Skip to content

Continue memory planning when unknown shape tensor is encountered. #6413

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

Merged
codemzs merged 3 commits into from
Jan 22, 2021
Merged

Continue memory planning when unknown shape tensor is encountered. #6413

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
cfb5fb7
Continue memory planning when unknown shape tensor is encountered.
Jan 22, 2021
377cb95
comment.
Jan 22, 2021
5386c95
PR feedback.
Jan 22, 2021
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
16 changes: 11 additions & 5 deletions onnxruntime/core/framework/session_state.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -389,6 +389,7 @@ void TryCalculateSizeFromResolvedShape(int ml_value_idx, std::unordered_map<int,

} // namespace

// If this function fails NO memory planning will take place, hence lets ONLY FAIL and stop training where warranted, example SIZE overflow.
Status SessionState::GeneratePatternGroupCache(const std::vector<std::reference_wrapper<const TensorShape>>& input_shape,
const std::vector<int>& feed_mlvalue_idxs,
MemoryPatternGroup* output,
Expand Down Expand Up @@ -425,12 +426,17 @@ Status SessionState::GeneratePatternGroupCache(const std::vector<std::reference_
auto* arg = node->OutputDefs()[i];
size_t is_resolved = 0;
std::vector<int64_t> resolved_shape;
ORT_RETURN_IF_ERROR(TryResolveShape(arg, map, is_resolved, resolved_shape));

// Store all valid resolved shapes. They will be queried in, for example,
// Recv operator to bypass the dependency of output shapes on inputs.
if (is_resolved != 0) {
resolved_shapes[ml_value_idx] = resolved_shape;
// Tensors whose shape cannot be resolved statically will be allocated at runtime.
if (TryResolveShape(arg, map, is_resolved, resolved_shape).IsOK()) {
// Store all valid resolved shapes. They will be queried in, for example,
// Recv operator to bypass the dependency of output shapes on inputs.
if (is_resolved != 0) {
resolved_shapes[ml_value_idx] = resolved_shape;
}
} else {
LOGS(logger_, INFO) << "[Static memory planning] Could not resolve shape for tensor with ML index "
<< ml_value_idx << ", will allocate dynamically.";
}
}
}
Expand Down