aboutsummaryrefslogtreecommitdiff
path: root/tests/validation
diff options
context:
space:
mode:
authorGian Marco Iodice <gianmarco.iodice@arm.com>2021-04-16 15:08:59 +0100
committerGian Marco Iodice <gianmarco.iodice@arm.com>2021-07-02 15:56:45 +0000
commit8155c0253c00aa9e26651361460c66feb39829a6 (patch)
tree41dacc432d4d1f1daa32d20d15e5120c11b9fa56 /tests/validation
parent2eb5d16b839cbc28c6cb7f0de7a0bf15290b425a (diff)
downloadComputeLibrary-8155c0253c00aa9e26651361460c66feb39829a6.tar.gz
Rework OpenCL Depthwise Convolution
- Remove dedicated kernels for NCHW. Now we only use NHWC with permute - Remove specialized kernels for 3x3 NHWC - Simplify CLDepthwiseConvolutionLayer.cpp to call just the native implementation for both floating-point and quantized data types - Develop two parametric opencl kernels for depthwise convolution layer NHWC (floating-point and quantized) - Add support to export the weights to cl_image - Extend test for depthwise convolution on opencl Resolves COMPMID-4417 Change-Id: Ibe533f79c2860f9cac8e921895d5a8f947753a5c Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5893 Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'tests/validation')
-rw-r--r--tests/validation/CL/DepthwiseConvolutionLayer.cpp7
-rw-r--r--tests/validation/CL/DepthwiseConvolutionLayerNative.cpp179
-rw-r--r--tests/validation/fixtures/DepthwiseConvolutionLayerFixture.h63
3 files changed, 207 insertions, 42 deletions
diff --git a/tests/validation/CL/DepthwiseConvolutionLayer.cpp b/tests/validation/CL/DepthwiseConvolutionLayer.cpp
index 22922f41a2..79a2678b44 100644
--- a/tests/validation/CL/DepthwiseConvolutionLayer.cpp
+++ b/tests/validation/CL/DepthwiseConvolutionLayer.cpp
@@ -48,14 +48,13 @@ constexpr RelativeTolerance<float> tolerance_f32(0.01f); /**<
constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(0); /**< Tolerance value for comparing reference's output against implementation's output for DataType::QASYMM8 */
constexpr float tolerance_num = 0.05f; /**< Tolerance number */
-const auto depth_multipliers = framework::dataset::make("DepthMultiplier", { 1, 2, 5 });
+const auto depth_multipliers = framework::dataset::make("DepthMultiplier", { 1, 5 });
const auto large_depth_multipliers = framework::dataset::make("DepthMultiplier", { 1, 2, 5, 8 });
//Activation Functions
const auto ActivationFunctionsDataset = framework::dataset::make("ActivationInfo",
{
ActivationLayerInfo(),
- ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.f, 0.f)
});
} // namespace
@@ -481,8 +480,8 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerQuantizedFixture
combine(combine(combine(combine(combine(combine(datasets::SmallDepthwiseConvolutionLayerDataset(),
depth_multipliers),
framework::dataset::make("DataType", DataType::QASYMM8)),
- framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.3f, 10), QuantizationInfo(2.2f, 10) })),
- framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(0.5f, 4) })),
+ framework::dataset::make("SrcQuantizationInfo", { QuantizationInfo(0.5f, 128), QuantizationInfo(2.2f, 10) })),
+ framework::dataset::make("DstQuantizationInfo", { QuantizationInfo(1.f, 128) })),
framework::dataset::make("DataLayout", { DataLayout::NHWC })), // NCHW is tested with int8
ActivationFunctionsDataset))
{
diff --git a/tests/validation/CL/DepthwiseConvolutionLayerNative.cpp b/tests/validation/CL/DepthwiseConvolutionLayerNative.cpp
index f640ee2b18..f565255719 100644
--- a/tests/validation/CL/DepthwiseConvolutionLayerNative.cpp
+++ b/tests/validation/CL/DepthwiseConvolutionLayerNative.cpp
@@ -62,7 +62,7 @@ RelativeTolerance<half_float::half> rel_tolerance_f16(half_float::half(0.01f));
constexpr float abs_tolerance_f16(0.03f);
/** Width values to test - Precommit */
-const auto width_values_precommit = framework::dataset::make("width", { 1U, 17U, 32U } );
+const auto width_values_precommit = framework::dataset::make("width", { 1U, 33U } );
/** Width values to test - Nightly */
const auto width_values_nightly = framework::dataset::make("width", { 53U, 47U } );
@@ -79,6 +79,12 @@ const auto channel_values_precommit = framework::dataset::make("channels", { 15U
/** Channel values to test - Nightly */
const auto channel_values_nightly = framework::dataset::make("channels", { 33U, 19U });
+/** Channel values to test with cl_image support - Precommit */
+const auto channel_values_export_to_cl_image_precommit = framework::dataset::make("channels", { 16U });
+
+/** Channel values to test with cl_image support - Nightly */
+const auto channel_values_export_to_cl_image_nightly = framework::dataset::make("channels", { 32U });
+
/** Batch values to test - Precommit */
