Skip to content

[Backport] Make populating the internal symbol table thread-safe #836

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 6 commits into
base: 12.9.x
Choose a base branch
from
Open

[Backport] Make populating the internal symbol table thread-safe #836

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
cd802e2
protect cuPythonInit in driver
leofang Aug 13, 2025
66a50b7
fixes for other modules
kkraus14 Aug 14, 2025
2459e09
fixes
kkraus14 Aug 14, 2025
58b7074
move init check inside of lock
kkraus14 Aug 14, 2025
1be81a0
change lock to RLock to handle recursion issue
kkraus14 Aug 14, 2025
cd45744
make cuPythonInit reentrant + ensure GIL is released when calling und…
leofang Aug 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16,507 changes: 8,258 additions & 8,249 deletions cuda_bindings/cuda/bindings/_bindings/cydriver.pyx.in
View file
Open in desktop

Large diffs are not rendered by default.

234 changes: 116 additions & 118 deletions cuda_bindings/cuda/bindings/_bindings/cynvrtc.pyx.in
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,8 @@ from libc.stdint cimport uintptr_t
{{endif}}
from cuda.pathfinder import load_nvidia_dynamic_lib
from libc.stdint cimport intptr_t

import threading
cdef object __symbol_lock = threading.Lock()
cdef bint __cuPythonInit = False
{{if 'nvrtcGetErrorString' in found_functions}}cdef void *__nvrtcGetErrorString = NULL{{endif}}
{{if 'nvrtcVersion' in found_functions}}cdef void *__nvrtcVersion = NULL{{endif}}
Expand Down Expand Up @@ -44,21 +45,13 @@ cdef int cuPythonInit() except -1 nogil:
global __cuPythonInit
if __cuPythonInit:
return 0
__cuPythonInit = True

# Load library
{{if 'Windows' == platform.system()}}
with gil:
with gil, __symbol_lock:
{{if 'Windows' == platform.system()}}
handle = load_nvidia_dynamic_lib("nvrtc")._handle_uint
{{else}}
with gil:
handle = <void*><uintptr_t>load_nvidia_dynamic_lib("nvrtc")._handle_uint
{{endif}}


# Load function
{{if 'Windows' == platform.system()}}
with gil:
# Load function
{{if 'nvrtcGetErrorString' in found_functions}}
try:
global __nvrtcGetErrorString
Expand Down Expand Up @@ -242,113 +235,118 @@ cdef int cuPythonInit() except -1 nogil:
pass
{{endif}}

