MLBEDSW-3858: Incorrect PAD usage bug
Fixed a bug where PAD having no consumers would result in a crash.
Now the constraint doesn't crash and thus the intended error message is shown, resulting in easier debugging.
Change-Id: I1e4403d47a6152e7adbf7bc065db86d4217d39cc
Signed-off-by: erik.andersson@arm.com <erik.andersson@arm.com>
diff --git a/ethosu/vela/supported_operators.py b/ethosu/vela/supported_operators.py
index 9d31518..1bebe9a 100644
--- a/ethosu/vela/supported_operators.py
+++ b/ethosu/vela/supported_operators.py
@@ -829,17 +829,15 @@
def constraint_pad_ofm(cls, op):
"Must be followed by one of the following operator types: {}"
consumers = op.ofm.consumers()
- consumers_to_pad = 0
- for consumer in consumers:
- if consumer.type in cls.supported_pad_consumers:
- if consumer.attrs["padding"] == Padding.VALID:
- consumers_to_pad += 1
- valid = len(consumers) > 0 and len(consumers) == consumers_to_pad
- return (
- valid,
- f"Operator is followed by {consumers_to_pad} consumers with "
- f"padding set to VALID, out of {len(consumers)} consumers",
- )
+ unsupported_consumers = [
+ cons.type
+ for cons in consumers
+ if cons is not None
+ if cons.type not in cls.supported_pad_consumers or cons.attrs["padding"] != Padding.VALID
+ ] + [None for cons in consumers if cons is None]
+ none_string = ", ".join(["NoneType" for cons in consumers if cons is None])
+ valid = len(unsupported_consumers) == 0
+ return valid, f"PAD operator is followed by: {_optype_formatter(unsupported_consumers)+none_string}"
@staticmethod
def constraint_stridedslice_inputs_const(op):
diff --git a/ethosu/vela/test/test_supported_operators.py b/ethosu/vela/test/test_supported_operators.py
index d8fbb98..36213b7 100644
--- a/ethosu/vela/test/test_supported_operators.py
+++ b/ethosu/vela/test/test_supported_operators.py
@@ -518,7 +518,13 @@
def create_pad_op(
- in_shape, out_shape, padding, in_dtype=DataType.int8, out_dtype=DataType.int8, pad_dtype=DataType.int32
+ in_shape,
+ out_shape,
+ padding,
+ in_dtype=DataType.int8,
+ out_dtype=DataType.int8,
+ pad_dtype=DataType.int32,
+ pad_setting=Padding.VALID,
):
qp = testutil.default_quant_params()
in0 = Tensor(in_shape, in_dtype, "in")
@@ -527,19 +533,17 @@
out = Tensor(out_shape, out_dtype, "out")
out.quantization = qp.clone()
op = testutil.create_op(Op.Pad, [in0, pad_tensor], out)
-
conv_out_tens = Tensor(in_shape, in_dtype, "output")
conv_out_tens.quantization = qp.clone()
weight_tens = Tensor(in_shape, in_dtype, "weights")
weight_tens.values = np.zeros(weight_tens.shape)
weight_tens.quant_values = np.zeros(weight_tens.shape, np.int8)
weight_tens.quantization = qp.clone()
- bias_tens = Tensor([in_shape[-1]], pad_dtype, "biases")
- attrs = {"padding": Padding.VALID, "stride_w": 2, "stride_h": 2, "dilation_w_factor": 1, "dilation_h_factor": 1}
+ bias_tens = Tensor(out_shape, pad_dtype, "biases")
+ attrs = {"padding": pad_setting, "stride_w": 2, "stride_h": 2, "dilation_w_factor": 1, "dilation_h_factor": 1}
attrs["strides"] = (1, attrs["stride_h"], attrs["stride_w"], 1)
- conv2d_op = testutil.create_op(Op.Conv2D, [out, weight_tens, bias_tens], conv_out_tens, attrs)
+ conv2d_op = testutil.create_op(Op.Conv2DBias, [out, weight_tens, bias_tens], conv_out_tens, attrs)
conv2d_op.add_input_tensor(out)
- conv2d_op.set_ifm_ofm_shapes()
return op
@@ -581,11 +585,16 @@
def test_constraint_pad_consumer():
# PAD operator must be followed by a valid consumer with Padding.VALID attribute
- op = create_pad_op(in_shape=[1, 1, 1, 1], out_shape=[1, 3, 3, 1], padding=[[0, 0], [1, 1], [1, 1], [0, 0], [0, 0]],)
- conv_op = op.ofm.consumers()[0]
- conv_op.attrs["Padding"] = Padding.SAME
+ op = create_pad_op(in_shape=[1, 1, 1, 1], out_shape=[1, 3, 3, 1], padding=[[0, 0], [1, 1], [1, 1], [0, 0]],)
+ assert support.is_operator_supported(op)
+ op = create_pad_op(
+ in_shape=[1, 1, 1, 1],
+ out_shape=[1, 3, 3, 1],
+ padding=[[0, 0], [1, 1], [1, 1], [0, 0]],
+ pad_setting=Padding.SAME,
+ )
assert not support.is_operator_supported(op)
- op_consumer = testutil.create_op_with_quant_tensors(Op.Concat, [1, 1, 1, 4], [1, 1, 1, 8])
+ op_consumer = testutil.create_op_with_quant_tensors(Op.ConcatTFLite, [1, 1, 1, 4], [1, 1, 1, 8])
op.ofm.consumer_list = [op_consumer]
assert not support.is_operator_supported(op)
op_consumer = testutil.create_op_with_quant_tensors(Op.AvgPool, [1, 8, 8, 8], [1, 8, 8, 8])