Coverage for cuda/core/_resource_handles.pyx: 97.94%

1# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. (empty)

2# 

3# SPDX-License-Identifier: Apache-2.0 

4  

5# This module compiles _cpp/resource_handles.cpp into a shared library. 

6# Consumer modules cimport the functions declared in _resource_handles.pxd. 

7# Since there is only one copy of the C++ code (in this .so), all static and 

8# thread-local state is shared correctly across all consumer modules. 

9# 

10# The cdef extern from declarations below satisfy the .pxd declarations directly, 

11# without needing separate wrapper functions. 

12  

13from cpython.pycapsule cimport PyCapsule_GetName, PyCapsule_GetPointer 

14from libc.stddef cimport size_t 

15  

16from cuda.bindings cimport cydriver 

17from cuda.bindings cimport cynvrtc 

18from cuda.bindings cimport cynvvm 

19from cuda.bindings cimport cynvjitlink 

20  

21import cuda.bindings.cydriver as cydriver (empty)

22import cuda.bindings.cynvrtc as cynvrtc (empty)

23import cuda.bindings.cynvvm as cynvvm (empty)

24import cuda.bindings.cynvjitlink as cynvjitlink (empty)

25  

26# ============================================================================= 

27# C++ function declarations (non-inline, implemented in resource_handles.cpp) 

28# 

29# These declarations satisfy the cdef function declarations in _resource_handles.pxd. 

30# Consumer modules cimport these functions and calls go through this .so. 

31# ============================================================================= 

32  

33cdef extern from "_cpp/resource_handles.hpp" namespace "cuda_core": 

34 # Thread-local error handling 

35 cydriver.CUresult get_last_error "cuda_core::get_last_error" () noexcept nogil 

36 cydriver.CUresult peek_last_error "cuda_core::peek_last_error" () noexcept nogil 

37 void clear_last_error "cuda_core::clear_last_error" () noexcept nogil 

38  

39 # Context handles 

40 ContextHandle create_context_handle_ref "cuda_core::create_context_handle_ref" ( 

41 cydriver.CUcontext ctx) except+ nogil 

42 ContextHandle create_context_handle_from_green_ctx "cuda_core::create_context_handle_from_green_ctx" ( 

43 const GreenCtxHandle& h_green_ctx) except+ nogil 

44 GreenCtxHandle get_context_green_ctx "cuda_core::get_context_green_ctx" ( 

45 const ContextHandle& h) noexcept nogil 

46 GreenCtxHandle create_green_ctx_handle "cuda_core::create_green_ctx_handle" ( 

47 cydriver.CUdevResource* resources, unsigned int nbResources, 

48 cydriver.CUdevice dev, unsigned int flags) except+ nogil 

49 GreenCtxHandle create_green_ctx_handle_ref "cuda_core::create_green_ctx_handle_ref" ( 

50 cydriver.CUgreenCtx ctx) except+ nogil 

51 ContextHandle get_primary_context "cuda_core::get_primary_context" ( 

52 int device_id) except+ nogil 

53 ContextHandle get_current_context "cuda_core::get_current_context" () except+ nogil 

54 cydriver.CUresult context_synchronize "cuda_core::context_synchronize" ( 

55 const ContextHandle& h_context) noexcept nogil 

56 cydriver.CUresult context_get_stream_priority_range "cuda_core::context_get_stream_priority_range" ( 

57 const ContextHandle& h_context, 

58 int* least_priority, 

59 int* greatest_priority) noexcept nogil 

60  

61 # Stream handles 

62 StreamHandle create_stream_handle "cuda_core::create_stream_handle" ( 

63 const ContextHandle& h_ctx, unsigned int flags, int priority) except+ nogil 

64 StreamHandle create_stream_handle_ref "cuda_core::create_stream_handle_ref" ( 

65 cydriver.CUstream stream) except+ nogil 

66 StreamHandle create_stream_handle_with_owner "cuda_core::create_stream_handle_with_owner" ( 

67 cydriver.CUstream stream, object owner) except+ nogil 

68 void py_object_user_object_destroy "cuda_core::py_object_user_object_destroy" ( 

69 void* py_object) noexcept nogil 

70 void initialize_deferred_cleanup "cuda_core::initialize_deferred_cleanup" () except+ 

71 void retry_deferred_cleanup "cuda_core::retry_deferred_cleanup" () noexcept 

72 ContextHandle get_stream_context "cuda_core::get_stream_context" ( 

73 const StreamHandle& h) noexcept nogil 

74 StreamHandle get_legacy_stream "cuda_core::get_legacy_stream" () except+ nogil 

75 StreamHandle get_per_thread_stream "cuda_core::get_per_thread_stream" () except+ nogil 

76 StreamHandle create_context_bound_legacy_stream "cuda_core::create_context_bound_legacy_stream" ( 

77 const ContextHandle& h_context) except+ nogil 

78  

79 # Event handles (note: _create_event_handle* are internal due to C++ overloading) 

80 EventHandle create_event_handle "cuda_core::create_event_handle" ( 

81 const ContextHandle& h_ctx, unsigned int flags, 

82 bint timing_enabled, bint is_blocking_sync, 

83 bint ipc_enabled, int device_id) except+ nogil 

84 EventHandle create_event_handle_for_stream "cuda_core::create_event_handle_for_stream" ( 

85 cydriver.CUstream stream, unsigned int flags) except+ nogil 

86 EventHandle create_event_handle_ref "cuda_core::create_event_handle_ref" ( 

87 cydriver.CUevent event) except+ nogil 

88 EventHandle create_event_handle_ipc "cuda_core::create_event_handle_ipc" ( 

89 const cydriver.CUipcEventHandle& ipc_handle, bint is_blocking_sync) except+ nogil 

