diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-01-13 14:42:54 +0000 |
|---|---|---|
| committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-01-13 15:58:08 +0000 |
| commit | d30405ac6eb38205676dfaa6e875b264caef431d (patch) | |
| tree | 7d2a92ae3f38346d991b8b04f0e3b912b3b50022 | |
| parent | 3d3a01c2122278febd6e77ec255d2557d0005546 (diff) | |
| download | ComputeLibrary-d30405ac6eb38205676dfaa6e875b264caef431d.tar.gz | |
Alter tolerance in QuantizationLayer to use int8_t for asymmetric signed
Tolerance values passed to the validation process of QuantizationLayer
for int8_t quantized types were in unsigned format leading in wrong
interpretation of negative values. Thus, alter the tolerance type to
use int8_t instead.
Resolves: COMPMID-4090
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I82388a6e59cc91ca5093f0c665267b620523b122
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4848
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
| -rw-r--r-- | tests/validation/NEON/QuantizationLayer.cpp | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/tests/validation/NEON/QuantizationLayer.cpp b/tests/validation/NEON/QuantizationLayer.cpp index 04b3a78972..36f2937cad 100644 --- a/tests/validation/NEON/QuantizationLayer.cpp +++ b/tests/validation/NEON/QuantizationLayer.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2020 Arm Limited. + * Copyright (c) 2017-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -97,7 +97,7 @@ FIXTURE_DATA_TEST_CASE(RunSmallQASYMM8Signed, NEQuantizationLayerQASYMM8SignedFi framework::dataset::make("QuantizationInfo", { QuantizationInfo(0.5f, 10) }))) { // Validate output - validate(Accessor(_target), _reference, tolerance_u8); + validate(Accessor(_target), _reference, tolerance_s8); } FIXTURE_DATA_TEST_CASE(RunSmallQASYMM16, NEQuantizationLayerQASYMM16Fixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(QuantizationSmallShapes, framework::dataset::make("DataType", DataType::F32)), @@ -140,7 +140,7 @@ FIXTURE_DATA_TEST_CASE(RunSmallQASYMM8Signed, NEQuantizationLayerQASYMM8SignedFi framework::dataset::make("QuantizationInfo", { QuantizationInfo(0.5f, 10) }))) { // Validate output - validate(Accessor(_target), _reference, tolerance_u8); + validate(Accessor(_target), _reference, tolerance_s8); } FIXTURE_DATA_TEST_CASE(RunSmallQASYMM16, NEQuantizationLayerQASYMM16Fixture<half>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(QuantizationSmallShapes, framework::dataset::make("DataType", DataType::F16)), |