diff options
| author | Tim Hall <tim.hall@arm.com> | 2023-01-13 17:57:25 +0000 |
|---|---|---|
| committer | tim.hall <tim.hall@arm.com> | 2023-01-20 14:07:21 +0000 |
| commit | 3b1578e44b4c6a8c8c9a8e0891d3866a89bd66af (patch) | |
| tree | 491c337bc854d435b80f0a535496084ea9ebc9ac /ethosu/vela/data_type.py | |
| parent | f34904717f643499f3ea6210322bbe1b635db088 (diff) | |
| download | ethos-u-vela-3b1578e44b4c6a8c8c9a8e0891d3866a89bd66af.tar.gz | |
MLBEDSW-7151: MLCE: Difference in model output between x86 & aarch64
- The issue is due to undefined behaviour when casting a NumPy float
to a NumPy unsigned integer which occurs in create_const_tensor()
- The fix is to make sure that the values are first cast to a Python
float
- In addition, the values datatype argument has been removed from
create_const_tensor() to stop the tensor and values datatypes getting
out of sync
Change-Id: I134b9be8c941b361929a5ae7db8cb35f2e9728f2
Signed-off-by: Tim Hall <tim.hall@arm.com>
Diffstat (limited to 'ethosu/vela/data_type.py')
| -rw-r--r-- | ethosu/vela/data_type.py | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/ethosu/vela/data_type.py b/ethosu/vela/data_type.py index 829cef38..5d0320b2 100644 --- a/ethosu/vela/data_type.py +++ b/ethosu/vela/data_type.py @@ -1,4 +1,4 @@ -# SPDX-FileCopyrightText: Copyright 2020-2021 Arm Limited and/or its affiliates <open-source-office@arm.com> +# SPDX-FileCopyrightText: Copyright 2020-2021, 2023 Arm Limited and/or its affiliates <open-source-office@arm.com> # # SPDX-License-Identifier: Apache-2.0 # @@ -110,7 +110,7 @@ class DataType: BaseType.Complex: "c", } assert self.type in numpy_dtype_code, f"Failed to interpret {self} as a numpy dtype" - return np.dtype(numpy_dtype_code[self.type] + str(self.size_in_bytes())) + return np.dtype(numpy_dtype_code[self.type] + str(self.size_in_bytes())).type stem_name = { BaseType.UnsignedInt: ("uint%s", True), |