mirror of
https://github.com/NixOS/nixpkgs.git
synced 2026-10-31 01:02:20 +00:00
6254244afd
cuda-bindings loads libcuda and libnvidia-ml by absolute path from the NixOS driver link. Unlike libraries that carry the driver link in their RUNPATH, that ignores LD_LIBRARY_PATH, so on non-NixOS hosts where the driver is exposed that way (nixglhost, container runtimes) every cuda.bindings driver and NVML call fails. Packages that probe the GPU through cuda.bindings then silently fall back: CUTLASS DSL, for one, compiles for its default sm_100a and the kernels fail to launch with cudaErrorNoKernelImageForDevice. Try the soname first and fall back to the configured path, matching the dynamic linker's order for RUNPATH-linked libraries. On NixOS the soname lookup finds nothing and the driver link is used as before. This affects NVML in the 12.9 and 13.1 through 13.4 patches, and libcuda in the 13.4 patch only; earlier versions load libcuda by soname, which the patches leave alone. 13.0 has no NVML module. Assisted-by: Claude Code (Claude Opus 5.5, claude-opus-5-5)
153 lines
6.1 KiB
Diff
153 lines
6.1 KiB
Diff
diff --git a/cuda_bindings/cuda/bindings/_bindings/cynvrtc.pyx.in b/cuda_bindings/cuda/bindings/_bindings/cynvrtc.pyx.in
|
|
index 608aebd1af..7f60917f71 100644
|
|
--- a/cuda_bindings/cuda/bindings/_bindings/cynvrtc.pyx.in
|
|
+++ b/cuda_bindings/cuda/bindings/_bindings/cynvrtc.pyx.in
|
|
@@ -8,7 +8,7 @@ cimport cuda.bindings._lib.windll as windll
|
|
{{else}}
|
|
cimport cuda.bindings._lib.dlfcn as dlfcn
|
|
{{endif}}
|
|
-from cuda.pathfinder import load_nvidia_dynamic_lib
|
|
+from ctypes import CDLL
|
|
from libc.stdint cimport intptr_t, uintptr_t
|
|
import threading
|
|
|
|
@@ -45,7 +45,7 @@ cdef int _cuPythonInit() except -1 nogil:
|
|
# Load library
|
|
with gil, __symbol_lock:
|
|
{{if 'Windows' == platform.system()}}
|
|
- handle = load_nvidia_dynamic_lib("nvrtc")._handle_uint
|
|
+ handle = CDLL("@libnvrtc@/lib/libnvrtc.so")._handle
|
|
|
|
# Load function
|
|
{{if 'nvrtcGetErrorString' in found_functions}}
|
|
@@ -146,7 +146,7 @@ cdef int _cuPythonInit() except -1 nogil:
|
|
{{endif}}
|
|
|
|
{{else}}
|
|
- handle = <void*><uintptr_t>(load_nvidia_dynamic_lib("nvrtc")._handle_uint)
|
|
+ handle = <void*><uintptr_t>(CDLL("@libnvrtc@/lib/libnvrtc.so")._handle)
|
|
|
|
# Load function
|
|
{{if 'nvrtcGetErrorString' in found_functions}}
|
|
diff --git a/cuda_bindings/cuda/bindings/_internal/_nvml_linux.pyx b/cuda_bindings/cuda/bindings/_internal/_nvml_linux.pyx
|
|
index 1622cc82aa..36eba31d32 100644
|
|
--- a/cuda_bindings/cuda/bindings/_internal/_nvml_linux.pyx
|
|
+++ b/cuda_bindings/cuda/bindings/_internal/_nvml_linux.pyx
|
|
@@ -10,7 +10,7 @@ import threading
|
|
|
|
from .utils import FunctionNotFoundError, NotSupportedError
|
|
|
|
-from cuda.pathfinder import load_nvidia_dynamic_lib
|
|
+from ctypes import CDLL
|
|
|
|
|
|
###############################################################################
|
|
@@ -410,7 +410,12 @@ cdef void* __nvmlDeviceGetSramUniqueUncorrectedEccErrorCounts = NULL
|
|
|
|
|
|
cdef void* load_library() except* with gil:
|
|
- return dlopen("libnvidia-ml.so.1", RTLD_NOW | RTLD_GLOBAL)
|
|
+ # Search like the dynamic linker would for a library with the driver link in its RUNPATH: LD_LIBRARY_PATH
|
|
+ # first (e.g. a host driver exposed by nixglhost or a container runtime), then the configured path.