90  

91 # Event metadata getters 

92 bint get_event_timing_enabled "cuda_core::get_event_timing_enabled" ( 

93 const EventHandle& h) noexcept nogil 

94 bint get_event_is_blocking_sync "cuda_core::get_event_is_blocking_sync" ( 

95 const EventHandle& h) noexcept nogil 

96 bint get_event_ipc_enabled "cuda_core::get_event_ipc_enabled" ( 

97 const EventHandle& h) noexcept nogil 

98 int get_event_device_id "cuda_core::get_event_device_id" ( 

99 const EventHandle& h) noexcept nogil 

100 ContextHandle get_event_context "cuda_core::get_event_context" ( 

101 const EventHandle& h) noexcept nogil 

102  

103 # Memory pool handles 

104 MemoryPoolHandle create_mempool_handle "cuda_core::create_mempool_handle" ( 

105 const cydriver.CUmemPoolProps& props) except+ nogil 

106 MemoryPoolHandle create_mempool_handle_ref "cuda_core::create_mempool_handle_ref" ( 

107 cydriver.CUmemoryPool pool) except+ nogil 

108 MemoryPoolHandle get_device_mempool "cuda_core::get_device_mempool" ( 

109 int device_id) except+ nogil 

110 MemoryPoolHandle create_mempool_handle_ipc "cuda_core::create_mempool_handle_ipc" ( 

111 int fd, cydriver.CUmemAllocationHandleType handle_type) except+ nogil 

112  

113 # Device pointer handles 

114 DevicePtrHandle deviceptr_alloc_from_pool "cuda_core::deviceptr_alloc_from_pool" ( 

115 size_t size, const MemoryPoolHandle& h_pool, const StreamHandle& h_stream) except+ nogil 

116 DevicePtrHandle deviceptr_alloc_async "cuda_core::deviceptr_alloc_async" ( 

117 size_t size, const StreamHandle& h_stream) except+ nogil 

118 cydriver.CUresult deviceptr_alloc_raw "cuda_core::deviceptr_alloc_raw" ( 

119 cydriver.CUdeviceptr* ptr, size_t size, const ContextHandle& h_context) noexcept nogil 

120 DevicePtrHandle deviceptr_alloc_host "cuda_core::deviceptr_alloc_host" (size_t size) except+ nogil 

121 DevicePtrHandle deviceptr_create_ref "cuda_core::deviceptr_create_ref" ( 

122 cydriver.CUdeviceptr ptr) except+ nogil 

123 DevicePtrHandle deviceptr_create_with_owner "cuda_core::deviceptr_create_with_owner" ( 

124 cydriver.CUdeviceptr ptr, object owner) except+ nogil 

125 DevicePtrHandle deviceptr_create_mapped_graphics "cuda_core::deviceptr_create_mapped_graphics" ( 

126 cydriver.CUdeviceptr ptr, 

127 const GraphicsResourceHandle& h_resource, 

128 const StreamHandle& h_stream) except+ nogil 

129  

130 # MR deallocation callback 

131 void register_mr_dealloc_callback "cuda_core::register_mr_dealloc_callback" ( 

132 MRDeallocCallback cb) noexcept 

133 DevicePtrHandle deviceptr_create_with_mr "cuda_core::deviceptr_create_with_mr" ( 

134 cydriver.CUdeviceptr ptr, size_t size, object mr) except+ nogil 

135  

136 DevicePtrHandle deviceptr_import_ipc "cuda_core::deviceptr_import_ipc" ( 

137 const MemoryPoolHandle& h_pool, const void* export_data, const StreamHandle& h_stream) except+ nogil 

138 StreamHandle deallocation_stream "cuda_core::deallocation_stream" ( 

139 const DevicePtrHandle& h) noexcept nogil 

140 cydriver.CUresult set_deallocation_stream "cuda_core::set_deallocation_stream" ( 

141 const DevicePtrHandle& h, const StreamHandle& h_stream) noexcept nogil 

142  

143 # Library handles 

144 LibraryHandle create_library_handle_from_file "cuda_core::create_library_handle_from_file" ( 

145 const char* path) except+ nogil 

146 LibraryHandle create_library_handle_from_data "cuda_core::create_library_handle_from_data" ( 

147 const void* data) except+ nogil 

148 LibraryHandle create_library_handle_ref "cuda_core::create_library_handle_ref" ( 

149 cydriver.CUlibrary library) except+ nogil 

150  

151 # Kernel handles 

152 KernelHandle create_kernel_handle "cuda_core::create_kernel_handle" ( 

153 const LibraryHandle& h_library, const char* name) except+ nogil 

154 KernelHandle create_kernel_handle_ref "cuda_core::create_kernel_handle_ref" ( 

155 cydriver.CUkernel kernel) except+ nogil 

156 LibraryHandle get_kernel_library "cuda_core::get_kernel_library" ( 

157 const KernelHandle& h) noexcept nogil 

158  

159 # Graph handles 

160 GraphHandle create_graph_handle "cuda_core::create_graph_handle" ( 

161 cydriver.CUgraph graph) except+ nogil 

162 GraphHandle create_child_graph_handle "cuda_core::create_child_graph_handle" ( 

163 cydriver.CUgraph child_graph, const GraphHandle& h_parent, 

164 cydriver.CUgraphNode owner_node) except+ nogil 

165  

166 # Graph node attachments 

167 OpaqueHandle make_opaque_py "cuda_core::make_opaque_py" (object obj) except+ 

168 OpaqueHandle make_opaque_malloc "cuda_core::make_opaque_malloc" (void* buf) except+ 

169 cydriver.CUresult graph_get_attachment "cuda_core::graph_get_attachment" ( 

170 const GraphHandle& h_graph, cydriver.CUgraphNode node, 

171 OpaqueHandle* owner0, OpaqueHandle* owner1) except+ 

172 cydriver.CUresult graph_prepare_attachment "cuda_core::graph_prepare_attachment" ( 