{{else}}
{{if 'nvrtcGetErrorString' in found_functions}}
global __nvrtcGetErrorString
__nvrtcGetErrorString = dlfcn.dlsym(handle, 'nvrtcGetErrorString')
{{endif}}
{{if 'nvrtcVersion' in found_functions}}
global __nvrtcVersion
__nvrtcVersion = dlfcn.dlsym(handle, 'nvrtcVersion')
{{endif}}
{{if 'nvrtcGetNumSupportedArchs' in found_functions}}
global __nvrtcGetNumSupportedArchs
__nvrtcGetNumSupportedArchs = dlfcn.dlsym(handle, 'nvrtcGetNumSupportedArchs')
{{endif}}
{{if 'nvrtcGetSupportedArchs' in found_functions}}
global __nvrtcGetSupportedArchs
__nvrtcGetSupportedArchs = dlfcn.dlsym(handle, 'nvrtcGetSupportedArchs')
{{endif}}
{{if 'nvrtcCreateProgram' in found_functions}}
global __nvrtcCreateProgram
__nvrtcCreateProgram = dlfcn.dlsym(handle, 'nvrtcCreateProgram')
{{endif}}
{{if 'nvrtcDestroyProgram' in found_functions}}
global __nvrtcDestroyProgram
__nvrtcDestroyProgram = dlfcn.dlsym(handle, 'nvrtcDestroyProgram')
{{endif}}
{{if 'nvrtcCompileProgram' in found_functions}}
global __nvrtcCompileProgram
__nvrtcCompileProgram = dlfcn.dlsym(handle, 'nvrtcCompileProgram')
{{endif}}
{{if 'nvrtcGetPTXSize' in found_functions}}
global __nvrtcGetPTXSize
__nvrtcGetPTXSize = dlfcn.dlsym(handle, 'nvrtcGetPTXSize')
{{endif}}
{{if 'nvrtcGetPTX' in found_functions}}
global __nvrtcGetPTX
__nvrtcGetPTX = dlfcn.dlsym(handle, 'nvrtcGetPTX')
{{endif}}
{{if 'nvrtcGetCUBINSize' in found_functions}}
global __nvrtcGetCUBINSize
__nvrtcGetCUBINSize = dlfcn.dlsym(handle, 'nvrtcGetCUBINSize')
{{endif}}
{{if 'nvrtcGetCUBIN' in found_functions}}
global __nvrtcGetCUBIN
__nvrtcGetCUBIN = dlfcn.dlsym(handle, 'nvrtcGetCUBIN')
{{endif}}
{{if 'nvrtcGetNVVMSize' in found_functions}}
global __nvrtcGetNVVMSize
__nvrtcGetNVVMSize = dlfcn.dlsym(handle, 'nvrtcGetNVVMSize')
{{endif}}
{{if 'nvrtcGetNVVM' in found_functions}}
global __nvrtcGetNVVM
__nvrtcGetNVVM = dlfcn.dlsym(handle, 'nvrtcGetNVVM')
{{endif}}
{{if 'nvrtcGetLTOIRSize' in found_functions}}
global __nvrtcGetLTOIRSize
__nvrtcGetLTOIRSize = dlfcn.dlsym(handle, 'nvrtcGetLTOIRSize')
{{endif}}
{{if 'nvrtcGetLTOIR' in found_functions}}
global __nvrtcGetLTOIR
__nvrtcGetLTOIR = dlfcn.dlsym(handle, 'nvrtcGetLTOIR')
{{endif}}
{{if 'nvrtcGetOptiXIRSize' in found_functions}}
global __nvrtcGetOptiXIRSize
__nvrtcGetOptiXIRSize = dlfcn.dlsym(handle, 'nvrtcGetOptiXIRSize')
{{endif}}
{{if 'nvrtcGetOptiXIR' in found_functions}}
global __nvrtcGetOptiXIR
__nvrtcGetOptiXIR = dlfcn.dlsym(handle, 'nvrtcGetOptiXIR')
{{endif}}
{{if 'nvrtcGetProgramLogSize' in found_functions}}
global __nvrtcGetProgramLogSize
__nvrtcGetProgramLogSize = dlfcn.dlsym(handle, 'nvrtcGetProgramLogSize')
{{endif}}
{{if 'nvrtcGetProgramLog' in found_functions}}
global __nvrtcGetProgramLog
__nvrtcGetProgramLog = dlfcn.dlsym(handle, 'nvrtcGetProgramLog')
{{endif}}
{{if 'nvrtcAddNameExpression' in found_functions}}
global __nvrtcAddNameExpression
__nvrtcAddNameExpression = dlfcn.dlsym(handle, 'nvrtcAddNameExpression')
{{endif}}
{{if 'nvrtcGetLoweredName' in found_functions}}
global __nvrtcGetLoweredName
__nvrtcGetLoweredName = dlfcn.dlsym(handle, 'nvrtcGetLoweredName')
{{endif}}
{{if 'nvrtcGetPCHHeapSize' in found_functions}}
global __nvrtcGetPCHHeapSize
__nvrtcGetPCHHeapSize = dlfcn.dlsym(handle, 'nvrtcGetPCHHeapSize')
{{endif}}
{{if 'nvrtcSetPCHHeapSize' in found_functions}}
global __nvrtcSetPCHHeapSize
__nvrtcSetPCHHeapSize = dlfcn.dlsym(handle, 'nvrtcSetPCHHeapSize')
{{endif}}
{{if 'nvrtcGetPCHCreateStatus' in found_functions}}
global __nvrtcGetPCHCreateStatus
__nvrtcGetPCHCreateStatus = dlfcn.dlsym(handle, 'nvrtcGetPCHCreateStatus')
{{endif}}
{{if 'nvrtcGetPCHHeapSizeRequired' in found_functions}}
global __nvrtcGetPCHHeapSizeRequired
__nvrtcGetPCHHeapSizeRequired = dlfcn.dlsym(handle, 'nvrtcGetPCHHeapSizeRequired')
{{endif}}
{{if 'nvrtcSetFlowCallback' in found_functions}}
global __nvrtcSetFlowCallback
__nvrtcSetFlowCallback = dlfcn.dlsym(handle, 'nvrtcSetFlowCallback')
{{endif}}
{{else}}
handle = <void*><uintptr_t>(load_nvidia_dynamic_lib("nvrtc")._handle_uint)

