Skip to content
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

Jump to bottom

[ONNX] Fix a bug with reshape imports when an initialized target shape is used more than once #7109

Merged
merged 2 commits into from
Dec 15, 2020
Merged

[ONNX] Fix a bug with reshape imports when an initialized target shape is used more than once #7109

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
293ff14
Fix a bug with reshape imports when an initialized target shape is us…
Dec 14, 2020
c3c5724
run autoformat
Dec 15, 2020
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
3 changes: 1 addition & 2 deletions python/tvm/relay/frontend/onnx.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -859,8 +859,7 @@ def _impl_v1(cls, inputs, attr, params):
@classmethod
def _impl_v5(cls, inputs, attr, params):
if get_name(inputs[1]) in params:
# pop shape out of parameters since it wont be needed later.
shape = tuple(params.pop(inputs[1].name_hint).asnumpy().astype("int32"))
shape = tuple(params[inputs[1].name_hint].asnumpy().astype("int32"))
out = _op.reshape(inputs[0], shape)
else:
out = _op.reshape(*inputs)
Expand Down
37 changes: 36 additions & 1 deletion tests/python/frontend/onnx/test_forward.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,7 +73,6 @@ def get_tvm_output(
input_names, shape_dict = get_input_data_shape_dict(graph_def, input_data)

mod, params = relay.frontend.from_onnx(graph_def, shape_dict, opset=opset)

with tvm.transform.PassContext(opt_level=1):
graph, lib, params = relay.build(mod, target, params=params)

Expand Down Expand Up @@ -234,6 +233,42 @@ def test_reshape():
tvm.testing.assert_allclose(ref_shape, tvm_out.shape)


@tvm.testing.uses_gpu
def test_double_reshape():
in_shape = (4, 3, 3, 4)
ref_shape = (6, 2, 4, 3)

ref_array = np.array(ref_shape)
ref_node = onnx.helper.make_node(
"Constant",
inputs=[],
outputs=["ref_in"],
value=onnx.helper.make_tensor(
name="const_tensor",
data_type=onnx.TensorProto.INT32,
dims=ref_array.shape,
vals=ref_array.flatten().astype(int),
),
)
reshape_node1 = helper.make_node("Reshape", ["in", "ref_in"], ["out1"])
reshape_node2 = helper.make_node("Reshape", ["in", "ref_in"], ["out2"])
add_node = helper.make_node("Add", ["out1", "out2"], ["out"])

graph = helper.make_graph(
[ref_node, reshape_node1, reshape_node2, add_node],
"reshape_test",
inputs=[helper.make_tensor_value_info("in", TensorProto.FLOAT, list(in_shape))],
outputs=[helper.make_tensor_value_info("out", TensorProto.FLOAT, list(ref_shape))],
)

model = helper.make_model(graph, producer_name="reshape_test")

for target, ctx in tvm.testing.enabled_targets():
x = np.random.uniform(size=in_shape).astype("int32")
tvm_out = get_tvm_output(model, x, target, ctx, ref_shape, "float32")
tvm.testing.assert_allclose(ref_shape, tvm_out.shape)


# TODO(mbrookhart): enable once VM supports heterogenous execution
# @tvm.testing.uses_gpu
def test_expand():
Expand Down