173 const GraphHandle& h_graph, OpaqueHandle owner0, OpaqueHandle owner1, 

174 PreparedAttachment* out_prepared) except+ 

175 cydriver.CUresult graph_commit_attachment "cuda_core::graph_commit_attachment" ( 

176 PreparedAttachment& prepared, cydriver.CUgraphNode node) except+ 

177 cydriver.CUresult graph_clone_attachments "cuda_core::graph_clone_attachments" ( 

178 const GraphHandle& h_clone, const GraphHandle& h_source) except+ 

179 cydriver.CUresult graph_prepare_child_graph_update "cuda_core::graph_prepare_child_graph_update" ( 

180 const GraphHandle& h_parent, const GraphHandle& h_old_child, 

181 cydriver.CUgraphNode owner_node, const GraphHandle& h_source, 

182 PreparedChildGraphUpdate* out_prepared) except+ 

183 cydriver.CUresult graph_commit_child_graph_update "cuda_core::graph_commit_child_graph_update" ( 

184 PreparedChildGraphUpdate& prepared, GraphHandle* out_child) except+ 

185 void invalidate_child_graph_state "cuda_core::invalidate_child_graph_state" ( 

186 const GraphHandle& h_parent, cydriver.CUgraphNode owner_node) noexcept 

187  

188 # Graph exec handles 

189 GraphExecHandle create_graph_exec_handle "cuda_core::create_graph_exec_handle" ( 

190 const GraphHandle& h_source, 

191 cydriver.CUDA_GRAPH_INSTANTIATE_PARAMS* params) except+ 

192 cydriver.CUresult graph_exec_update "cuda_core::graph_exec_update" ( 

193 const GraphExecHandle& h_exec, 

194 const GraphHandle& h_source, 

195 cydriver.CUgraphExecUpdateResultInfo* result_info) except+ 

196 cydriver.CUresult graph_prepare_exec_attachment "cuda_core::graph_prepare_exec_attachment" ( 

197 const GraphExecHandle& h_exec, 

198 OpaqueHandle owner0, 

199 OpaqueHandle owner1, 

200 PreparedExecAttachment* out_prepared) except+ 

201 void graph_commit_exec_attachment "cuda_core::graph_commit_exec_attachment" ( 

202 PreparedExecAttachment& prepared) noexcept 

203  

204 # Graph node handles 

205 GraphNodeHandle create_graph_node_handle "cuda_core::create_graph_node_handle" ( 

206 cydriver.CUgraphNode node, const GraphHandle& h_graph) except+ nogil 

207 GraphHandle graph_node_get_graph "cuda_core::graph_node_get_graph" ( 

208 const GraphNodeHandle& h) noexcept nogil 

209 void invalidate_graph_node "cuda_core::invalidate_graph_node" ( 

210 const GraphNodeHandle& h) noexcept nogil 

211  

212 # Graphics resource handles 

213 GraphicsResourceHandle create_graphics_resource_handle "cuda_core::create_graphics_resource_handle" ( 

214 cydriver.CUgraphicsResource resource) except+ nogil 

215  

216 # NVRTC Program handles 

217 NvrtcProgramHandle create_nvrtc_program_handle "cuda_core::create_nvrtc_program_handle" ( 

218 cynvrtc.nvrtcProgram prog) except+ nogil 

219 NvrtcProgramHandle create_nvrtc_program_handle_ref "cuda_core::create_nvrtc_program_handle_ref" ( 

220 cynvrtc.nvrtcProgram prog) except+ nogil 

221  

222 # NVVM Program handles 

223 NvvmProgramHandle create_nvvm_program_handle "cuda_core::create_nvvm_program_handle" ( 

224 cynvvm.nvvmProgram prog) except+ nogil 

225 NvvmProgramHandle create_nvvm_program_handle_ref "cuda_core::create_nvvm_program_handle_ref" ( 

226 cynvvm.nvvmProgram prog) except+ nogil 

227  

228 # nvJitLink handles 

229 NvJitLinkHandle create_nvjitlink_handle "cuda_core::create_nvjitlink_handle" ( 

230 cynvjitlink.nvJitLinkHandle handle) except+ nogil 

231 NvJitLinkHandle create_nvjitlink_handle_ref "cuda_core::create_nvjitlink_handle_ref" ( 

232 cynvjitlink.nvJitLinkHandle handle) except+ nogil 

233  

234 # cuLink handles 

235 CuLinkHandle create_culink_handle "cuda_core::create_culink_handle" ( 

236 cydriver.CUlinkState state) except+ nogil 

237 CuLinkHandle create_culink_handle_ref "cuda_core::create_culink_handle_ref" ( 

238 cydriver.CUlinkState state) except+ nogil 

239  

240 # File descriptor handles 

241 FileDescriptorHandle create_fd_handle "cuda_core::create_fd_handle" ( 

242 int fd) except+ nogil 

243 FileDescriptorHandle create_fd_handle_ref "cuda_core::create_fd_handle_ref" ( 

244 int fd) except+ nogil 

245  

246 # SM resource split (13.1+ wrapper — avoids direct cydriver cimport) 

247 # groupParams is void* to avoid referencing CU_DEV_SM_RESOURCE_GROUP_PARAMS 

248 # (which doesn't exist in cuda-bindings 13.0 .pxd). The C++ side casts it. 

