Skip to content

[13/N] Update gather with CPU/CUDA implementations #86409

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
H-Huang wants to merge 4 commits into from
Closed

[13/N] Update gather with CPU/CUDA implementations #86409

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
eb5e95a
[13/N] Update gather with CPU/CUDA implementations
H-Huang Oct 6, 2022
c228bb5
Update on "[13/N] Update gather with CPU/CUDA implementations"
H-Huang Oct 7, 2022
3f83c9f
Update on "[13/N] Update gather with CPU/CUDA implementations"
H-Huang Oct 31, 2022
64f9d46
Update on "[13/N] Update gather with CPU/CUDA implementations"
H-Huang Nov 9, 2022
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
2 changes: 1 addition & 1 deletion test/distributed/test_c10d_common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1466,7 +1466,7 @@ def _call_collective_with_varying_tensors(self, backend, collective, *args):
# multi tensor collectives
if collective == dist.barrier:
collective()
elif collective == dist.all_gather:
elif collective in (dist.all_gather, dist.gather):
collective([tensor], tensor, *args)
elif collective == dist.scatter:
collective(tensor, [tensor], *args)
Expand Down
32 changes: 32 additions & 0 deletions torch/csrc/distributed/c10d/OpsImpl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,6 +219,30 @@ reduce_scatter_cuda_(
output_tensors, work);
}

c10::intrusive_ptr<Work> gather_cpu_(
const std::vector<std::vector<at::Tensor>>& output_tensors,
const std::vector<at::Tensor>& input_tensors,
const c10::intrusive_ptr<ProcessGroup>& process_group,
int64_t root_rank,
int64_t timeout) {
return process_group->gather(
const_cast<std::vector<std::vector<at::Tensor>>&>(output_tensors),
const_cast<std::vector<at::Tensor>&>(input_tensors),
GatherOptions{root_rank, std::chrono::milliseconds(timeout)});
}

c10::intrusive_ptr<Work> gather_cuda_(
const std::vector<std::vector<at::Tensor>>& output_tensors,
const std::vector<at::Tensor>& input_tensors,
const c10::intrusive_ptr<ProcessGroup>& process_group,
int64_t root_rank,
int64_t timeout) {
return process_group->gather(
const_cast<std::vector<std::vector<at::Tensor>>&>(output_tensors),
const_cast<std::vector<at::Tensor>&>(input_tensors),
GatherOptions{root_rank, std::chrono::milliseconds(timeout)});
}

std::tuple<std::vector<at::Tensor>, c10::intrusive_ptr<Work>> scatter_cpu_(
const std::vector<at::Tensor>& output_tensors,
const std::vector<std::vector<at::Tensor>>& input_tensors,
Expand Down Expand Up @@ -359,6 +383,14 @@ TORCH_LIBRARY_IMPL(c10d, CUDA, m) {
m.impl("reduce_scatter_", reduce_scatter_cuda_);
}

TORCH_LIBRARY_IMPL(c10d, CPU, m) {
m.impl("gather_", gather_cpu_);
}

TORCH_LIBRARY_IMPL(c10d, CUDA, m) {
m.impl("gather_", gather_cuda_);
}

TORCH_LIBRARY_IMPL(c10d, CPU, m) {
m.impl("scatter_", scatter_cpu_);
}
Expand Down