Skip to content

Support for zero-sized dimensions in aten.empty.memory_format #134

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 6 commits into
base: feature/backport_ea1_ops
Choose a base branch
from
Open

Support for zero-sized dimensions in aten.empty.memory_format #134

Changes from 4 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
c369bdd
Support for zero-sized dimensions in aten.empty.memory_format
TinaAMD Sep 4, 2023
b8c9731
Merge branch 'feature/backport_ea1_ops' into tina.FXML-2699-fix-empty…
TinaAMD Sep 6, 2023
0fe676d
Add match failure for 0-sized tensor dimensions
TinaAMD Sep 6, 2023
c59311e
Merge branch 'feature/backport_ea1_ops' into tina.FXML-2699-fix-empty…
mgehre-amd Sep 8, 2023
2656928
Add test case where some dimensions are zero
TinaAMD Sep 8, 2023
bf896be
Update fail sets to account for stable release
TinaAMD Sep 11, 2023
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
8 changes: 8 additions & 0 deletions lib/Conversion/TorchToTosa/TorchToTosa.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5392,6 +5392,14 @@ LogicalResult ConvertAtenOp<AtenEmptyMemoryFormatOp>::matchAndRewrite(
auto resultType =
typeConverter->convertType(op.getType()).template cast<RankedTensorType>();

// TOSA does not allow empty dimensions, so we can't lower this while
// preserving the shape.
if (llvm::any_of(resultType.getShape(),
[](int dimSize) { return dimSize == 0; })) {
return rewriter.notifyMatchFailure(
op, "Cannot lower tensors with 0-sized dimensions to TOSA.");
}

DenseElementsAttr emptyVal;
if (op.getDtype().getType().template isa<Torch::NoneType>()) {
emptyVal = DenseFPElementsAttr::get(resultType, {0.0F});
Expand Down