Coverage for cuda/core/_resource_handles.pyx: 97.73%

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  

55 # Stream handles 

56 StreamHandle create_stream_handle "cuda_core::create_stream_handle" ( 

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

58 StreamHandle create_stream_handle_ref "cuda_core::create_stream_handle_ref" ( 

59 cydriver.CUstream stream) except+ nogil 

60 StreamHandle create_stream_handle_with_owner "cuda_core::create_stream_handle_with_owner" ( 

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

62 void py_object_user_object_destroy "cuda_core::py_object_user_object_destroy" ( 

63 void* py_object) noexcept nogil 

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

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

66 ContextHandle get_stream_context "cuda_core::get_stream_context" ( 

67 const StreamHandle& h) noexcept nogil 

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

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

70  

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

72 EventHandle create_event_handle "cuda_core::create_event_handle" ( 

73 const ContextHandle& h_ctx, unsigned int flags, 

74 bint timing_enabled, bint is_blocking_sync, 

75 bint ipc_enabled, int device_id) except+ nogil 

76 EventHandle create_event_handle_noctx "cuda_core::create_event_handle_noctx" ( 

77 unsigned int flags) except+ nogil 

78 EventHandle create_event_handle_ref "cuda_core::create_event_handle_ref" ( 

79 cydriver.CUevent event) except+ nogil 

80 EventHandle create_event_handle_ipc "cuda_core::create_event_handle_ipc" ( 

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

82  

83 # Event metadata getters 

84 bint get_event_timing_enabled "cuda_core::get_event_timing_enabled" ( 

85 const EventHandle& h) noexcept nogil 

86 bint get_event_is_blocking_sync "cuda_core::get_event_is_blocking_sync" ( 

87 const EventHandle& h) noexcept nogil 

88 bint get_event_ipc_enabled "cuda_core::get_event_ipc_enabled" ( 

89 const EventHandle& h) noexcept nogil 

90 int get_event_device_id "cuda_core::get_event_device_id" ( 

91 const EventHandle& h) noexcept nogil 

92 ContextHandle get_event_context "cuda_core::get_event_context" ( 

93 const EventHandle& h) noexcept nogil 

94  

95 # Memory pool handles 

96 MemoryPoolHandle create_mempool_handle "cuda_core::create_mempool_handle" ( 

97 const cydriver.CUmemPoolProps& props) except+ nogil 

98 MemoryPoolHandle create_mempool_handle_ref "cuda_core::create_mempool_handle_ref" ( 

99 cydriver.CUmemoryPool pool) except+ nogil 

100 MemoryPoolHandle get_device_mempool "cuda_core::get_device_mempool" ( 

101 int device_id) except+ nogil 

102 MemoryPoolHandle create_mempool_handle_ipc "cuda_core::create_mempool_handle_ipc" ( 

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

104  

105 # Device pointer handles 

106 DevicePtrHandle deviceptr_alloc_from_pool "cuda_core::deviceptr_alloc_from_pool" ( 

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

108 DevicePtrHandle deviceptr_alloc_async "cuda_core::deviceptr_alloc_async" ( 

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

110 DevicePtrHandle deviceptr_alloc "cuda_core::deviceptr_alloc" (size_t size) except+ nogil 

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

112 DevicePtrHandle deviceptr_create_ref "cuda_core::deviceptr_create_ref" ( 

113 cydriver.CUdeviceptr ptr) except+ nogil 

114 DevicePtrHandle deviceptr_create_with_owner "cuda_core::deviceptr_create_with_owner" ( 

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

116 DevicePtrHandle deviceptr_create_mapped_graphics "cuda_core::deviceptr_create_mapped_graphics" ( 

117 cydriver.CUdeviceptr ptr, 

118 const GraphicsResourceHandle& h_resource, 

119 const StreamHandle& h_stream) except+ nogil 

120  

121 # MR deallocation callback 

122 void register_mr_dealloc_callback "cuda_core::register_mr_dealloc_callback" ( 

