
 
 // TODO return a union of nvtx/cuda/nvrtc errors or a string?
 CCCL_C_API CUresult cccl_device_segmented_reduce_build(
   cccl_device_segmented_reduce_build_result_t* build,
   cccl_iterator_t d_in,
   cccl_iterator_t d_out,
-  cccl_iterator_t begin_offset_it,
-  cccl_iterator_t end_offset_it,
+  cccl_iterator_t begin_offset_in,
+  cccl_iterator_t end_offset_in,
   cccl_op_t op,
   cccl_value_t init,
   int cc_major,
   int cc_minor,
   const char* cub_path,
   const char* thrust_path,
   const char* libcudacxx_path,
-  const char* ctk_path) noexcept;
+  const char* ctk_path);
 
 CCCL_C_API CUresult cccl_device_segmented_reduce(
   cccl_device_segmented_reduce_build_result_t build,
@@ -55,12 +54,12 @@ CCCL_C_API CUresult cccl_device_segmented_reduce(
   cccl_iterator_t d_in,
   cccl_iterator_t d_out,
   unsigned long long num_offsets,
-  cccl_iterator_t start_offset_it,
-  cccl_iterator_t end_offset_it,
+  cccl_iterator_t start_offset_in,
+  cccl_iterator_t end_offset_in,
   cccl_op_t op,
   cccl_value_t init,
-  CUstream stream) noexcept;
+  CUstream stream);
 
-CCCL_C_API CUresult cccl_device_segmented_reduce_cleanup(cccl_device_segmented_reduce_build_result_t* bld_ptr) noexcept;
+CCCL_C_API CUresult cccl_device_segmented_reduce_cleanup(cccl_device_segmented_reduce_build_result_t* bld_ptr);
 
 CCCL_C_EXTERN_C_END
diff --git a/c/parallel/include/cccl/c/types.h b/c/parallel/include/cccl/c/types.h
@@ -37,20 +37,20 @@ typedef enum cccl_type_enum
   CCCL_FLOAT32 = 8,
   CCCL_FLOAT64 = 9,
   CCCL_STORAGE = 10
-} ccclType;
+} cccl_type_enum;
 
 typedef struct cccl_type_info
 {
   int size;
   int alignment;
   cccl_type_enum type;
-} ccclTypeInfo;
+} cccl_type_info;
 
 typedef enum cccl_op_kind_t
 {
   CCCL_STATELESS = 0,
   CCCL_STATEFUL  = 1
-} ccclOpKind;
+} cccl_op_kind_t;
 
 typedef struct cccl_op_t
 {
@@ -61,19 +61,19 @@ typedef struct cccl_op_t
   int size;
   int alignment;
   void* state;
-} ccclOp;
+} cccl_op_t;
 
 typedef enum cccl_iterator_kind_t
 {
   CCCL_POINTER  = 0,
   CCCL_ITERATOR = 1
-} ccclIteratorKind;
+} cccl_iterator_kind_t;
 
 typedef struct cccl_value_t
 {
   cccl_type_info type;
   void* state;
-} ccclValue;
+} cccl_value_t;
 
 typedef struct cccl_iterator_t
 {
@@ -84,6 +84,6 @@ typedef struct cccl_iterator_t
   cccl_op_t dereference;
   cccl_type_info value_type;
   void* state;
-} ccclIterator;
+} cccl_iterator_t;
 
 CCCL_C_EXTERN_C_END
diff --git a/c/parallel/src/for.cu b/c/parallel/src/for.cu
@@ -66,15 +66,15 @@ static std::string get_device_for_kernel_name()
 }
 
 CUresult cccl_device_for_build(
-  cccl_device_for_build_result_t* build,
+  cccl_device_for_build_result_t* build_ptr,
   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) noexcept
+  const char* ctk_path)
 {
   CUresult error = CUDA_SUCCESS;
 
@@ -124,12 +124,12 @@ CUresult cccl_device_for_build(
       result = cl.finalize_program(num_lto_args, lopts);
     }
 
-    cuLibraryLoadData(&build->library, result.data.get(), nullptr, nullptr, 0, nullptr, nullptr, 0);
-    check(cuLibraryGetKernel(&build->static_kernel, build->library, lowered_name.c_str()));
+    cuLibraryLoadData(&build_ptr->library, result.data.get(), nullptr, nullptr, 0, nullptr, nullptr, 0);
+    check(cuLibraryGetKernel(&build_ptr->static_kernel, build_ptr->library, lowered_name.c_str()));
 
-    build->cc         = cc;
-    build->cubin      = (void*) result.data.release();
-    build->cubin_size = result.size;
+    build_ptr->cc         = cc;
+    build_ptr->cubin      = (void*) result.data.release();
+    build_ptr->cubin_size = result.size;
   }
   catch (...)
   {
@@ -139,11 +139,7 @@ CUresult cccl_device_for_build(
 }
 
 CUresult cccl_device_for(
-  cccl_device_for_build_result_t build,
-  cccl_iterator_t d_data,
-  int64_t num_items,
-  cccl_op_t op,
-  CUstream stream) noexcept
+  cccl_device_for_build_result_t build, cccl_iterator_t d_data, int64_t num_items, cccl_op_t op, CUstream stream)
 {
   bool pushed    = false;
   CUresult error = CUDA_SUCCESS;
@@ -167,17 +163,17 @@ CUresult cccl_device_for(
   return error;
 }
 
-CUresult cccl_device_for_cleanup(cccl_device_for_build_result_t* bld_ptr)
+CUresult cccl_device_for_cleanup(cccl_device_for_build_result_t* build_ptr)
 {
   try
   {
-    if (bld_ptr == nullptr)
+    if (build_ptr == nullptr)
     {
       return CUDA_ERROR_INVALID_VALUE;
     }
 
-    std::unique_ptr<char[]> cubin(reinterpret_cast<char*>(bld_ptr->cubin));
-    check(cuLibraryUnload(bld_ptr->library));
+    std::unique_ptr<char[]> cubin(reinterpret_cast<char*>(build_ptr->cubin));
+    check(cuLibraryUnload(build_ptr->library));
   }
   catch (...)
   {