const auto batch_values_precommit = framework::dataset::make("batch", { 1U, 2U });
@@ -115,11 +121,17 @@ const auto n0_values_precommit = framework::dataset::make("N0", {2, 4});
/** N0 values to test - Nightly */
const auto n0_values_nightly = framework::dataset::make("N0", {3, 8});
+/** N0 values to test with cl_image support - Precommit */
+const auto n0_values_export_to_cl_image_precommit = framework::dataset::make("N0", {4});
+
+/** N0 values to test with cl_image support - Nightly */
+const auto n0_values_export_to_cl_image_nightly = framework::dataset::make("N0", {8});
+
/** Activation values to test */
const auto act_values = framework::dataset::make("Activation",
{
ActivationLayerInfo(),
- ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 8.f, 2.f),
+ ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 6.0f, 0.5f),
});
} // namespace
@@ -129,7 +141,7 @@ TEST_SUITE(DepthwiseConvolutionLayerNative)
TEST_SUITE(Float)
TEST_SUITE(FP32)
FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::ALL,
- combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
width_values_precommit,
height_values_precommit),
channel_values_precommit),
@@ -142,14 +154,15 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<fl
framework::dataset::make("DataType", DataType::F32)),
data_layout_values),
act_values),
- n0_values_precommit))
+ n0_values_precommit),
+ framework::dataset::make("ExportToCLImage", false)))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
}
FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::NIGHTLY,
- combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
width_values_nightly,
height_values_nightly),
channel_values_nightly),
@@ -162,16 +175,79 @@ FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<fl
framework::dataset::make("DataType", DataType::F32)),
data_layout_values),
act_values),
- n0_values_nightly))
+ n0_values_nightly),
+ framework::dataset::make("ExportToCLImage", false)))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
}
+
+TEST_SUITE(ExportWeightsToCLImage)
+FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::ALL,
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ width_values_precommit,
+ height_values_precommit),
+ channel_values_export_to_cl_image_precommit),
+ batch_values_precommit),
+ kernel_sz_values_precommit),
+ framework::dataset::make("depth_multiplier", 1)),
+ dilation_values),
+ stride_values),
+ padding_valid_values),
+ framework::dataset::make("DataType", DataType::F32)),
+ data_layout_values),
+ act_values),
+ n0_values_export_to_cl_image_precommit),
+ framework::dataset::make("ExportToCLImage", true)))
+{
+ // Validate output
+ if(_validate_output)
+ {
+ // Validate output
+ validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
+ }
+ else
+ {
+ ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped");
+ framework::ARM_COMPUTE_PRINT_INFO();
+ }
+}
+
+FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::NIGHTLY,
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ width_values_nightly,
+ height_values_nightly),
+ channel_values_export_to_cl_image_nightly),
+ batch_values_nightly),
+ kernel_sz_values_nightly),
+ framework::dataset::make("depth_multiplier", 1)),
+ dilation_values),
+ stride_values),
+ padding_valid_values),
+ framework::dataset::make("DataType", DataType::F32)),
+ data_layout_values),
+ act_values),
+ n0_values_export_to_cl_image_nightly),
+ framework::dataset::make("ExportToCLImage", true)))
+{
+ // Validate output
+ if(_validate_output)
+ {
+ // Validate output
+ validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
+ }
+ else
+ {
+ ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped");
+ framework::ARM_COMPUTE_PRINT_INFO();
+ }
+}
+TEST_SUITE_END() // ExportWeightsToCLImage
TEST_SUITE_END() // FP32
TEST_SUITE(FP16)
FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::ALL,
- combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
width_values_precommit,
height_values_precommit),
channel_values_precommit),
@@ -184,14 +260,15 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<ha
framework::dataset::make("DataType", DataType::F16)),
data_layout_values),
act_values),
- n0_values_precommit))
+ n0_values_precommit),
+ framework::dataset::make("ExportToCLImage", false)))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f16);
}
FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::NIGHTLY,
- combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
width_values_nightly,
height_values_nightly),
channel_values_nightly),
@@ -204,18 +281,80 @@ FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<ha
framework::dataset::make("DataType", DataType::F16)),
data_layout_values),
act_values),
