Skip to content

Try to propagate input data layout to outputs for FINN ops #229

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
jmitrevs merged 1 commit into from
Jan 29, 2026
Merged

Try to propagate input data layout to outputs for FINN ops #229

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
31 changes: 27 additions & 4 deletions src/qonnx/transformation/infer_data_layouts.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,33 @@ def _dims_to_layout(model, node, ndims):
return DataLayout.NC
else:
return DataLayout.UNKNOWN
else:
# Also try to propagate input data layout for "FINN ops" (if number of dims matches)
input_ndims = len(model.get_tensor_shape(node.input[0]))
if input_ndims == ndims and (layout := model.get_tensor_layout(node.input[0])):
# TODO: There are multi-input operations, why should the first
# determine the output layout?
return layout
else:
# Fallback: guess based on number of output dims
if ndims == 4:
return DataLayout.NHWC
elif ndims == 3:
return DataLayout.NWC
elif ndims == 2:
return DataLayout.NC
else:
return DataLayout.UNKNOWN
else:
# Check whether there is a layout annotation for the first input
# TODO: There are multi-input operations, why should the first
# determine the output layout?
# TODO: Shouldn't we at least check that the number of dims matches?
if layout := model.get_tensor_layout(node.input[0]):
# If annotation present: propagate input layout to output
# TODO: this won't work for concat, squeeze/unsqueeze/reshape...
return layout
# Fallback to the same defaults as for the FINN-Ops above
else:
if ndims == 4:
return DataLayout.NHWC
Expand All @@ -62,10 +89,6 @@ def _dims_to_layout(model, node, ndims):
return DataLayout.NC
else:
return DataLayout.UNKNOWN
else:
# propagate input layout to output
# TODO this won't work for concat, squeeze/unsqueeze/reshape...
return model.get_tensor_layout(node.input[0])


def _infer_node_data_layout(model, node):
Expand Down
Loading