Skip to content

Fix #938: Call win32 APIs directly #942

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

Merged
merged 7 commits into from
Sep 10, 2025
Merged

Fix #938: Call win32 APIs directly #942

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
7828876
Fix #938: Call win32 APIs directly
mdboom Sep 4, 2025
8c7ea2e
Address comments from PR
mdboom Sep 4, 2025
e0e868a
Address comments from PR
mdboom Sep 5, 2025
7a46173
Remove APIs
mdboom Sep 5, 2025
488bdc2
Don't check return type
mdboom Sep 5, 2025
673974c
Address comments in PR
mdboom Sep 8, 2025
dcce4f5
Merge branch 'main' into issue938
mdboom Sep 10, 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
3,898 changes: 1,115 additions & 2,783 deletions cuda_bindings/cuda/bindings/_bindings/cydriver.pyx.in
View file
Open in desktop

Large diffs are not rendered by default.

173 changes: 50 additions & 123 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 @@ -4,13 +4,12 @@
# This code was automatically generated with version 13.0.0. Do not modify it directly.
{{if 'Windows' == platform.system()}}
import os
import win32api
cimport cuda.bindings._lib.windll as windll
{{else}}
cimport cuda.bindings._lib.dlfcn as dlfcn
from libc.stdint cimport uintptr_t
{{endif}}
from cuda.pathfinder import load_nvidia_dynamic_lib
from libc.stdint cimport intptr_t
from libc.stdint cimport intptr_t, uintptr_t
import threading

cdef object __symbol_lock = threading.Lock()
Expand Down Expand Up @@ -50,172 +49,100 @@ cdef int _cuPythonInit() except -1 nogil:

# Load function
{{if 'nvrtcGetErrorString' in found_functions}}
try:
global __nvrtcGetErrorString
__nvrtcGetErrorString = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetErrorString')
except:
pass
global __nvrtcGetErrorString
__nvrtcGetErrorString = windll.GetProcAddress(handle, 'nvrtcGetErrorString')
{{endif}}
{{if 'nvrtcVersion' in found_functions}}
try:
global __nvrtcVersion
__nvrtcVersion = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcVersion')
except:
pass
global __nvrtcVersion
__nvrtcVersion = windll.GetProcAddress(handle, 'nvrtcVersion')
{{endif}}
{{if 'nvrtcGetNumSupportedArchs' in found_functions}}
try:
global __nvrtcGetNumSupportedArchs
__nvrtcGetNumSupportedArchs = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetNumSupportedArchs')
except:
pass
global __nvrtcGetNumSupportedArchs
__nvrtcGetNumSupportedArchs = windll.GetProcAddress(handle, 'nvrtcGetNumSupportedArchs')
{{endif}}
{{if 'nvrtcGetSupportedArchs' in found_functions}}
try:
global __nvrtcGetSupportedArchs
__nvrtcGetSupportedArchs = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetSupportedArchs')
except:
pass
global __nvrtcGetSupportedArchs
__nvrtcGetSupportedArchs = windll.GetProcAddress(handle, 'nvrtcGetSupportedArchs')
{{endif}}
{{if 'nvrtcCreateProgram' in found_functions}}
try:
global __nvrtcCreateProgram
__nvrtcCreateProgram = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcCreateProgram')
except:
pass
global __nvrtcCreateProgram
__nvrtcCreateProgram = windll.GetProcAddress(handle, 'nvrtcCreateProgram')
{{endif}}
{{if 'nvrtcDestroyProgram' in found_functions}}
try:
global __nvrtcDestroyProgram
__nvrtcDestroyProgram = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcDestroyProgram')
except:
pass
global __nvrtcDestroyProgram
__nvrtcDestroyProgram = windll.GetProcAddress(handle, 'nvrtcDestroyProgram')
{{endif}}
{{if 'nvrtcCompileProgram' in found_functions}}
try:
global __nvrtcCompileProgram
__nvrtcCompileProgram = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcCompileProgram')
except:
pass
global __nvrtcCompileProgram
__nvrtcCompileProgram = windll.GetProcAddress(handle, 'nvrtcCompileProgram')
{{endif}}
{{if 'nvrtcGetPTXSize' in found_functions}}
try:
global __nvrtcGetPTXSize
__nvrtcGetPTXSize = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetPTXSize')
except:
pass
global __nvrtcGetPTXSize
__nvrtcGetPTXSize = windll.GetProcAddress(handle, 'nvrtcGetPTXSize')
{{endif}}
{{if 'nvrtcGetPTX' in found_functions}}
try:
global __nvrtcGetPTX
__nvrtcGetPTX = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetPTX')
except:
pass
global __nvrtcGetPTX
__nvrtcGetPTX = windll.GetProcAddress(handle, 'nvrtcGetPTX')
{{endif}}
{{if 'nvrtcGetCUBINSize' in found_functions}}
try:
global __nvrtcGetCUBINSize
__nvrtcGetCUBINSize = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetCUBINSize')
except:
pass
global __nvrtcGetCUBINSize
__nvrtcGetCUBINSize = windll.GetProcAddress(handle, 'nvrtcGetCUBINSize')
{{endif}}
{{if 'nvrtcGetCUBIN' in found_functions}}
try:
global __nvrtcGetCUBIN
__nvrtcGetCUBIN = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetCUBIN')
except:
pass
global __nvrtcGetCUBIN
__nvrtcGetCUBIN = windll.GetProcAddress(handle, 'nvrtcGetCUBIN')
{{endif}}
{{if 'nvrtcGetLTOIRSize' in found_functions}}
try:
global __nvrtcGetLTOIRSize
__nvrtcGetLTOIRSize = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetLTOIRSize')
except:
pass
global __nvrtcGetLTOIRSize
__nvrtcGetLTOIRSize = windll.GetProcAddress(handle, 'nvrtcGetLTOIRSize')
{{endif}}
{{if 'nvrtcGetLTOIR' in found_functions}}
try:
global __nvrtcGetLTOIR
__nvrtcGetLTOIR = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetLTOIR')
except:
pass
global __nvrtcGetLTOIR
__nvrtcGetLTOIR = windll.GetProcAddress(handle, 'nvrtcGetLTOIR')
{{endif}}
{{if 'nvrtcGetOptiXIRSize' in found_functions}}
try:
global __nvrtcGetOptiXIRSize
__nvrtcGetOptiXIRSize = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetOptiXIRSize')
except:
pass
global __nvrtcGetOptiXIRSize
__nvrtcGetOptiXIRSize = windll.GetProcAddress(handle, 'nvrtcGetOptiXIRSize')
{{endif}}
{{if 'nvrtcGetOptiXIR' in found_functions}}
try:
global __nvrtcGetOptiXIR
__nvrtcGetOptiXIR = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetOptiXIR')
except:
pass
global __nvrtcGetOptiXIR
__nvrtcGetOptiXIR = windll.GetProcAddress(handle, 'nvrtcGetOptiXIR')
{{endif}}
{{if 'nvrtcGetProgramLogSize' in found_functions}}
try:
global __nvrtcGetProgramLogSize
__nvrtcGetProgramLogSize = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetProgramLogSize')
except:
pass
global __nvrtcGetProgramLogSize
__nvrtcGetProgramLogSize = windll.GetProcAddress(handle, 'nvrtcGetProgramLogSize')
{{endif}}
{{if 'nvrtcGetProgramLog' in found_functions}}
try:
global __nvrtcGetProgramLog
__nvrtcGetProgramLog = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetProgramLog')
except:
pass
global __nvrtcGetProgramLog
__nvrtcGetProgramLog = windll.GetProcAddress(handle, 'nvrtcGetProgramLog')
{{endif}}
{{if 'nvrtcAddNameExpression' in found_functions}}
try:
global __nvrtcAddNameExpression
__nvrtcAddNameExpression = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcAddNameExpression')
except:
pass
global __nvrtcAddNameExpression
__nvrtcAddNameExpression = windll.GetProcAddress(handle, 'nvrtcAddNameExpression')
{{endif}}
{{if 'nvrtcGetLoweredName' in found_functions}}
try:
global __nvrtcGetLoweredName
__nvrtcGetLoweredName = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetLoweredName')
except:
pass
global __nvrtcGetLoweredName
__nvrtcGetLoweredName = windll.GetProcAddress(handle, 'nvrtcGetLoweredName')
{{endif}}
{{if 'nvrtcGetPCHHeapSize' in found_functions}}
try:
global __nvrtcGetPCHHeapSize
__nvrtcGetPCHHeapSize = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetPCHHeapSize')
except:
pass
global __nvrtcGetPCHHeapSize
__nvrtcGetPCHHeapSize = windll.GetProcAddress(handle, 'nvrtcGetPCHHeapSize')
{{endif}}
{{if 'nvrtcSetPCHHeapSize' in found_functions}}
try:
global __nvrtcSetPCHHeapSize
__nvrtcSetPCHHeapSize = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcSetPCHHeapSize')
except:
pass
global __nvrtcSetPCHHeapSize
__nvrtcSetPCHHeapSize = windll.GetProcAddress(handle, 'nvrtcSetPCHHeapSize')
{{endif}}
{{if 'nvrtcGetPCHCreateStatus' in found_functions}}
try:
global __nvrtcGetPCHCreateStatus
__nvrtcGetPCHCreateStatus = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetPCHCreateStatus')
except:
pass
global __nvrtcGetPCHCreateStatus
__nvrtcGetPCHCreateStatus = windll.GetProcAddress(handle, 'nvrtcGetPCHCreateStatus')
{{endif}}
{{if 'nvrtcGetPCHHeapSizeRequired' in found_functions}}
try:
global __nvrtcGetPCHHeapSizeRequired
__nvrtcGetPCHHeapSizeRequired = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcGetPCHHeapSizeRequired')
except:
pass
global __nvrtcGetPCHHeapSizeRequired
__nvrtcGetPCHHeapSizeRequired = windll.GetProcAddress(handle, 'nvrtcGetPCHHeapSizeRequired')
{{endif}}
{{if 'nvrtcSetFlowCallback' in found_functions}}
try:
global __nvrtcSetFlowCallback
__nvrtcSetFlowCallback = <void*><unsigned long long>win32api.GetProcAddress(handle, 'nvrtcSetFlowCallback')
except:
pass
global __nvrtcSetFlowCallback
__nvrtcSetFlowCallback = windll.GetProcAddress(handle, 'nvrtcSetFlowCallback')
{{endif}}

