Skip to content

Xccl option #20

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 3 commits into
base: master
Choose a base branch
from
Open

Xccl option #20

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
836ed81
add options for xccl work
Chao1Han Jun 4, 2025
70764ff
add comm split support
Chao1Han Jul 7, 2025
aba0961
update
Chao1Han Jul 16, 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
60 changes: 58 additions & 2 deletions torch/csrc/distributed/c10d/init.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3369,6 +3369,24 @@ Example::
auto processGroupXCCL =
intrusive_ptr_no_gil_destructor_class_<::c10d::ProcessGroupXCCL>(
module, "ProcessGroupXCCL", backend)
.def(
py::init([](const c10::intrusive_ptr<::c10d::Store>& store,
int rank,
int size,
c10::intrusive_ptr<::c10d::ProcessGroupXCCL::Options>
options) {
// gil_scoped_release is not safe as a call_guard in init.
// https://github.com/pybind/pybind11/issues/5473
py::gil_scoped_release nogil{};

return c10::make_intrusive<::c10d::ProcessGroupXCCL>(
store, rank, size, std::move(options));
}),
py::arg("store"),
py::arg("rank"),
py::arg("size"),
py::arg("options"),
R"(Create a new ProcessGroupXCCL instance.)")
.def(
py::init([](const c10::intrusive_ptr<::c10d::Store>& store,
int rank,
Expand All @@ -3377,12 +3395,50 @@ Example::
// https://github.com/pybind/pybind11/issues/5473
py::gil_scoped_release nogil{};

auto options = ::c10d::ProcessGroupXCCL::Options::create();
options->is_high_priority_stream = false;
return c10::make_intrusive<::c10d::ProcessGroupXCCL>(
store, rank, size);
store, rank, size, options);
}),
py::arg("store"),
py::arg("rank"),
py::arg("size"));
py::arg("size"),
R"(Create a new ProcessGroupXCCL instance.)")
.def(
"comm_split_count",
&::c10d::ProcessGroupXCCL::getCommSplitCounter)
.def_property_readonly(
"options",
&::c10d::ProcessGroupXCCL::getOptions,
R"(Return the options used to create this ProcessGroupXCCL instance.)")
.def_property(
"bound_device_id",
&::c10d::ProcessGroupXCCL::getBoundDeviceId,
&::c10d::ProcessGroupXCCL::setBoundDeviceId,
R"(Return the bound device id.)")
.def(
"perform_nocolor_split",
&::c10d::ProcessGroupXCCL::performNocolorSplit)
.def(
"_is_initialized",
&::c10d::ProcessGroupXCCL::isInitialized,
py::call_guard<py::gil_scoped_release>());
intrusive_ptr_class_<::c10d::ProcessGroupXCCL::Options>(
processGroupXCCL, "Options", backendOptions)
.def(py::init<bool>(), py::arg("is_high_priority_stream") = false)
.def_readwrite("config", &::c10d::ProcessGroupXCCL::Options::config)
.def_readwrite(
"is_high_priority_stream",
&::c10d::ProcessGroupXCCL::Options::is_high_priority_stream)
.def_readwrite(
"split_from", &::c10d::ProcessGroupXCCL::Options::split_from)
.def_readwrite(
"split_color", &::c10d::ProcessGroupXCCL::Options::split_color)
.def_readwrite(
"global_ranks_in_group",
&::c10d::ProcessGroupXCCL::Options::global_ranks_in_group)
.def_readwrite(
"group_name", &::c10d::ProcessGroupXCCL::Options::group_name);
#endif

#ifdef USE_C10D_UCC
Expand Down
27 changes: 24 additions & 3 deletions torch/distributed/distributed_c10d.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2033,8 +2033,18 @@ def _new_process_group_helper(
elif backend_str == Backend.XCCL:
if not is_xccl_available():
raise RuntimeError("Distributed package doesn't have XCCL built in")
if backend_options is not None:
assert isinstance(backend_options, ProcessGroupXCCL.Options), (
"Expected backend_options argument to be of type ProcessGroupXCCL.Options"
)
else:
# default backend_options for XCCL
backend_options = ProcessGroupXCCL.Options()
backend_options.is_high_priority_stream = False
backend_options.global_ranks_in_group = global_ranks_in_group
backend_options.group_name = group_name
backend_class = ProcessGroupXCCL(
backend_prefix_store, group_rank, group_size
backend_prefix_store, group_rank, group_size, backend_options
)
backend_type = ProcessGroup.BackendType.XCCL
else:
Expand Down Expand Up @@ -5042,7 +5052,7 @@ def split_group(
)

parent_group_rank = parent_global_to_group_ranks[global_rank]
parent_backend = parent_pg._get_backend(torch.device("cuda"))
parent_backend = parent_pg._get_backend(device_id)

# if the parent backend does not support splitting, raise error
# currently this API only support NCCL backend
Expand Down Expand Up @@ -5123,6 +5133,15 @@ def split_group(
backend_class = ProcessGroupNCCL(
prefix_store, group_rank, len(my_group), pg_options
)
elif parent_backend_str == Backend.XCCL:
backend_type = ProcessGroup.BackendType.XCCL
if not isinstance(pg_options, ProcessGroupXCCL.Options):
raise RuntimeError(
"Expected pg_options argument to be of type ProcessGroupXCCL.Options"
)
backend_class = ProcessGroupXCCL(
prefix_store, group_rank, len(my_group), pg_options
)
else:
assert parent_backend_str.upper() in Backend._plugins, (
f"Unknown c10d backend type {parent_backend_str.upper()}"
Expand All @@ -5143,7 +5162,9 @@ def split_group(
pg._set_default_backend(backend_type)
backend_class._set_sequence_number_for_group()

pg._register_backend(torch.device("cuda"), backend_type, backend_class)
pg._register_backend(
torch.accelerator.current_accelerator(), backend_type, backend_class
)

# set group_name and group_desc to backend
assert group_name is not None
Expand Down