249 cydriver.CUresult sm_resource_split "cuda_core::sm_resource_split" ( 

250 cydriver.CUdevResource* result, unsigned int nbGroups, 

251 const cydriver.CUdevResource* input, cydriver.CUdevResource* remainder, 

252 unsigned int flags, void* groupParams) nogil 

253 bint has_sm_resource_split "cuda_core::has_sm_resource_split" () noexcept nogil 

254  

255 # cuMemcpyWithAttributesAsync (13.2+ wrapper — avoids direct cydriver cimport) 

256 # attr is void* to avoid referencing CUmemcpyAttributes (absent from 

257 # cuda-bindings built against CUDA < 12.8). The C++ side casts it. 

258 cydriver.CUresult memcpy_with_attributes_async "cuda_core::memcpy_with_attributes_async" ( 

259 cydriver.CUdeviceptr dst, cydriver.CUdeviceptr src, size_t size, 

260 void* attr, cydriver.CUstream hStream) nogil 

261 bint has_memcpy_with_attributes_async "cuda_core::has_memcpy_with_attributes_async" () noexcept nogil 

262  

263 # Array / mipmapped-array / texture / surface handles (PR #467) 

264 OpaqueArrayHandle create_array_handle "cuda_core::create_array_handle" ( 

265 const ContextHandle& h_context, const cydriver.CUDA_ARRAY3D_DESCRIPTOR& desc) except+ nogil 

266 OpaqueArrayHandle create_array_handle_ref "cuda_core::create_array_handle_ref" ( 

267 cydriver.CUarray arr) except+ nogil 

268 OpaqueArrayHandle create_array_handle_owning "cuda_core::create_array_handle_owning" ( 

269 cydriver.CUarray arr) except+ nogil 

270 ContextHandle get_array_context "cuda_core::get_array_context" ( 

271 const OpaqueArrayHandle& h) noexcept nogil 

272 OpaqueArrayHandle create_array_level_handle "cuda_core::create_array_level_handle" ( 

273 const MipmappedArrayHandle& h_mip, unsigned int level) except+ nogil 

274 MipmappedArrayHandle create_mipmapped_array_handle "cuda_core::create_mipmapped_array_handle" ( 

275 const ContextHandle& h_context, const cydriver.CUDA_ARRAY3D_DESCRIPTOR& desc, 

276 unsigned int num_levels) except+ nogil 

277 ContextHandle get_mipmapped_array_context "cuda_core::get_mipmapped_array_context" ( 

278 const MipmappedArrayHandle& h) noexcept nogil 

279 TexObjectHandle create_tex_object_handle_array "cuda_core::create_tex_object_handle_array" ( 

280 const ContextHandle& h_context, const cydriver.CUDA_RESOURCE_DESC& res, 

281 const cydriver.CUDA_TEXTURE_DESC& tex, const OpaqueArrayHandle& h_backing) except+ nogil 

282 TexObjectHandle create_tex_object_handle_mipmap "cuda_core::create_tex_object_handle_mipmap" ( 

283 const ContextHandle& h_context, const cydriver.CUDA_RESOURCE_DESC& res, 

284 const cydriver.CUDA_TEXTURE_DESC& tex, const MipmappedArrayHandle& h_backing) except+ nogil 

285 TexObjectHandle create_tex_object_handle_linear "cuda_core::create_tex_object_handle_linear" ( 

286 const ContextHandle& h_context, const cydriver.CUDA_RESOURCE_DESC& res, 

287 const cydriver.CUDA_TEXTURE_DESC& tex, const DevicePtrHandle& h_backing) except+ nogil 

288 SurfObjectHandle create_surf_object_handle "cuda_core::create_surf_object_handle" ( 

289 const ContextHandle& h_context, const cydriver.CUDA_RESOURCE_DESC& res, 

290 const OpaqueArrayHandle& h_backing) except+ nogil 

291  

292  

293# ============================================================================= 

294# CUDA Driver API capsule 

295# 

296# This provides resolved CUDA driver function pointers to the C++ code. 

297# ============================================================================= 

298  

299cdef const char* _CUDA_DRIVER_API_V1_NAME = b"cuda.core._resource_handles._CUDA_DRIVER_API_V1" (empty)

300  

301  

302# ============================================================================= 

303# CUDA driver function pointer initialization 

304# 

305# The C++ code declares extern function pointers (p_cuXxx) that need to be 

306# populated before any handle creation functions are called. We extract these 

307# from cuda.bindings.cydriver.__pyx_capi__ at module import time. 

308# 

309# The Cython string substitution (e.g., "reinterpret_cast<void*&>(...)") 

310# allows us to assign void* values to typed function pointer variables. 