123 MRDeallocCallback cb) noexcept 

124 DevicePtrHandle deviceptr_create_with_mr "cuda_core::deviceptr_create_with_mr" ( 

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

126  

127 DevicePtrHandle deviceptr_import_ipc "cuda_core::deviceptr_import_ipc" ( 

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

129 StreamHandle deallocation_stream "cuda_core::deallocation_stream" ( 

130 const DevicePtrHandle& h) noexcept nogil 

131 void set_deallocation_stream "cuda_core::set_deallocation_stream" ( 

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

133  

134 # Library handles 

135 LibraryHandle create_library_handle_from_file "cuda_core::create_library_handle_from_file" ( 

136 const char* path) except+ nogil 

137 LibraryHandle create_library_handle_from_data "cuda_core::create_library_handle_from_data" ( 

138 const void* data) except+ nogil 

139 LibraryHandle create_library_handle_ref "cuda_core::create_library_handle_ref" ( 

140 cydriver.CUlibrary library) except+ nogil 

141  

142 # Kernel handles 

143 KernelHandle create_kernel_handle "cuda_core::create_kernel_handle" ( 

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

145 KernelHandle create_kernel_handle_ref "cuda_core::create_kernel_handle_ref" ( 

146 cydriver.CUkernel kernel) except+ nogil 

147 LibraryHandle get_kernel_library "cuda_core::get_kernel_library" ( 

148 const KernelHandle& h) noexcept nogil 

149  

150 # Graph handles 

151 GraphHandle create_graph_handle "cuda_core::create_graph_handle" ( 

152 cydriver.CUgraph graph) except+ nogil 

153 GraphHandle create_child_graph_handle "cuda_core::create_child_graph_handle" ( 

154 cydriver.CUgraph child_graph, const GraphHandle& h_parent, 

155 cydriver.CUgraphNode owner_node) except+ nogil 

156  

157 # Graph node attachments 

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

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

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

161 const GraphHandle& h_graph, cydriver.CUgraphNode node, 

162 OpaqueHandle* owner0, OpaqueHandle* owner1) except+ 

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

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

165 PreparedAttachment* out_prepared) except+ 

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

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

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

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

170 void invalidate_child_graph_state "cuda_core::invalidate_child_graph_state" ( 