- n0_values_nightly))
+ n0_values_nightly),
+ framework::dataset::make("ExportToCLImage", false)))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16);
}
+TEST_SUITE(ExportWeightsToCLImage)
+FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::ALL,
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ width_values_precommit,
+ height_values_precommit),
+ channel_values_export_to_cl_image_precommit),
+ batch_values_precommit),
+ kernel_sz_values_precommit),
+ framework::dataset::make("depth_multiplier", 1)),
+ dilation_values),
+ stride_values),
+ padding_valid_values),
+ framework::dataset::make("DataType", DataType::F16)),
+ data_layout_values),
+ act_values),
+ n0_values_export_to_cl_image_precommit),
+ framework::dataset::make("ExportToCLImage", true)))
+{
+ // Validate output
+ if(_validate_output)
+ {
+ // Validate output
+ validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16);
+ }
+ else
+ {
+ ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped");
+ framework::ARM_COMPUTE_PRINT_INFO();
+ }
+}
+
+FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::NIGHTLY,
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ width_values_nightly,
+ height_values_nightly),
+ channel_values_export_to_cl_image_nightly),
+ batch_values_nightly),
+ kernel_sz_values_nightly),
+ framework::dataset::make("depth_multiplier", 1)),
+ dilation_values),
+ stride_values),
+ padding_valid_values),
+ framework::dataset::make("DataType", DataType::F16)),
+ data_layout_values),
+ act_values),
+ n0_values_export_to_cl_image_nightly),
+ framework::dataset::make("ExportToCLImage", true)))
+{
+ // Validate output
+ if(_validate_output)
+ {
+ // Validate output
+ validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16);
+ }
+ else
+ {
+ ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped");
+ framework::ARM_COMPUTE_PRINT_INFO();
+ }
+}
+TEST_SUITE_END() // ExportWeightsToCLImage
TEST_SUITE_END() // FP16
TEST_SUITE_END() // Float
TEST_SUITE(DepthMultiplier)
TEST_SUITE(Float)
TEST_SUITE(FP32)
FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::ALL,
- combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
width_values_precommit,
height_values_precommit),
channel_values_precommit),
@@ -228,14 +367,15 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<fl
framework::dataset::make("DataType", DataType::F32)),
data_layout_values),
act_values),
- framework::dataset::make("N0", 1)))
+ framework::dataset::make("N0", 1)),
+ framework::dataset::make("ExportToCLImage", false)))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
}
FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::NIGHTLY,
- combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
width_values_nightly,
height_values_nightly),
channel_values_nightly),
@@ -248,7 +388,8 @@ FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<fl
framework::dataset::make("DataType", DataType::F32)),
data_layout_values),
act_values),
- framework::dataset::make("N0", 1)))
+ framework::dataset::make("N0", 1)),
+ framework::dataset::make("ExportToCLImage", false)))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
@@ -257,7 +398,7 @@ TEST_SUITE_END() // FP32
TEST_SUITE(FP16)
FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::ALL,
- combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
width_values_precommit,
height_values_precommit),
channel_values_precommit),
@@ -270,14 +411,15 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<ha
framework::dataset::make("DataType", DataType::F16)),
data_layout_values),
act_values),
- framework::dataset::make("N0", 1)))
+ framework::dataset::make("N0", 1)),
+ framework::dataset::make("ExportToCLImage", false)))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f16);
}
FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::NIGHTLY,
- combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+ combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
width_values_nightly,
height_values_nightly),
channel_values_nightly),
@@ -290,7 +432,8 @@ FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<ha
framework::dataset::make("DataType", DataType::F16)),
data_layout_values),
act_values),
- framework::dataset::make("N0", 1)))
+ framework::dataset::make("N0", 1)),
+ framework::dataset::make("ExportToCLImage", false)))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16);
diff --git a/tests/validation/fixtures/DepthwiseConvolutionLayerFixture.h b/tests/validation/fixtures/DepthwiseConvolutionLayerFixture.h
index c255cc5c13..0e02ae28ca 100644
--- a/tests/validation/fixtures/DepthwiseConvolutionLayerFixture.h
+++ b/tests/validation/fixtures/DepthwiseConvolutionLayerFixture.h
@@ -198,7 +198,7 @@ protected:
{
case DataType::QASYMM8:
{
- std::uniform_int_distribution<uint8_t> distribution(0, 10);
+ std::uniform_int_distribution<uint8_t> distribution(0, 15);
library->fill(tensor, distribution, i);
break;
}
@@ -292,7 +292,7 @@ public:
if(padding_valid)
{
- _conv_info = PadStrideInfo();
+ _conv_info = PadStrideInfo(stride.width, stride.height);
}
else
{
@@ -333,6 +333,7 @@ public:
void allocate_and_run_target()
{
add_padding_x({ &_src, &_weights, &_biases, &_target }, _data_layout);
+ add_padding_y({ &_src, &_target }, _data_layout);
// Allocate tensors
_src.allocator()->allocate();
@@ -416,15 +417,16 @@ class DepthwiseConvolutionLayerNativeConfigurableValidationFixture : public Dept
public:
template <typename...>
void setup(size_t width, size_t height, size_t channel, size_t batch, Size2D kernel_size, size_t depth_multiplier, Size2D dilation, Size2D stride, bool padding_valid, DataType data_type,
- DataLayout data_layout, const ActivationLayerInfo &act_info, unsigned int n0)
+ DataLayout data_layout, const ActivationLayerInfo &act_info, unsigned int n0, bool export_to_cl_image)
{
- _dilation = dilation;
- _depth_multiplier = depth_multiplier;
- _data_type = data_type;
- _data_layout = data_layout;
- _act_info = act_info;
- _n0 = n0;
- _in_place = in_place;
+ _dilation = dilation;
+ _depth_multiplier = depth_multiplier;
+ _data_type = data_type;
+ _data_layout = data_layout;
+ _act_info = act_info;
+ _n0 = n0;
+ _export_to_cl_image = export_to_cl_image;
+ _in_place = in_place;
_input_shape = TensorShape(width, height, channel, batch);
_weights_shape = TensorShape(kernel_size.width, kernel_size.height, channel * _depth_multiplier);
@@ -432,11 +434,11 @@ public:
if(padding_valid)
{
- _conv_info = PadStrideInfo();
+ _conv_info = calculate_same_pad(_input_shape, _weights_shape, PadStrideInfo(stride.width, stride.height), DataLayout::NCHW, _dilation);
}
else
{
- _conv_info = calculate_same_pad(_input_shape, _weights_shape, PadStrideInfo(stride.width, stride.height), DataLayout::NCHW, _dilation);
+ _conv_info = PadStrideInfo(stride.width, stride.height);
}
}
@@ -462,14 +464,26 @@ public:
target_to_use = &_target;
}
- DWCWeightsKernelInfo dwc_weights_info;
- dwc_weights_info.n0 = _n0;
+ DWCComputeKernelInfo dwc_info;
+ dwc_info.n0 = _n0;
+ dwc_info.m0 = _conv_info.stride().first == 1 && _dilation.x() == 1 ? 8 : 1;
+ dwc_info.export_weights_to_cl_image = _export_to_cl_image;
- DWCKernelInfo dwc_info;
- dwc_info.activation_info = _act_info;
+#if defined(ARM_COMPUTE_OPENCL_ENABLED)
+ if(_export_to_cl_image)
+ {
+ _validate_output |= image2d_from_buffer_supported(CLKernelLibrary::get().get_device());
+ _validate_output |= (get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device()) != 0);
+ }
+#endif // ARM_COMPUTE_OPENCL_ENABLED
+
+ const ConvolutionInfo conv_kernel_info
+ {
+ _conv_info, _depth_multiplier, _act_info, _dilation
+ };
// Create Depthwise Convolution configure function
- _dwc.configure(&_src, &_weights, &_biases, target_to_use, dwc_weights_info, dwc_info, _conv_info, _depth_multiplier, _dilation);
+ _dwc.configure(&_src, &_weights, &_biases, target_to_use, dwc_info, conv_kernel_info);
ARM_COMPUTE_ASSERT(_src.info()->is_resizable());
ARM_COMPUTE_ASSERT(_weights.info()->is_resizable());
@@ -479,7 +493,8 @@ public:
void allocate_and_run_target()
{
- add_padding_x({ &_src, &_weights, &_biases, &_target }, _data_layout);
+ add_padding_x({ &_src, &_biases, &_target }, _data_layout);
+ add_padding_x({ &_weights }, _data_layout, _export_to_cl_image); // Don't add left padding if cl image will be used
// Allocate tensors
_src.allocator()->allocate();
@@ -508,7 +523,10 @@ public:
}
// Compute function
- _dwc.run();
+ if(_validate_output)
+ {
+ _dwc.run();
+ }
// Reinstating original data layout for the test suite to properly check the values
if(!_in_place)
@@ -529,7 +547,10 @@ public:
const ConvolutionInfo info{ _conv_info, _depth_multiplier, _act_info, _dilation };
const TensorShape dst_shape = compute_depthwise_convolution_shape(TensorInfo(_input_shape, 1, _data_type), TensorInfo(_weights_shape, 1, _data_type), info);
- _reference = reference::activation_layer(reference::depthwise_convolution(src, weights, biases, dst_shape, _conv_info, _depth_multiplier, _dilation), _act_info);
+ if(_validate_output)
+ {
+ _reference = reference::activation_layer(reference::depthwise_convolution(src, weights, biases, dst_shape, _conv_info, _depth_multiplier, _dilation), _act_info);
+ }
}
protected:
@@ -573,6 +594,8 @@ protected:
Size2D _dilation{};
unsigned int _depth_multiplier{};
unsigned int _n0{};
+ bool _export_to_cl_image{};
+ bool _validate_output{ true };
bool _in_place{ false };
};