311# ============================================================================= 

312  

313# Declare extern variables with reinterpret_cast to allow void* assignment 

314cdef extern from "_cpp/resource_handles.hpp" namespace "cuda_core": 

315 # Error formatting 

316 void* p_cuGetErrorName "reinterpret_cast<void*&>(cuda_core::p_cuGetErrorName)" 

317 void* p_cuGetErrorString "reinterpret_cast<void*&>(cuda_core::p_cuGetErrorString)" 

318  

319 # Context 

320 void* p_cuDevicePrimaryCtxRetain "reinterpret_cast<void*&>(cuda_core::p_cuDevicePrimaryCtxRetain)" 

321 void* p_cuDevicePrimaryCtxRelease "reinterpret_cast<void*&>(cuda_core::p_cuDevicePrimaryCtxRelease)" 

322 void* p_cuCtxGetCurrent "reinterpret_cast<void*&>(cuda_core::p_cuCtxGetCurrent)" 

323 void* p_cuCtxSetCurrent "reinterpret_cast<void*&>(cuda_core::p_cuCtxSetCurrent)" 

324 void* p_cuCtxSynchronize "reinterpret_cast<void*&>(cuda_core::p_cuCtxSynchronize)" 

325 void* p_cuCtxGetStreamPriorityRange "reinterpret_cast<void*&>(cuda_core::p_cuCtxGetStreamPriorityRange)" 

326 void* p_cuGreenCtxCreate "reinterpret_cast<void*&>(cuda_core::p_cuGreenCtxCreate)" 

327 void* p_cuGreenCtxDestroy "reinterpret_cast<void*&>(cuda_core::p_cuGreenCtxDestroy)" 

328 void* p_cuCtxFromGreenCtx "reinterpret_cast<void*&>(cuda_core::p_cuCtxFromGreenCtx)" 

329 void* p_cuDevResourceGenerateDesc "reinterpret_cast<void*&>(cuda_core::p_cuDevResourceGenerateDesc)" 

330 void* p_cuGreenCtxStreamCreate "reinterpret_cast<void*&>(cuda_core::p_cuGreenCtxStreamCreate)" 

331  

332 # Stream 

333 void* p_cuStreamCreateWithPriority "reinterpret_cast<void*&>(cuda_core::p_cuStreamCreateWithPriority)" 

334 void* p_cuStreamDestroy "reinterpret_cast<void*&>(cuda_core::p_cuStreamDestroy)" 

335 void* p_cuStreamGetCtx "reinterpret_cast<void*&>(cuda_core::p_cuStreamGetCtx)" 

336  

337 # Event 

338 void* p_cuEventCreate "reinterpret_cast<void*&>(cuda_core::p_cuEventCreate)" 

339 void* p_cuEventDestroy "reinterpret_cast<void*&>(cuda_core::p_cuEventDestroy)" 

340 void* p_cuIpcOpenEventHandle "reinterpret_cast<void*&>(cuda_core::p_cuIpcOpenEventHandle)" 

341  

342 # Device 

343 void* p_cuDeviceGetCount "reinterpret_cast<void*&>(cuda_core::p_cuDeviceGetCount)" 

344  

345 # Memory pool 

346 void* p_cuMemPoolSetAccess "reinterpret_cast<void*&>(cuda_core::p_cuMemPoolSetAccess)" 

347 void* p_cuMemPoolDestroy "reinterpret_cast<void*&>(cuda_core::p_cuMemPoolDestroy)" 

348 void* p_cuMemPoolCreate "reinterpret_cast<void*&>(cuda_core::p_cuMemPoolCreate)" 

349 void* p_cuDeviceGetMemPool "reinterpret_cast<void*&>(cuda_core::p_cuDeviceGetMemPool)" 

350 void* p_cuMemPoolImportFromShareableHandle "reinterpret_cast<void*&>(cuda_core::p_cuMemPoolImportFromShareableHandle)" 

351  

352 # Memory allocation 

353 void* p_cuMemAllocFromPoolAsync "reinterpret_cast<void*&>(cuda_core::p_cuMemAllocFromPoolAsync)" 

354 void* p_cuMemAllocAsync "reinterpret_cast<void*&>(cuda_core::p_cuMemAllocAsync)" 

355 void* p_cuMemAlloc "reinterpret_cast<void*&>(cuda_core::p_cuMemAlloc)" 

356 void* p_cuMemAllocHost "reinterpret_cast<void*&>(cuda_core::p_cuMemAllocHost)" 

357  

358 # Memory deallocation 

359 void* p_cuMemFreeAsync "reinterpret_cast<void*&>(cuda_core::p_cuMemFreeAsync)" 

360 void* p_cuMemFree "reinterpret_cast<void*&>(cuda_core::p_cuMemFree)" 

361 void* p_cuMemFreeHost "reinterpret_cast<void*&>(cuda_core::p_cuMemFreeHost)" 

362  

363 # IPC 

364 void* p_cuMemPoolImportPointer "reinterpret_cast<void*&>(cuda_core::p_cuMemPoolImportPointer)" 

365  

366 # Library 

367 void* p_cuLibraryLoadFromFile "reinterpret_cast<void*&>(cuda_core::p_cuLibraryLoadFromFile)" 

368 void* p_cuLibraryLoadData "reinterpret_cast<void*&>(cuda_core::p_cuLibraryLoadData)" 

369 void* p_cuLibraryUnload "reinterpret_cast<void*&>(cuda_core::p_cuLibraryUnload)" 

370 void* p_cuLibraryGetKernel "reinterpret_cast<void*&>(cuda_core::p_cuLibraryGetKernel)" 

371  

372 # Graph 

373 void* p_cuGraphDestroy "reinterpret_cast<void*&>(cuda_core::p_cuGraphDestroy)" 

374 void* p_cuGraphInstantiateWithParams "reinterpret_cast<void*&>(cuda_core::p_cuGraphInstantiateWithParams)" 

375 void* p_cuGraphExecUpdate "reinterpret_cast<void*&>(cuda_core::p_cuGraphExecUpdate)" 

