blob: b45183029c41bbcca394eac7b98318733d6a6bce [file] [log] [blame]
Fredrik Svedberg6ef02302022-09-29 13:29:37 +02001# Copyright (C) 2022 Arm Limited or its affiliates. All rights reserved.
2#
3# SPDX-License-Identifier: Apache-2.0
4#
5# Licensed under the Apache License, Version 2.0 (the License); you may
6# not use this file except in compliance with the License.
7# You may obtain a copy of the License at
8#
9# www.apache.org/licenses/LICENSE-2.0
10#
11# Unless required by applicable law or agreed to in writing, software
12# distributed under the License is distributed on an AS IS BASIS, WITHOUT
13# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14# See the License for the specific language governing permissions and
15# limitations under the License.
16# Description:
17# Unit tests for scaling
18from ethosu.vela.scaling import quantise_scale
19from ethosu.vela.scaling import reduced_quantise_scale
20
21
22def test_scaling():
23 multiplier, shift = quantise_scale(1)
24 assert multiplier == 1073741824 and shift == 30
25 multiplier, shift = quantise_scale(0.5)
26 assert multiplier == 1073741824 and shift == 31
27 multiplier, shift = quantise_scale(0.001)
28 assert multiplier == 1099511628 and shift == 40
29 multiplier, shift = quantise_scale(0.008)
30 assert multiplier == 1099511628 and shift == 37
31 multiplier, shift = quantise_scale(0.00097652)
32 assert multiplier == 2147390190 and shift == 41
33 multiplier, shift = quantise_scale(0.0009765615959827986)
34 assert multiplier == 2147481660 and shift == 41
35
36
37def test_reduced_scaling():
38 multiplier, shift = reduced_quantise_scale(1)
39 assert multiplier == 16384 and shift == 14
40 multiplier, shift = reduced_quantise_scale(0.5)
41 assert multiplier == 16384 and shift == 15
42 multiplier, shift = reduced_quantise_scale(0.001)
43 assert multiplier == 16777 and shift == 24
44 multiplier, shift = reduced_quantise_scale(0.008)
45 assert multiplier == 16777 and shift == 21
46 multiplier, shift = reduced_quantise_scale(0.00097652)
47 assert multiplier == 32767 and shift == 25
48 # multiplier saturated
49 multiplier, shift = reduced_quantise_scale(0.0009765615959827986)
50 assert multiplier == 32767 and shift == 25