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

172  

173 # Graph exec handles 

174 GraphExecHandle create_graph_exec_handle "cuda_core::create_graph_exec_handle" ( 

175 cydriver.CUgraphExec graph_exec) except+ nogil 

176  

177 # Graph node handles 

178 GraphNodeHandle create_graph_node_handle "cuda_core::create_graph_node_handle" ( 

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

180 GraphHandle graph_node_get_graph "cuda_core::graph_node_get_graph" ( 

181 const GraphNodeHandle& h) noexcept nogil 

182 void invalidate_graph_node "cuda_core::invalidate_graph_node" ( 

183 const GraphNodeHandle& h) noexcept nogil 

184  

185 # Graphics resource handles 

186 GraphicsResourceHandle create_graphics_resource_handle "cuda_core::create_graphics_resource_handle" ( 

187 cydriver.CUgraphicsResource resource) except+ nogil 

188  

189 # NVRTC Program handles 

190 NvrtcProgramHandle create_nvrtc_program_handle "cuda_core::create_nvrtc_program_handle" ( 

191 cynvrtc.nvrtcProgram prog) except+ nogil 

192 NvrtcProgramHandle create_nvrtc_program_handle_ref "cuda_core::create_nvrtc_program_handle_ref" ( 

193 cynvrtc.nvrtcProgram prog) except+ nogil 

194  

195 # NVVM Program handles 

196 NvvmProgramHandle create_nvvm_program_handle "cuda_core::create_nvvm_program_handle" ( 

197 cynvvm.nvvmProgram prog) except+ nogil 

198 NvvmProgramHandle create_nvvm_program_handle_ref "cuda_core::create_nvvm_program_handle_ref" ( 

199 cynvvm.nvvmProgram prog) except+ nogil 

200  

201 # nvJitLink handles 

202 NvJitLinkHandle create_nvjitlink_handle "cuda_core::create_nvjitlink_handle" ( 

203 cynvjitlink.nvJitLinkHandle handle) except+ nogil 

204 NvJitLinkHandle create_nvjitlink_handle_ref "cuda_core::create_nvjitlink_handle_ref" ( 

205 cynvjitlink.nvJitLinkHandle handle) except+ nogil 

206  

207 # cuLink handles 

208 CuLinkHandle create_culink_handle "cuda_core::create_culink_handle" ( 

209 cydriver.CUlinkState state) except+ nogil 

210 CuLinkHandle create_culink_handle_ref "cuda_core::create_culink_handle_ref" ( 

211 cydriver.CUlinkState state) except+ nogil 

212  

213 # File descriptor handles 

214 FileDescriptorHandle create_fd_handle "cuda_core::create_fd_handle" ( 

215 int fd) except+ nogil 

216 FileDescriptorHandle create_fd_handle_ref "cuda_core::create_fd_handle_ref" ( 

217 int fd) except+ nogil 

218  

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

220 # groupParams is void* to avoid referencing CU_DEV_SM_RESOURCE_GROUP_PARAMS 

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

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

223 cydriver.CUdevResource* result, unsigned int nbGroups, 

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

225 unsigned int flags, void* groupParams) nogil 

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

227  

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

229 OpaqueArrayHandle create_array_handle "cuda_core::create_array_handle" ( 

230 const cydriver.CUDA_ARRAY3D_DESCRIPTOR& desc) except+ nogil 

231 OpaqueArrayHandle create_array_handle_ref "cuda_core::create_array_handle_ref" ( 

232 cydriver.CUarray arr) except+ nogil 

233 OpaqueArrayHandle create_array_handle_owning "cuda_core::create_array_handle_owning" ( 

234 cydriver.CUarray arr) except+ nogil 

235 OpaqueArrayHandle create_array_level_handle "cuda_core::create_array_level_handle" ( 

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

237 MipmappedArrayHandle create_mipmapped_array_handle "cuda_core::create_mipmapped_array_handle" ( 

238 const cydriver.CUDA_ARRAY3D_DESCRIPTOR& desc, unsigned int num_levels) except+ nogil 

239 TexObjectHandle create_tex_object_handle_array "cuda_core::create_tex_object_handle_array" ( 

240 const cydriver.CUDA_RESOURCE_DESC& res, const cydriver.CUDA_TEXTURE_DESC& tex, 

241 const OpaqueArrayHandle& h_backing) except+ nogil 

242 TexObjectHandle create_tex_object_handle_mipmap "cuda_core::create_tex_object_handle_mipmap" ( 

243 const cydriver.CUDA_RESOURCE_DESC& res, const cydriver.CUDA_TEXTURE_DESC& tex, 

244 const MipmappedArrayHandle& h_backing) except+ nogil 

245 TexObjectHandle create_tex_object_handle_linear "cuda_core::create_tex_object_handle_linear" ( 

246 const cydriver.CUDA_RESOURCE_DESC& res, const cydriver.CUDA_TEXTURE_DESC& tex, 

247 const DevicePtrHandle& h_backing) except+ nogil 

248 SurfObjectHandle create_surf_object_handle "cuda_core::create_surf_object_handle" ( 

249 const cydriver.CUDA_RESOURCE_DESC& res, const OpaqueArrayHandle& h_backing) except+ nogil 

250  

251  

252# ============================================================================= 

253# CUDA Driver API capsule 

254# 

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

256# ============================================================================= 

257  

258cdef const char* _CUDA_DRIVER_API_V1_NAME = b"cuda.core._resource_handles._CUDA_DRIVER_API_V1" (empty)

259  

260  

261# ============================================================================= 

262# CUDA driver function pointer initialization 

263# 

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

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

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

267# 

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

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

270# ============================================================================= 

271  

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

273cdef extern from "_cpp/resource_handles.hpp" namespace "cuda_core": 

274 # Context 

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

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

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

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

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

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

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

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

283  

284 # Stream 

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

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

287  

288 # Event 

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

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

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

292  

293 # Device 

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

295  

296 # Memory pool 

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

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

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

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

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

302  

303 # Memory allocation 

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

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

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

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

308  

309 # Memory deallocation 

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

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

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

313  

314 # IPC 

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

316  

317 # Library 

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

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

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

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

322  

323 # Graph 

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

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

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

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

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

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

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

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

332  

333 # Linker 

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

335  

336 # Graphics interop 

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

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

339  

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

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

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

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

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

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

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

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

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

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

350  

351 # SM resource split (13.1+) 

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

353  

354 # NVRTC 

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

356  

357 # NVVM 

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

359  

360 # nvJitLink 

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

362  

363  

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

365cdef void* _get_driver_fn(str name): (empty)