376 void* p_cuGraphExecDestroy "reinterpret_cast<void*&>(cuda_core::p_cuGraphExecDestroy)" 

377 void* p_cuUserObjectCreate "reinterpret_cast<void*&>(cuda_core::p_cuUserObjectCreate)" 

378 void* p_cuUserObjectRelease "reinterpret_cast<void*&>(cuda_core::p_cuUserObjectRelease)" 

379 void* p_cuGraphRetainUserObject "reinterpret_cast<void*&>(cuda_core::p_cuGraphRetainUserObject)" 

380 void* p_cuGraphReleaseUserObject "reinterpret_cast<void*&>(cuda_core::p_cuGraphReleaseUserObject)" 

381 void* p_cuGraphNodeFindInClone "reinterpret_cast<void*&>(cuda_core::p_cuGraphNodeFindInClone)" 

382 void* p_cuGraphChildGraphNodeGetGraph "reinterpret_cast<void*&>(cuda_core::p_cuGraphChildGraphNodeGetGraph)" 

383  

384 # Linker 

385 void* p_cuLinkDestroy "reinterpret_cast<void*&>(cuda_core::p_cuLinkDestroy)" 

386  

387 # Graphics interop 

388 void* p_cuGraphicsUnmapResources "reinterpret_cast<void*&>(cuda_core::p_cuGraphicsUnmapResources)" 

389 void* p_cuGraphicsUnregisterResource "reinterpret_cast<void*&>(cuda_core::p_cuGraphicsUnregisterResource)" 

390  

391 # Texture / surface / array (PR #467) 

392 void* p_cuArray3DCreate "reinterpret_cast<void*&>(cuda_core::p_cuArray3DCreate)" 

393 void* p_cuArrayDestroy "reinterpret_cast<void*&>(cuda_core::p_cuArrayDestroy)" 

394 void* p_cuMipmappedArrayCreate "reinterpret_cast<void*&>(cuda_core::p_cuMipmappedArrayCreate)" 

395 void* p_cuMipmappedArrayDestroy "reinterpret_cast<void*&>(cuda_core::p_cuMipmappedArrayDestroy)" 

396 void* p_cuMipmappedArrayGetLevel "reinterpret_cast<void*&>(cuda_core::p_cuMipmappedArrayGetLevel)" 

397 void* p_cuTexObjectCreate "reinterpret_cast<void*&>(cuda_core::p_cuTexObjectCreate)" 

398 void* p_cuTexObjectDestroy "reinterpret_cast<void*&>(cuda_core::p_cuTexObjectDestroy)" 

399 void* p_cuSurfObjectCreate "reinterpret_cast<void*&>(cuda_core::p_cuSurfObjectCreate)" 

400 void* p_cuSurfObjectDestroy "reinterpret_cast<void*&>(cuda_core::p_cuSurfObjectDestroy)" 

401  

402 # SM resource split (13.1+) 

403 void* p_cuDevSmResourceSplit "reinterpret_cast<void*&>(cuda_core::p_cuDevSmResourceSplit)" 

404  

405 # cuMemcpyWithAttributesAsync (13.2+) 

406 void* p_cuMemcpyWithAttributesAsync "reinterpret_cast<void*&>(cuda_core::p_cuMemcpyWithAttributesAsync)" 

407  

408 # NVRTC 

409 void* p_nvrtcDestroyProgram "reinterpret_cast<void*&>(cuda_core::p_nvrtcDestroyProgram)" 

410  

411 # NVVM 

412 void* p_nvvmDestroyProgram "reinterpret_cast<void*&>(cuda_core::p_nvvmDestroyProgram)" 

413  

414 # nvJitLink 

415 void* p_nvJitLinkDestroy "reinterpret_cast<void*&>(cuda_core::p_nvJitLinkDestroy)" 

416  

417  

418# Initialize driver function pointers from cydriver.__pyx_capi__ at module load 

419cdef void* _get_driver_fn(str name): (empty)

420 capsule = cydriver.__pyx_capi__[name] (empty)

421 return PyCapsule_GetPointer(capsule, PyCapsule_GetName(capsule)) (empty)

422  

423  

424cdef void* _get_optional_driver_fn(str name): (empty)

425 try: (empty)

426 capsule = cydriver.__pyx_capi__[name] (empty)

427 except KeyError: 

428 return NULL 

429 return PyCapsule_GetPointer(capsule, PyCapsule_GetName(capsule)) (empty)

430  

431  

432cdef void _init_driver_fn_pointers() noexcept: (empty)

433 global p_cuGetErrorName, p_cuGetErrorString 

434 global p_cuDevicePrimaryCtxRetain, p_cuDevicePrimaryCtxRelease, p_cuCtxGetCurrent 

435 global p_cuCtxSetCurrent, p_cuCtxSynchronize, p_cuCtxGetStreamPriorityRange 

436 global p_cuGreenCtxCreate, p_cuGreenCtxDestroy, p_cuCtxFromGreenCtx 

437 global p_cuDevResourceGenerateDesc, p_cuGreenCtxStreamCreate 

438 global p_cuStreamCreateWithPriority, p_cuStreamDestroy, p_cuStreamGetCtx 

439 global p_cuEventCreate, p_cuEventDestroy, p_cuIpcOpenEventHandle 

440 global p_cuDeviceGetCount 

441 global p_cuMemPoolSetAccess, p_cuMemPoolDestroy, p_cuMemPoolCreate 

442 global p_cuDeviceGetMemPool, p_cuMemPoolImportFromShareableHandle 

443 global p_cuMemAllocFromPoolAsync, p_cuMemAllocAsync, p_cuMemAlloc, p_cuMemAllocHost 