{{endif}}
# Load function
{{if 'nvrtcGetErrorString' in found_functions}}
global __nvrtcGetErrorString
__nvrtcGetErrorString = dlfcn.dlsym(handle, 'nvrtcGetErrorString')
{{endif}}
{{if 'nvrtcVersion' in found_functions}}
global __nvrtcVersion
__nvrtcVersion = dlfcn.dlsym(handle, 'nvrtcVersion')
{{endif}}
{{if 'nvrtcGetNumSupportedArchs' in found_functions}}
global __nvrtcGetNumSupportedArchs
__nvrtcGetNumSupportedArchs = dlfcn.dlsym(handle, 'nvrtcGetNumSupportedArchs')
{{endif}}
{{if 'nvrtcGetSupportedArchs' in found_functions}}
global __nvrtcGetSupportedArchs
__nvrtcGetSupportedArchs = dlfcn.dlsym(handle, 'nvrtcGetSupportedArchs')
{{endif}}
{{if 'nvrtcCreateProgram' in found_functions}}
global __nvrtcCreateProgram
__nvrtcCreateProgram = dlfcn.dlsym(handle, 'nvrtcCreateProgram')
{{endif}}
{{if 'nvrtcDestroyProgram' in found_functions}}
global __nvrtcDestroyProgram
__nvrtcDestroyProgram = dlfcn.dlsym(handle, 'nvrtcDestroyProgram')
{{endif}}
{{if 'nvrtcCompileProgram' in found_functions}}
global __nvrtcCompileProgram
__nvrtcCompileProgram = dlfcn.dlsym(handle, 'nvrtcCompileProgram')
{{endif}}
{{if 'nvrtcGetPTXSize' in found_functions}}
global __nvrtcGetPTXSize
__nvrtcGetPTXSize = dlfcn.dlsym(handle, 'nvrtcGetPTXSize')
{{endif}}
{{if 'nvrtcGetPTX' in found_functions}}
global __nvrtcGetPTX
__nvrtcGetPTX = dlfcn.dlsym(handle, 'nvrtcGetPTX')
{{endif}}
{{if 'nvrtcGetCUBINSize' in found_functions}}
global __nvrtcGetCUBINSize
__nvrtcGetCUBINSize = dlfcn.dlsym(handle, 'nvrtcGetCUBINSize')
{{endif}}
{{if 'nvrtcGetCUBIN' in found_functions}}
global __nvrtcGetCUBIN
__nvrtcGetCUBIN = dlfcn.dlsym(handle, 'nvrtcGetCUBIN')
{{endif}}
{{if 'nvrtcGetNVVMSize' in found_functions}}
global __nvrtcGetNVVMSize
__nvrtcGetNVVMSize = dlfcn.dlsym(handle, 'nvrtcGetNVVMSize')
{{endif}}
{{if 'nvrtcGetNVVM' in found_functions}}
global __nvrtcGetNVVM
__nvrtcGetNVVM = dlfcn.dlsym(handle, 'nvrtcGetNVVM')
{{endif}}
{{if 'nvrtcGetLTOIRSize' in found_functions}}
global __nvrtcGetLTOIRSize
__nvrtcGetLTOIRSize = dlfcn.dlsym(handle, 'nvrtcGetLTOIRSize')
{{endif}}
{{if 'nvrtcGetLTOIR' in found_functions}}
global __nvrtcGetLTOIR
__nvrtcGetLTOIR = dlfcn.dlsym(handle, 'nvrtcGetLTOIR')
{{endif}}
{{if 'nvrtcGetOptiXIRSize' in found_functions}}
global __nvrtcGetOptiXIRSize
__nvrtcGetOptiXIRSize = dlfcn.dlsym(handle, 'nvrtcGetOptiXIRSize')
{{endif}}
{{if 'nvrtcGetOptiXIR' in found_functions}}
global __nvrtcGetOptiXIR
__nvrtcGetOptiXIR = dlfcn.dlsym(handle, 'nvrtcGetOptiXIR')
{{endif}}
{{if 'nvrtcGetProgramLogSize' in found_functions}}
global __nvrtcGetProgramLogSize
__nvrtcGetProgramLogSize = dlfcn.dlsym(handle, 'nvrtcGetProgramLogSize')
{{endif}}
{{if 'nvrtcGetProgramLog' in found_functions}}
global __nvrtcGetProgramLog
__nvrtcGetProgramLog = dlfcn.dlsym(handle, 'nvrtcGetProgramLog')
{{endif}}
{{if 'nvrtcAddNameExpression' in found_functions}}
global __nvrtcAddNameExpression
__nvrtcAddNameExpression = dlfcn.dlsym(handle, 'nvrtcAddNameExpression')
{{endif}}
{{if 'nvrtcGetLoweredName' in found_functions}}
global __nvrtcGetLoweredName
__nvrtcGetLoweredName = dlfcn.dlsym(handle, 'nvrtcGetLoweredName')
{{endif}}
{{if 'nvrtcGetPCHHeapSize' in found_functions}}
global __nvrtcGetPCHHeapSize
__nvrtcGetPCHHeapSize = dlfcn.dlsym(handle, 'nvrtcGetPCHHeapSize')
{{endif}}
{{if 'nvrtcSetPCHHeapSize' in found_functions}}
global __nvrtcSetPCHHeapSize
__nvrtcSetPCHHeapSize = dlfcn.dlsym(handle, 'nvrtcSetPCHHeapSize')
{{endif}}
{{if 'nvrtcGetPCHCreateStatus' in found_functions}}
global __nvrtcGetPCHCreateStatus
__nvrtcGetPCHCreateStatus = dlfcn.dlsym(handle, 'nvrtcGetPCHCreateStatus')
{{endif}}
{{if 'nvrtcGetPCHHeapSizeRequired' in found_functions}}
global __nvrtcGetPCHHeapSizeRequired
__nvrtcGetPCHHeapSizeRequired = dlfcn.dlsym(handle, 'nvrtcGetPCHHeapSizeRequired')
{{endif}}
{{if 'nvrtcSetFlowCallback' in found_functions}}
global __nvrtcSetFlowCallback
__nvrtcSetFlowCallback = dlfcn.dlsym(handle, 'nvrtcSetFlowCallback')
{{endif}}
{{endif}}

__cuPythonInit = True
return 0

{{if 'nvrtcGetErrorString' in found_functions}}

Expand Down
Loading
Loading