Coverage for cuda/core/_resource_handles.pyx: 97.83%

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

171 const GraphHandle& h_parent, const GraphHandle& h_old_child, 

172 cydriver.CUgraphNode owner_node, const GraphHandle& h_source, 

173 PreparedChildGraphUpdate* out_prepared) except+ 

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

175 PreparedChildGraphUpdate& prepared, GraphHandle* out_child) except+ 

176 void invalidate_child_graph_state "cuda_core::invalidate_child_graph_state" ( 

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

178  

179 # Graph exec handles 

180 GraphExecHandle create_graph_exec_handle "cuda_core::create_graph_exec_handle" ( 

181 const GraphHandle& h_source, 

182 cydriver.CUDA_GRAPH_INSTANTIATE_PARAMS* params) except+ 

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

184 const GraphExecHandle& h_exec, 

185 const GraphHandle& h_source, 

186 cydriver.CUgraphExecUpdateResultInfo* result_info) except+ 

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

188 const GraphExecHandle& h_exec, 

189 OpaqueHandle owner0, 

190 OpaqueHandle owner1, 

191 PreparedExecAttachment* out_prepared) except+ 

192 void graph_commit_exec_attachment "cuda_core::graph_commit_exec_attachment" ( 

193 PreparedExecAttachment& prepared) noexcept 

194  

195 # Graph node handles 

196 GraphNodeHandle create_graph_node_handle "cuda_core::create_graph_node_handle" ( 

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

198 GraphHandle graph_node_get_graph "cuda_core::graph_node_get_graph" ( 

199 const GraphNodeHandle& h) noexcept nogil 

200 void invalidate_graph_node "cuda_core::invalidate_graph_node" ( 

201 const GraphNodeHandle& h) noexcept nogil 

202  

203 # Graphics resource handles 

204 GraphicsResourceHandle create_graphics_resource_handle "cuda_core::create_graphics_resource_handle" ( 

205 cydriver.CUgraphicsResource resource) except+ nogil 

206  

207 # NVRTC Program handles 

208 NvrtcProgramHandle create_nvrtc_program_handle "cuda_core::create_nvrtc_program_handle" ( 

209 cynvrtc.nvrtcProgram prog) except+ nogil 

210 NvrtcProgramHandle create_nvrtc_program_handle_ref "cuda_core::create_nvrtc_program_handle_ref" ( 

211 cynvrtc.nvrtcProgram prog) except+ nogil 

212  

213 # NVVM Program handles 

214 NvvmProgramHandle create_nvvm_program_handle "cuda_core::create_nvvm_program_handle" ( 

215 cynvvm.nvvmProgram prog) except+ nogil 

216 NvvmProgramHandle create_nvvm_program_handle_ref "cuda_core::create_nvvm_program_handle_ref" ( 

217 cynvvm.nvvmProgram prog) except+ nogil 

218  

219 # nvJitLink handles 

220 NvJitLinkHandle create_nvjitlink_handle "cuda_core::create_nvjitlink_handle" ( 

221 cynvjitlink.nvJitLinkHandle handle) except+ nogil 

222 NvJitLinkHandle create_nvjitlink_handle_ref "cuda_core::create_nvjitlink_handle_ref" ( 

223 cynvjitlink.nvJitLinkHandle handle) except+ nogil 

224  

225 # cuLink handles 

226 CuLinkHandle create_culink_handle "cuda_core::create_culink_handle" ( 

227 cydriver.CUlinkState state) except+ nogil 

228 CuLinkHandle create_culink_handle_ref "cuda_core::create_culink_handle_ref" ( 

229 cydriver.CUlinkState state) except+ nogil 

230  

231 # File descriptor handles 

232 FileDescriptorHandle create_fd_handle "cuda_core::create_fd_handle" ( 

233 int fd) except+ nogil 

234 FileDescriptorHandle create_fd_handle_ref "cuda_core::create_fd_handle_ref" ( 

235 int fd) except+ nogil 

236  

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

238 # groupParams is void* to avoid referencing CU_DEV_SM_RESOURCE_GROUP_PARAMS 

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

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

241 cydriver.CUdevResource* result, unsigned int nbGroups, 

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

243 unsigned int flags, void* groupParams) nogil 

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

245  

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

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

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

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

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

251 void* attr, cydriver.CUstream hStream) nogil 

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

253  

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

255 OpaqueArrayHandle create_array_handle "cuda_core::create_array_handle" ( 

256 const cydriver.CUDA_ARRAY3D_DESCRIPTOR& desc) except+ nogil 

257 OpaqueArrayHandle create_array_handle_ref "cuda_core::create_array_handle_ref" ( 

258 cydriver.CUarray arr) except+ nogil 

259 OpaqueArrayHandle create_array_handle_owning "cuda_core::create_array_handle_owning" ( 

260 cydriver.CUarray arr) except+ nogil 

261 OpaqueArrayHandle create_array_level_handle "cuda_core::create_array_level_handle" ( 

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

263 MipmappedArrayHandle create_mipmapped_array_handle "cuda_core::create_mipmapped_array_handle" ( 

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

265 TexObjectHandle create_tex_object_handle_array "cuda_core::create_tex_object_handle_array" ( 

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

267 const OpaqueArrayHandle& h_backing) except+ nogil 

268 TexObjectHandle create_tex_object_handle_mipmap "cuda_core::create_tex_object_handle_mipmap" ( 

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

270 const MipmappedArrayHandle& h_backing) except+ nogil 

271 TexObjectHandle create_tex_object_handle_linear "cuda_core::create_tex_object_handle_linear" ( 

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