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

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

368  

369  

370cdef void* _get_optional_driver_fn(str name): (empty)

371 try: (empty)

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

373 except KeyError: 

374 return NULL 

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

376  

377  

378cdef void _init_driver_fn_pointers() noexcept: (empty)

379 global p_cuDevicePrimaryCtxRetain, p_cuDevicePrimaryCtxRelease, p_cuCtxGetCurrent 

380 global p_cuGreenCtxCreate, p_cuGreenCtxDestroy, p_cuCtxFromGreenCtx 

381 global p_cuDevResourceGenerateDesc, p_cuGreenCtxStreamCreate 

382 global p_cuStreamCreateWithPriority, p_cuStreamDestroy 

383 global p_cuEventCreate, p_cuEventDestroy, p_cuIpcOpenEventHandle 

384 global p_cuDeviceGetCount 

385 global p_cuMemPoolSetAccess, p_cuMemPoolDestroy, p_cuMemPoolCreate 

386 global p_cuDeviceGetMemPool, p_cuMemPoolImportFromShareableHandle 

387 global p_cuMemAllocFromPoolAsync, p_cuMemAllocAsync, p_cuMemAlloc, p_cuMemAllocHost 

388 global p_cuMemFreeAsync, p_cuMemFree, p_cuMemFreeHost 

389 global p_cuMemPoolImportPointer 

390 global p_cuLibraryLoadFromFile, p_cuLibraryLoadData, p_cuLibraryUnload, p_cuLibraryGetKernel 

391 global p_cuGraphDestroy, p_cuGraphExecDestroy 

392 global p_cuUserObjectCreate, p_cuUserObjectRelease 

393 global p_cuGraphRetainUserObject, p_cuGraphReleaseUserObject 

394 global p_cuGraphNodeFindInClone, p_cuGraphChildGraphNodeGetGraph 

395 global p_cuLinkDestroy 

396 global p_cuGraphicsUnmapResources, p_cuGraphicsUnregisterResource 

397 global p_cuDevSmResourceSplit 

398 global p_cuArray3DCreate, p_cuArrayDestroy 

399 global p_cuMipmappedArrayCreate, p_cuMipmappedArrayDestroy, p_cuMipmappedArrayGetLevel 

400 global p_cuTexObjectCreate, p_cuTexObjectDestroy 

401 global p_cuSurfObjectCreate, p_cuSurfObjectDestroy 

402  

403 # Context 

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

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

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

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

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

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

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

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

412  

413 # Stream 

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

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

416  

417 # Event 

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

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

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

421  

422 # Device 

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

424  

425 # Memory pool 

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

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

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

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

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

431  

432 # Memory allocation 

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

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

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

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

437  

438 # Memory deallocation 

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

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

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

442  

443 # IPC 

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

445  

446 # Library 

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

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

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

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

451  

452 # Graph 

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

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

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

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

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

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

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

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

461  

462 # Linker 

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

464  

465 # Graphics interop 

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

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

468  

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

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

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

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

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

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

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

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

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

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

479  

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

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

482  

483_init_driver_fn_pointers() (empty)

484initialize_deferred_cleanup() (empty)

485  

486# ============================================================================= 

487# NVRTC function pointer initialization 

488# ============================================================================= 

489  

490cdef void* _get_nvrtc_fn(str name): (empty)

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

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

493  

494cdef void _init_nvrtc_fn_pointers() noexcept: (empty)

495 global p_nvrtcDestroyProgram 

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

497  

498_init_nvrtc_fn_pointers() (empty)

499  

500# ============================================================================= 

501# NVVM function pointer initialization 

502# 

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

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

505# ============================================================================= 

506  

507cdef void* _get_nvvm_fn(str name): (empty)

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

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

510  

511cdef void _init_nvvm_fn_pointers() noexcept: (empty)

512 global p_nvvmDestroyProgram 

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

514  

515_init_nvvm_fn_pointers() (empty)

516  

517# ============================================================================= 

518# nvJitLink function pointer initialization 

519# 

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

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

522# ============================================================================= 

523  

524cdef void* _get_nvjitlink_fn(str name): (empty)

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

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

527  

528cdef void _init_nvjitlink_fn_pointers() noexcept: (empty)

529 global p_nvJitLinkDestroy 

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

531  

532_init_nvjitlink_fn_pointers() (empty)