444 global p_cuMemFreeAsync, p_cuMemFree, p_cuMemFreeHost 

445 global p_cuMemPoolImportPointer 

446 global p_cuLibraryLoadFromFile, p_cuLibraryLoadData, p_cuLibraryUnload, p_cuLibraryGetKernel 

447 global p_cuGraphDestroy, p_cuGraphInstantiateWithParams 

448 global p_cuGraphExecUpdate, p_cuGraphExecDestroy 

449 global p_cuUserObjectCreate, p_cuUserObjectRelease 

450 global p_cuGraphRetainUserObject, p_cuGraphReleaseUserObject 

451 global p_cuGraphNodeFindInClone, p_cuGraphChildGraphNodeGetGraph 

452 global p_cuLinkDestroy 

453 global p_cuGraphicsUnmapResources, p_cuGraphicsUnregisterResource 

454 global p_cuDevSmResourceSplit 

455 global p_cuMemcpyWithAttributesAsync 

456 global p_cuArray3DCreate, p_cuArrayDestroy 

457 global p_cuMipmappedArrayCreate, p_cuMipmappedArrayDestroy, p_cuMipmappedArrayGetLevel 

458 global p_cuTexObjectCreate, p_cuTexObjectDestroy 

459 global p_cuSurfObjectCreate, p_cuSurfObjectDestroy 

460  

461 # Error formatting 

462 p_cuGetErrorName = _get_driver_fn("cuGetErrorName") (empty)

463 p_cuGetErrorString = _get_driver_fn("cuGetErrorString") (empty)

464  

465 # Context 

466 p_cuDevicePrimaryCtxRetain = _get_driver_fn("cuDevicePrimaryCtxRetain") (empty)

467 p_cuDevicePrimaryCtxRelease = _get_driver_fn("cuDevicePrimaryCtxRelease") (empty)

468 p_cuCtxGetCurrent = _get_driver_fn("cuCtxGetCurrent") (empty)

469 p_cuCtxSetCurrent = _get_driver_fn("cuCtxSetCurrent") (empty)

470 p_cuCtxSynchronize = _get_driver_fn("cuCtxSynchronize") (empty)

471 p_cuCtxGetStreamPriorityRange = _get_driver_fn("cuCtxGetStreamPriorityRange") (empty)

472 p_cuGreenCtxCreate = _get_optional_driver_fn("cuGreenCtxCreate") (empty)

473 p_cuGreenCtxDestroy = _get_optional_driver_fn("cuGreenCtxDestroy") (empty)

474 p_cuCtxFromGreenCtx = _get_optional_driver_fn("cuCtxFromGreenCtx") (empty)

475 p_cuDevResourceGenerateDesc = _get_optional_driver_fn("cuDevResourceGenerateDesc") (empty)

476 p_cuGreenCtxStreamCreate = _get_optional_driver_fn("cuGreenCtxStreamCreate") (empty)

477  

478 # Stream 

479 p_cuStreamCreateWithPriority = _get_driver_fn("cuStreamCreateWithPriority") (empty)

480 p_cuStreamDestroy = _get_driver_fn("cuStreamDestroy") (empty)

481 p_cuStreamGetCtx = _get_driver_fn("cuStreamGetCtx") (empty)

482  

483 # Event 

484 p_cuEventCreate = _get_driver_fn("cuEventCreate") (empty)

485 p_cuEventDestroy = _get_driver_fn("cuEventDestroy") (empty)

486 p_cuIpcOpenEventHandle = _get_driver_fn("cuIpcOpenEventHandle") (empty)

487  

488 # Device 

489 p_cuDeviceGetCount = _get_driver_fn("cuDeviceGetCount") (empty)

490  

491 # Memory pool 

492 p_cuMemPoolSetAccess = _get_driver_fn("cuMemPoolSetAccess") (empty)

493 p_cuMemPoolDestroy = _get_driver_fn("cuMemPoolDestroy") (empty)

494 p_cuMemPoolCreate = _get_driver_fn("cuMemPoolCreate") (empty)

495 p_cuDeviceGetMemPool = _get_driver_fn("cuDeviceGetMemPool") (empty)

496 p_cuMemPoolImportFromShareableHandle = _get_driver_fn("cuMemPoolImportFromShareableHandle") (empty)

497  

498 # Memory allocation 

499 p_cuMemAllocFromPoolAsync = _get_driver_fn("cuMemAllocFromPoolAsync") (empty)

500 p_cuMemAllocAsync = _get_driver_fn("cuMemAllocAsync") (empty)

501 p_cuMemAlloc = _get_driver_fn("cuMemAlloc") (empty)

502 p_cuMemAllocHost = _get_driver_fn("cuMemAllocHost") (empty)

503  

504 # Memory deallocation 

505 p_cuMemFreeAsync = _get_driver_fn("cuMemFreeAsync") (empty)

506 p_cuMemFree = _get_driver_fn("cuMemFree") (empty)

507 p_cuMemFreeHost = _get_driver_fn("cuMemFreeHost") (empty)

508  

509 # IPC 

510 p_cuMemPoolImportPointer = _get_driver_fn("cuMemPoolImportPointer") (empty)

511  

512 # Library 

513 p_cuLibraryLoadFromFile = _get_driver_fn("cuLibraryLoadFromFile") (empty)

514 p_cuLibraryLoadData = _get_driver_fn("cuLibraryLoadData") (empty)

