arm_compute v18.11
diff --git a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
index fe7bba3..8f87a7d 100644
--- a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
+++ b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp
@@ -45,8 +45,9 @@
namespace
{
-constexpr RelativeTolerance<float> tolerance_f32(0.01f); /**< Tolerance value for comparing reference's output against implementation's output for DataType::F32 */
-constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(1); /**< Tolerance value for comparing reference's output against implementation's output for DataType::QASYMM8 */
+RelativeTolerance<half_float::half> tolerance_f16(half_float::half(0.001)); /**< Tolerance value for comparing reference's output against implementation's output for DataType::F16 */
+constexpr RelativeTolerance<float> tolerance_f32(0.01f); /**< Tolerance value for comparing reference's output against implementation's output for DataType::F32 */
+constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(1); /**< Tolerance value for comparing reference's output against implementation's output for DataType::QASYMM8 */
const auto depth_multipliers = framework::dataset::make("DepthMultiplier", { 1, 2, 3 });
} // namespace
@@ -209,7 +210,7 @@
{
validate(Accessor(_target), _reference, tolerance_f32);
}
-TEST_SUITE_END()
+TEST_SUITE_END() // Generic
TEST_SUITE(W3x3)
template <typename T>
@@ -238,10 +239,43 @@
{
validate(Accessor(_target), _reference, tolerance_f32);
}
-TEST_SUITE_END()
-TEST_SUITE_END()
+TEST_SUITE_END() // W3x3
+TEST_SUITE_END() // F32
-TEST_SUITE_END()
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+TEST_SUITE(F16)
+TEST_SUITE(W3x3)
+template <typename T>
+using NEDepthwiseConvolutionLayerFixture3x3 = DepthwiseConvolutionLayerValidationFixture<Tensor, Accessor, NEDepthwiseConvolutionLayer3x3, T>;
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::ALL, combine(combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset3x3(),
+ depth_multipliers),
+ framework::dataset::make("DataType",
+ DataType::F16)),
+ framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+ validate(Accessor(_target), _reference, tolerance_f16);
+}
+FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(datasets::LargeDepthwiseConvolutionLayerDataset3x3(),
+ depth_multipliers),
+ framework::dataset::make("DataType",
+ DataType::F16)),
+ framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+ validate(Accessor(_target), _reference, tolerance_f16);
+}
+FIXTURE_DATA_TEST_CASE(RunOptimized, NEDepthwiseConvolutionLayerFixture3x3<half>, framework::DatasetMode::ALL, combine(combine(combine(datasets::OptimizedDepthwiseConvolutionLayerDataset3x3(),
+ framework::dataset::make("DepthMultiplier", 1)),
+ framework::dataset::make("DataType",
+ DataType::F16)),
+ framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+ validate(Accessor(_target), _reference, tolerance_f16);
+}
+TEST_SUITE_END() // W3x3
+TEST_SUITE_END() // FP16
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+TEST_SUITE_END() // Float
template <typename T>
using NEDepthwiseConvolutionLayerQuantizedFixture3x3 = DepthwiseConvolutionLayerValidationQuantizedFixture<Tensor, Accessor, NEDepthwiseConvolutionLayer3x3, T>;
@@ -270,6 +304,16 @@
{
validate(Accessor(_target), _reference, tolerance_qasymm8);
}
+FIXTURE_DATA_TEST_CASE(RunOptimized, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::ALL,
+ combine(combine(combine(combine(datasets::OptimizedDepthwiseConvolutionLayerDataset3x3(),
+ framework::dataset::make("DepthMultiplier", 1)),
+ framework::dataset::make("DataType",
+ DataType::QASYMM8)),
+ framework::dataset::make("QuantizationInfo", { QuantizationInfo(0.5f, 10) })),
+ framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })))
+{
+ validate(Accessor(_target), _reference, tolerance_qasymm8);
+}
FIXTURE_DATA_TEST_CASE(RunLarge, NEDepthwiseConvolutionLayerQuantizedFixture3x3<uint8_t>, framework::DatasetMode::NIGHTLY,
combine(combine(combine(combine(datasets::LargeDepthwiseConvolutionLayerDataset3x3(),
depth_multipliers),