{{else}}
Expand Down
36 changes: 19 additions & 17 deletions cuda_bindings/cuda/bindings/_internal/cufile_linux.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,14 +32,31 @@ cdef extern from "<dlfcn.h>" nogil:

const void* RTLD_DEFAULT 'RTLD_DEFAULT'

cdef int get_cuda_version():
cdef void* handle = NULL
cdef int err, driver_ver = 0

# Load driver to check version
handle = dlopen('libcuda.so.1', RTLD_NOW | RTLD_GLOBAL)
if handle == NULL:
err_msg = dlerror()
raise NotSupportedError(f'CUDA driver is not found ({err_msg.decode()})')
cuDriverGetVersion = dlsym(handle, "cuDriverGetVersion")
if cuDriverGetVersion == NULL:
raise RuntimeError('something went wrong')
err = (<int (*)(int*) noexcept nogil>cuDriverGetVersion)(&driver_ver)
if err != 0:
raise RuntimeError('something went wrong')

return driver_ver


###############################################################################
# Wrapper init
###############################################################################

cdef object __symbol_lock = threading.Lock()
cdef bint __py_cufile_init = False
cdef void* __cuDriverGetVersion = NULL

cdef void* __cuFileHandleRegister = NULL
cdef void* __cuFileHandleDeregister = NULL
Expand Down Expand Up @@ -97,24 +114,9 @@ cdef int _check_or_init_cufile() except -1 nogil:
return 0

cdef void* handle = NULL
cdef int err, driver_ver = 0

with gil, __symbol_lock:
# Load driver to check version
handle = dlopen('libcuda.so.1', RTLD_NOW | RTLD_GLOBAL)
if handle == NULL:
err_msg = dlerror()
raise NotSupportedError(f'CUDA driver is not found ({err_msg.decode()})')
global __cuDriverGetVersion
if __cuDriverGetVersion == NULL:
__cuDriverGetVersion = dlsym(handle, "cuDriverGetVersion")
if __cuDriverGetVersion == NULL:
raise RuntimeError('something went wrong')
err = (<int (*)(int*) noexcept nogil>__cuDriverGetVersion)(&driver_ver)
if err != 0:
raise RuntimeError('something went wrong')
#dlclose(handle)
handle = NULL
driver_ver = get_cuda_version()

# Load function
global __cuFileHandleRegister
Expand Down
36 changes: 19 additions & 17 deletions cuda_bindings/cuda/bindings/_internal/nvjitlink_linux.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,14 +30,31 @@ cdef extern from "<dlfcn.h>" nogil:

const void* RTLD_DEFAULT 'RTLD_DEFAULT'

cdef int get_cuda_version():
cdef void* handle = NULL
cdef int err, driver_ver = 0

# Load driver to check version
handle = dlopen('libcuda.so.1', RTLD_NOW | RTLD_GLOBAL)
if handle == NULL:
err_msg = dlerror()
raise NotSupportedError(f'CUDA driver is not found ({err_msg.decode()})')
cuDriverGetVersion = dlsym(handle, "cuDriverGetVersion")
if cuDriverGetVersion == NULL:
raise RuntimeError('something went wrong')
err = (<int (*)(int*) noexcept nogil>cuDriverGetVersion)(&driver_ver)
if err != 0:
raise RuntimeError('something went wrong')

return driver_ver


###############################################################################
# Wrapper init
###############################################################################

cdef object __symbol_lock = threading.Lock()
cdef bint __py_nvjitlink_init = False
cdef void* __cuDriverGetVersion = NULL

cdef void* __nvJitLinkCreate = NULL
cdef void* __nvJitLinkDestroy = NULL
Expand Down Expand Up @@ -66,24 +83,9 @@ cdef int _check_or_init_nvjitlink() except -1 nogil:
return 0

cdef void* handle = NULL
cdef int err, driver_ver = 0

with gil, __symbol_lock:
# Load driver to check version
handle = dlopen('libcuda.so.1', RTLD_NOW | RTLD_GLOBAL)
if handle == NULL:
err_msg = dlerror()
raise NotSupportedError(f'CUDA driver is not found ({err_msg.decode()})')
global __cuDriverGetVersion
if __cuDriverGetVersion == NULL:
__cuDriverGetVersion = dlsym(handle, "cuDriverGetVersion")
if __cuDriverGetVersion == NULL:
raise RuntimeError('something went wrong')
err = (<int (*)(int*) noexcept nogil>__cuDriverGetVersion)(&driver_ver)
if err != 0:
raise RuntimeError('something went wrong')
#dlclose(handle)
handle = NULL
driver_ver = get_cuda_version()

# Load function
global __nvJitLinkCreate
Expand Down
Loading
Loading