Skip to content

Unconditionally return 1 for ParallelType::Stream for unshardedSizes #5706

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

Draft
zasdfgbnm wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Unconditionally return 1 for ParallelType::Stream for unshardedSizes #5706

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
24 changes: 1 addition & 23 deletions csrc/multidevice/execution_utils.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,29 +70,7 @@ std::vector<int64_t> unshardedSizes(

auto multiplier = [&]() -> int64_t {
if (parallel_type == ParallelType::Stream) {
// TODO(#5525): hack for MultiDeviceExecutor. MultiDeviceExecutor looks
// for ParallelType::Stream only in logical domains and assumes a
// stream-parallelized dimension is always fully allocated. So we set
// the multiplier to 1 when `sharded_id` is a logical IterDomain. This
// will have to change when FusionExecutorCache requires a logical
// dimension to be stream-parallelized, both loop and allocation. Refer
// to
// https://github.com/NVIDIA/Fuser/blob/f8e84e52296cdecd318dd2ce904139616d7bd434/tests/cpp/test_overlap.cpp#L155
// for an example. An alternative to consider is to create a new
// ParallelType for stream parallelization and use it in
// FusionExecutorCache.
if (std::find(
tv->getLogicalDomain().begin(),
tv->getLogicalDomain().end(),
sharded_id) != tv->getLogicalDomain().end()) {
return 1;
}

NVF_ERROR(
sharded_id->extent()->isConstInt(),
"DIDs/Stream extent is expected to be constant: ",
sharded_id);
return sharded_id->extent()->evaluate().as<int64_t>();
return 1;
}

if (isParallelTypeDeviceDim(parallel_type)) {
Expand Down