273 const DevicePtrHandle& h_backing) except+ nogil 

274 SurfObjectHandle create_surf_object_handle "cuda_core::create_surf_object_handle" ( 

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

276  

277  

278# ============================================================================= 

279# CUDA Driver API capsule 

280# 

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

282# ============================================================================= 

283  

284cdef const char* _CUDA_DRIVER_API_V1_NAME = b"cuda.core._resource_handles._CUDA_DRIVER_API_V1" (empty)

285  

286  

287# ============================================================================= 

288# CUDA driver function pointer initialization 

289# 

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

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

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

293# 

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

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

296# ============================================================================= 

297  

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

299cdef extern from "_cpp/resource_handles.hpp" namespace "cuda_core": 

300 # Context 

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

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

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

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

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

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

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

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

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

310  

311 # Stream 

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

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

314  

315 # Event 

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

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

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

319  

320 # Device 

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

322  

323 # Memory pool 

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

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

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

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

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

329  

330 # Memory allocation 

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

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

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

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

335  

336 # Memory deallocation 

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

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

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

340  

341 # IPC 

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

343  

344 # Library 

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

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

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

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

349  

350 # Graph 

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

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

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

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

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

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

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

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

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

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

361  

362 # Linker 

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

364  

365 # Graphics interop 

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

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

368  

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

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

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

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

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

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

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

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

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

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

379  

380 # SM resource split (13.1+) 

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

382  

383 # cuMemcpyWithAttributesAsync (13.2+) 

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

385  

386 # NVRTC 

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

388  

389 # NVVM 

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

391  

392 # nvJitLink 

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

394  

395  

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

397cdef void* _get_driver_fn(str name): (empty)

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

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

400  

401  

402cdef void* _get_optional_driver_fn(str name): (empty)

403 try: (empty)

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

405 except KeyError: 

406 return NULL 

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

408  

409  

410cdef void _init_driver_fn_pointers() noexcept: (empty)

411 global p_cuDevicePrimaryCtxRetain, p_cuDevicePrimaryCtxRelease, p_cuCtxGetCurrent 