|
|
+ cdef void* handle = dlopen("libnvidia-ml.so.1", RTLD_NOW | RTLD_GLOBAL)
|
|
+ if handle == NULL:
|
|
+ handle = dlopen("@libnvml@/lib/libnvidia-ml.so.1", RTLD_NOW | RTLD_GLOBAL)
|
|
+ return handle
|
|
|
|
|
|
cdef int _init_nvml() except -1 nogil:
|
|
diff --git a/cuda_bindings/cuda/bindings/_internal/cufile_linux.pyx b/cuda_bindings/cuda/bindings/_internal/cufile_linux.pyx
|
|
index c64eb00c5c..9ba4ea2727 100644
|
|
--- a/cuda_bindings/cuda/bindings/_internal/cufile_linux.pyx
|
|
+++ b/cuda_bindings/cuda/bindings/_internal/cufile_linux.pyx
|
|
@@ -9,7 +9,7 @@ import threading
|
|
|
|
from .utils import FunctionNotFoundError, NotSupportedError
|
|
|
|
-from cuda.pathfinder import load_nvidia_dynamic_lib
|
|
+from ctypes import CDLL
|
|
|
|
import cython
|
|
|
|
@@ -107,7 +107,7 @@ cdef void* __cuFileGetParameterPosixPoolSlabArray = NULL
|
|
|
|
|
|
cdef void* load_library() except* with gil:
|
|
- cdef uintptr_t handle = load_nvidia_dynamic_lib("cufile")._handle_uint
|
|
+ cdef uintptr_t handle = CDLL("@libcufile@/lib/libcufile.so")._handle
|
|
return <void*>handle
|
|
|
|
|
|
diff --git a/cuda_bindings/cuda/bindings/_internal/nvjitlink_linux.pyx b/cuda_bindings/cuda/bindings/_internal/nvjitlink_linux.pyx
|
|
index 8c96b6d640..96761a95ae 100644
|
|
--- a/cuda_bindings/cuda/bindings/_internal/nvjitlink_linux.pyx
|
|
+++ b/cuda_bindings/cuda/bindings/_internal/nvjitlink_linux.pyx
|
|
@@ -9,7 +9,7 @@ from libc.stdint cimport intptr_t, uintptr_t
|
|
import threading
|
|
from .utils import FunctionNotFoundError, NotSupportedError
|
|
|
|
-from cuda.pathfinder import load_nvidia_dynamic_lib
|
|
+from ctypes import CDLL
|
|
|
|
|
|
###############################################################################
|
|
@@ -76,7 +76,7 @@ cdef void* __nvJitLinkVersion = NULL
|
|
|
|
|
|
cdef void* load_library() except* with gil:
|
|
- cdef uintptr_t handle = load_nvidia_dynamic_lib("nvJitLink")._handle_uint
|
|
+ cdef uintptr_t handle = CDLL("@libnvjitlink@/lib/libnvJitLink.so")._handle
|
|
return <void*>handle
|
|
|
|
|
|
diff --git a/cuda_bindings/cuda/bindings/_internal/nvvm_linux.pyx b/cuda_bindings/cuda/bindings/_internal/nvvm_linux.pyx
|
|
index 408a2cb592..2b31888e1f 100644
|
|
--- a/cuda_bindings/cuda/bindings/_internal/nvvm_linux.pyx
|
|
+++ b/cuda_bindings/cuda/bindings/_internal/nvvm_linux.pyx
|
|
@@ -9,7 +9,7 @@ from libc.stdint cimport intptr_t, uintptr_t
|
|
import threading
|
|
from .utils import FunctionNotFoundError, NotSupportedError
|
|
|
|
-from cuda.pathfinder import load_nvidia_dynamic_lib
|
|
+from ctypes import CDLL
|
|
|
|
|
|
###############################################################################
|
|
@@ -75,7 +75,7 @@ cdef void* __nvvmGetProgramLog = NULL
|
|
|
|
|
|
cdef void* load_library() except* with gil:
|
|
- cdef uintptr_t handle = load_nvidia_dynamic_lib("nvvm")._handle_uint
|
|
+ cdef uintptr_t handle = CDLL("@libnvvm@/lib/libnvvm.so")._handle
|
|
return <void*>handle
|
|
|
|
|
|
diff --git a/cuda_bindings/cuda/bindings/cyruntime.pyx.in b/cuda_bindings/cuda/bindings/cyruntime.pyx.in
|
|
index 248346d274..e041b41d69 100644
|
|
--- a/cuda_bindings/cuda/bindings/cyruntime.pyx.in
|
|
+++ b/cuda_bindings/cuda/bindings/cyruntime.pyx.in
|
|
@@ -2012,7 +2012,7 @@ cdef cudaError_t cudaGraphicsVDPAURegisterOutputSurface(cudaGraphicsResource** r
|
|
{{if True}}
|
|
|
|
from libc.stdint cimport uintptr_t
|
|
-from cuda.pathfinder import load_nvidia_dynamic_lib
|
|
+from ctypes import CDLL
|
|
{{if 'Windows' == platform.system()}}
|
|
cimport cuda.bindings._lib.windll as windll
|
|
{{else}}
|
|
@@ -2022,11 +2022,11 @@ cimport cuda.bindings._lib.dlfcn as dlfcn
|
|
cdef cudaError_t getLocalRuntimeVersion(int* runtimeVersion) except ?cudaErrorCallRequiresNewerDriver nogil:
|
|
# Load
|
|
with gil:
|
|
- loaded_dl = load_nvidia_dynamic_lib("cudart")
|
|
+ loaded_dl = CDLL("@libcudart@/lib/libcudart.so")
|
|
{{if 'Windows' == platform.system()}}
|
|
- handle = <uintptr_t>loaded_dl._handle_uint
|
|
+ handle = <uintptr_t>loaded_dl._handle
|
|
{{else}}
|
|
- handle = <void *><uintptr_t>loaded_dl._handle_uint
|
|
+ handle = <void *><uintptr_t>loaded_dl._handle
|
|
{{endif}}
|
|
|
|
{{if 'Windows' == platform.system()}}
|