Skip to content

[bugfix]put nccl window register/deregister behind cuda platform #25608

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

Closed
wants to merge 2 commits into from
Closed

[bugfix]put nccl window register/deregister behind cuda platform #25608

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
29fc66a
put nccl window register/deregister behind cuda platform
Amir-19 Sep 24, 2025
5f7499c
typo
Amir-19 Sep 24, 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
10 changes: 8 additions & 2 deletions vllm/distributed/device_communicators/pynccl_wrapper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,7 @@
from torch.distributed import ReduceOp

from vllm.logger import init_logger
from vllm.platforms import current_platform
from vllm.utils import find_nccl_library

logger = init_logger(__name__)
Expand Down Expand Up @@ -223,6 +224,9 @@ class NCCLLibrary:
Function("ncclGroupStart", ncclResult_t, []),
# ncclResult_t ncclGroupEnd();
Function("ncclGroupEnd", ncclResult_t, []),
]

exported_functions_cuda_specific = [
# ncclResult_t ncclCommWindowRegister(
# ncclComm_t comm, void* buff, size_t size,
# ncclWindow_t* win, int winFlags);
Expand Down Expand Up @@ -271,10 +275,12 @@ def __init__(self, so_file: Optional[str] = None):
" to point to the correct nccl library path.", so_file,
platform.platform())
raise e

function_specs = list(NCCLLibrary.exported_functions)
if current_platform.is_cuda():
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

These functions should exist on both platforms in NCCL >= 2.27.03.
The original issue was that on ROCm trying to import a non-existent symbol causes a crash, while on CUDA (apparently) it does not.

function_specs.extend(NCCLLibrary.exported_functions_cuda_specific)
if so_file not in NCCLLibrary.path_to_dict_mapping:
_funcs: dict[str, Any] = {}
for func in NCCLLibrary.exported_functions:
for func in function_specs:
f = getattr(self.lib, func.name)
f.restype = func.restype
f.argtypes = func.argtypes
Expand Down