412 global p_cuCtxSetCurrent 

413 global p_cuGreenCtxCreate, p_cuGreenCtxDestroy, p_cuCtxFromGreenCtx 

414 global p_cuDevResourceGenerateDesc, p_cuGreenCtxStreamCreate 

415 global p_cuStreamCreateWithPriority, p_cuStreamDestroy 

416 global p_cuEventCreate, p_cuEventDestroy, p_cuIpcOpenEventHandle 

417 global p_cuDeviceGetCount 

418 global p_cuMemPoolSetAccess, p_cuMemPoolDestroy, p_cuMemPoolCreate 

419 global p_cuDeviceGetMemPool, p_cuMemPoolImportFromShareableHandle 

420 global p_cuMemAllocFromPoolAsync, p_cuMemAllocAsync, p_cuMemAlloc, p_cuMemAllocHost 

421 global p_cuMemFreeAsync, p_cuMemFree, p_cuMemFreeHost 

422 global p_cuMemPoolImportPointer 

423 global p_cuLibraryLoadFromFile, p_cuLibraryLoadData, p_cuLibraryUnload, p_cuLibraryGetKernel 

424 global p_cuGraphDestroy, p_cuGraphInstantiateWithParams 

425 global p_cuGraphExecUpdate, p_cuGraphExecDestroy 

426 global p_cuUserObjectCreate, p_cuUserObjectRelease 

427 global p_cuGraphRetainUserObject, p_cuGraphReleaseUserObject 

428 global p_cuGraphNodeFindInClone, p_cuGraphChildGraphNodeGetGraph 

429 global p_cuLinkDestroy 

430 global p_cuGraphicsUnmapResources, p_cuGraphicsUnregisterResource 

431 global p_cuDevSmResourceSplit 

432 global p_cuMemcpyWithAttributesAsync 

433 global p_cuArray3DCreate, p_cuArrayDestroy 

434 global p_cuMipmappedArrayCreate, p_cuMipmappedArrayDestroy, p_cuMipmappedArrayGetLevel 

435 global p_cuTexObjectCreate, p_cuTexObjectDestroy 

436 global p_cuSurfObjectCreate, p_cuSurfObjectDestroy 

437  

438 # Context 

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

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

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

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

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

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

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

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

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

448  

449 # Stream 

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

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

452  

453 # Event 

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

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

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

457  

458 # Device 

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

460  

461 # Memory pool 

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

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

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

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

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

467  

468 # Memory allocation 

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

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

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

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

473  

474 # Memory deallocation 

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

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

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

478  

479 # IPC 

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

481  

482 # Library 

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

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

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

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

487  

488 # Graph 

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

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

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

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

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

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

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

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

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

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

499  

500 # Linker 

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

502  

503 # Graphics interop 

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

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

506  

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

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

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

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

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

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

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

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

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

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

517  

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

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

520  

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

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

523  

524_init_driver_fn_pointers() (empty)

525initialize_deferred_cleanup() (empty)

526  

527# ============================================================================= 

528# NVRTC function pointer initialization 

529# ============================================================================= 

530  

531cdef void* _get_nvrtc_fn(str name): (empty)

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

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

534  

535cdef void _init_nvrtc_fn_pointers() noexcept: (empty)

536 global p_nvrtcDestroyProgram 

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

538  

539_init_nvrtc_fn_pointers() (empty)

540  

541# ============================================================================= 

542# NVVM function pointer initialization 

543# 

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

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

546# ============================================================================= 

547  

548cdef void* _get_nvvm_fn(str name): (empty)

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

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

551  

552cdef void _init_nvvm_fn_pointers() noexcept: (empty)

553 global p_nvvmDestroyProgram 

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

555  

556_init_nvvm_fn_pointers() (empty)

557  

558# ============================================================================= 

559# nvJitLink function pointer initialization 

560# 

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

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

563# ============================================================================= 

564  

565cdef void* _get_nvjitlink_fn(str name): (empty)

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

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

568  

569cdef void _init_nvjitlink_fn_pointers() noexcept: (empty)

570 global p_nvJitLinkDestroy 

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

572  

573_init_nvjitlink_fn_pointers() (empty)