515 p_cuLibraryUnload = _get_driver_fn("cuLibraryUnload") (empty)

516 p_cuLibraryGetKernel = _get_driver_fn("cuLibraryGetKernel") (empty)

517  

518 # Graph 

519 p_cuGraphDestroy = _get_driver_fn("cuGraphDestroy") (empty)

520 p_cuGraphInstantiateWithParams = _get_driver_fn("cuGraphInstantiateWithParams") (empty)

521 p_cuGraphExecUpdate = _get_driver_fn("cuGraphExecUpdate") (empty)

522 p_cuGraphExecDestroy = _get_driver_fn("cuGraphExecDestroy") (empty)

523 p_cuUserObjectCreate = _get_driver_fn("cuUserObjectCreate") (empty)

524 p_cuUserObjectRelease = _get_driver_fn("cuUserObjectRelease") (empty)

525 p_cuGraphRetainUserObject = _get_driver_fn("cuGraphRetainUserObject") (empty)

526 p_cuGraphReleaseUserObject = _get_driver_fn("cuGraphReleaseUserObject") (empty)

527 p_cuGraphNodeFindInClone = _get_driver_fn("cuGraphNodeFindInClone") (empty)

528 p_cuGraphChildGraphNodeGetGraph = _get_driver_fn("cuGraphChildGraphNodeGetGraph") (empty)

529  

530 # Linker 

531 p_cuLinkDestroy = _get_driver_fn("cuLinkDestroy") (empty)

532  

533 # Graphics interop 

534 p_cuGraphicsUnmapResources = _get_driver_fn("cuGraphicsUnmapResources") (empty)

535 p_cuGraphicsUnregisterResource = _get_driver_fn("cuGraphicsUnregisterResource") (empty)

536  

537 # Texture / surface / array (PR #467) 

538 p_cuArray3DCreate = _get_driver_fn("cuArray3DCreate") (empty)

539 p_cuArrayDestroy = _get_driver_fn("cuArrayDestroy") (empty)

540 p_cuMipmappedArrayCreate = _get_driver_fn("cuMipmappedArrayCreate") (empty)

541 p_cuMipmappedArrayDestroy = _get_driver_fn("cuMipmappedArrayDestroy") (empty)

542 p_cuMipmappedArrayGetLevel = _get_driver_fn("cuMipmappedArrayGetLevel") (empty)

543 p_cuTexObjectCreate = _get_driver_fn("cuTexObjectCreate") (empty)

544 p_cuTexObjectDestroy = _get_driver_fn("cuTexObjectDestroy") (empty)

545 p_cuSurfObjectCreate = _get_driver_fn("cuSurfObjectCreate") (empty)

546 p_cuSurfObjectDestroy = _get_driver_fn("cuSurfObjectDestroy") (empty)

547  

548 # SM resource split (13.1+ — may not exist in older cuda-bindings) 

549 p_cuDevSmResourceSplit = _get_optional_driver_fn("cuDevSmResourceSplit") (empty)

550  

551 # cuMemcpyWithAttributesAsync (13.2+ — may not exist in older cuda-bindings) 

552 p_cuMemcpyWithAttributesAsync = _get_optional_driver_fn("cuMemcpyWithAttributesAsync") (empty)

553  

554_init_driver_fn_pointers() (empty)

555initialize_deferred_cleanup() (empty)

556  

557# ============================================================================= 

558# NVRTC function pointer initialization 

559# ============================================================================= 

560  

561cdef void* _get_nvrtc_fn(str name): (empty)

562 capsule = cynvrtc.__pyx_capi__[name] (empty)

563 return PyCapsule_GetPointer(capsule, PyCapsule_GetName(capsule)) (empty)

564  

565cdef void _init_nvrtc_fn_pointers() noexcept: (empty)

566 global p_nvrtcDestroyProgram 

567 p_nvrtcDestroyProgram = _get_nvrtc_fn("nvrtcDestroyProgram") (empty)

568  

569_init_nvrtc_fn_pointers() (empty)

570  

571# ============================================================================= 

572# NVVM function pointer initialization 

573# 

574# NVVM may not be available at runtime, so we handle missing function pointers 

575# gracefully. The C++ deleter checks for null before calling. 

576# ============================================================================= 

577  

578cdef void* _get_nvvm_fn(str name): (empty)

579 capsule = cynvvm.__pyx_capi__[name] (empty)

580 return PyCapsule_GetPointer(capsule, PyCapsule_GetName(capsule)) (empty)

581  

582cdef void _init_nvvm_fn_pointers() noexcept: (empty)

583 global p_nvvmDestroyProgram 

584 p_nvvmDestroyProgram = _get_nvvm_fn("nvvmDestroyProgram") (empty)

585  

586_init_nvvm_fn_pointers() (empty)

587  

588# ============================================================================= 

589# nvJitLink function pointer initialization 

590# 

591# nvJitLink may not be available at runtime, so we handle missing function 

592# pointers gracefully. The C++ deleter checks for null before calling. 

593# ============================================================================= 

594  

595cdef void* _get_nvjitlink_fn(str name): (empty)

596 capsule = cynvjitlink.__pyx_capi__[name] (empty)

597 return PyCapsule_GetPointer(capsule, PyCapsule_GetName(capsule)) (empty)

598  

599cdef void _init_nvjitlink_fn_pointers() noexcept: (empty)

600 global p_nvJitLinkDestroy 

601 p_nvJitLinkDestroy = _get_nvjitlink_fn("nvJitLinkDestroy") (empty)

602  

603_init_nvjitlink_fn_pointers() (empty)