Skip to content

support premul sum #23

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 1 commit into
base: master
Choose a base branch
from
Open

support premul sum #23

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
19 changes: 16 additions & 3 deletions torch/csrc/distributed/c10d/init.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -882,7 +882,7 @@ This class does not support ``__members__`` property.)");
}
TORCH_CHECK(r.supplement_.defined(), "Invalid PREMUL_SUM ReduceOp");
const auto* preMulSupplement =
reinterpret_cast<::c10d::NCCLPreMulSumSupplement*>(
reinterpret_cast<::c10d::XCCLPreMulSumSupplement*>(
r.supplement_.get());
if (!preMulSupplement->tensor_factor.defined()) {
return py::make_tuple(r.op_, preMulSupplement->double_factor);
Expand All @@ -900,9 +900,9 @@ This class does not support ``__members__`` property.)");
}
const auto preMulSupplement_factor = t[1];
if (py::isinstance<py::float_>(preMulSupplement_factor)) {
return ::c10d::makeNCCLPreMulSum(t[1].cast<double>());
return ::c10d::makeXCCLPreMulSum(t[1].cast<double>());
} else {
return ::c10d::makeNCCLPreMulSum(t[1].cast<at::Tensor>());
return ::c10d::makeXCCLPreMulSum(t[1].cast<at::Tensor>());
}
}));

Expand Down Expand Up @@ -938,6 +938,19 @@ This class does not support ``__members__`` property.)");
py::arg("factor").noconvert(),
py::return_value_policy::copy, // seems safest
py::call_guard<py::gil_scoped_release>());
module
.def(
"_make_xccl_premul_sum",
&::c10d::makeXCCLPreMulSum<double>,
py::arg("factor").noconvert(),
py::return_value_policy::copy, // seems safest
py::call_guard<py::gil_scoped_release>())
.def(
"_make_xccl_premul_sum",
&::c10d::makeXCCLPreMulSum<at::Tensor>,
py::arg("factor").noconvert(),
py::return_value_policy::copy, // seems safest
py::call_guard<py::gil_scoped_release>());

module.def(
"_set_thread_isolation_mode",
Expand Down
1 change: 1 addition & 0 deletions torch/distributed/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@ def is_available() -> bool:
_ControlCollectives,
_DEFAULT_FIRST_BUCKET_BYTES,
_make_nccl_premul_sum,
_make_xccl_premul_sum,
_register_builtin_comm_hook,
_register_comm_hook,
_StoreCollectives,
Expand Down