COMPMID-464 Implement Depthwise separable convolution on NEON

Change-Id: Icdaf7c2666a7fbb9b3c32ba2d70b667a59a6eeb9
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/110468
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com>

5 files changed, 197 insertions, 1 deletions

diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 203df33692..4b36202fd9 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -47,6 +47,7 @@
 #include "arm_compute/runtime/NEON/functions/NEDepthConcatenate.h"
 #include "arm_compute/runtime/NEON/functions/NEDepthConvert.h"
 #include "arm_compute/runtime/NEON/functions/NEDepthwiseConvolution.h"
+#include "arm_compute/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDequantizationLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDerivative.h"
 #include "arm_compute/runtime/NEON/functions/NEDilate.h"
diff --git a/arm_compute/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.h
new file mode 100644
index 0000000000..3f4c1389f0
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.h
@@ -0,0 +1,79 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_NEON_DEPTHWISE_SEPARABLE_CONVOLUTION_H__
+#define __ARM_COMPUTE_NEON_DEPTHWISE_SEPARABLE_CONVOLUTION_H__
+
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/NEON/INESimpleFunction.h"
+#include "arm_compute/runtime/NEON/functions/NEDepthwiseConvolution.h"
+#include "arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h"
+#include "arm_compute/runtime/Tensor.h"
+
+#include <cstdint>
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Basic function to execute depthwise convolution. This function calls the following NEON kernels and function:
+ *
+ * -# @ref NEDepthwiseConvolution
+ * -# @ref NEDirectConvolutionLayer
+ *
+ */
+class NEDepthwiseSeparableConvolutionLayer : public IFunction
+{
+public:
+    /** Default constructor */
+    NEDepthwiseSeparableConvolutionLayer();
+    /** Set the input and output tensors.
+     *
+     * @param[in]  input               Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
+     *                                 while every optional dimension from 4 and above represent a batch of inputs. Data types supported: F32.
+     * @param[in]  depthwise_weights   Depthwise convolution weights tensor. These are 3D tensors with dimensions [kernel_x, kernel_y, IFM]. Data type supported: Same as @p input.
+     * @param[in]  depthwise_biases    (Optional) Biases tensor.Biases are 1D tensor with dimensions [IFM]. Must be nullptr if not needed.
+     *                                 Data type supported: Same as @p weights.
+     * @param[out] depthwise_out       Depthwise destination tensor.
+     * @param[in]  pointwise_weights   Pointwise convolution weights tensor. These are 4D tensors with dimensions [1, 1, IFM, OFM]. Data type supported: Same as @p input.
+     * @param[in]  pointwise_biases    (Optional) Biases tensor. Biases are 1D tensor with dimensions [OFM]. Must be nullptr if not needed.
+     *                                 Data type supported: Same as @p weights.
+     * @param[out] output              Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
+     *                                 Data types supported: Same as @p input.
+     * @param[in]  depthwise_conv_info Contains padding and stride information described in @ref PadStrideInfo for depthwise convolution.
+     * @param[in]  pointwise_conv_info Contains padding and stride information described in @ref PadStrideInfo for pointwise convolution.
+     */
+    void configure(ITensor *input, const ITensor *depthwise_weights, const ITensor *depthwise_biases, ITensor *depthwise_out,
+                   const ITensor *pointwise_weights, const ITensor *pointwise_biases, ITensor *output,
+                   const PadStrideInfo &depthwise_conv_info, const PadStrideInfo &pointwise_conv_info);
+
+    // Inherited methods overriden:
+    void run() override;
+
+private:
+    NEDepthwiseConvolution   _depthwise_conv;
+    NEDirectConvolutionLayer _pointwise_conv;
+};
+}
+#endif /*__ARM_COMPUTE_NEON_DEPTHWISE_SEPARABLE_CONVOLUTION_H__ */
diff --git a/src/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.cpp b/src/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.cpp
new file mode 100644
index 0000000000..d70a6689ac
--- /dev/null
+++ b/src/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.cpp
@@ -0,0 +1,50 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.h"
+
+#include "arm_compute/core/ITensor.h"
+#include "arm_compute/core/PixelValue.h"
+#include "arm_compute/runtime/NEON/NEScheduler.h"
+#include "support/ToolchainSupport.h"
+
+using namespace arm_compute;
+
+NEDepthwiseSeparableConvolutionLayer::NEDepthwiseSeparableConvolutionLayer()
+    : _depthwise_conv(), _pointwise_conv()
+{
+}
+
+void NEDepthwiseSeparableConvolutionLayer::configure(ITensor *input, const ITensor *depthwise_weights, const ITensor *depthwise_biases, ITensor *depthwise_out,
+                                                     const ITensor *pointwise_weights, const ITensor *pointwise_biases, ITensor *output,
+                                                     const PadStrideInfo &depthwise_conv_info, const PadStrideInfo &pointwise_conv_info)
+{
+    _depthwise_conv.configure(input, depthwise_weights, depthwise_biases, depthwise_out, depthwise_conv_info);
+    _pointwise_conv.configure(depthwise_out, pointwise_weights, pointwise_biases, output, pointwise_conv_info);
+}
+
+void NEDepthwiseSeparableConvolutionLayer::run()
+{
+    _depthwise_conv.run();
+    _pointwise_conv.run();
+}
\ No newline at end of file
diff --git a/tests/validation/NEON/DepthwiseSeparableConvolutionLayer.cpp b/tests/validation/NEON/DepthwiseSeparableConvolutionLayer.cpp
new file mode 100644
index 0000000000..c3e6705775
--- /dev/null
+++ b/tests/validation/NEON/DepthwiseSeparableConvolutionLayer.cpp
@@ -0,0 +1,66 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONCLCTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NEDepthwiseSeparableConvolutionLayer.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+#include "tests/NEON/Accessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/MobileNetDepthwiseSeparableConvolutionLayerDataset.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/DepthwiseSeparableConvolutionLayerFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+RelativeTolerance<float> tolerance_f32(0.1f); /**< Tolerance value for comparing reference's output against implementation's output for DataType::F32 */
+const float              tolerance_num = 0.001f;
+} // namespace
+
+TEST_SUITE(NEON)
+TEST_SUITE(DepthwiseSeparableConvolutionLayer)
+
+// Configuration test to do
+
+template <typename T>
+using NEDepthwiseSeparableConvolutionLayerFixture = DepthwiseSeparableConvolutionValidationFixture<Tensor, Accessor, NEDepthwiseSeparableConvolutionLayer, T>;
+
+FIXTURE_DATA_TEST_CASE(RunSmall, NEDepthwiseSeparableConvolutionLayerFixture<float>, framework::DatasetMode::PRECOMMIT, datasets::MobileNetDepthwiseSeparableConvolutionLayerDataset())
+{
+    // Validate output
+    validate(Accessor(_target), _reference, tolerance_f32, tolerance_num);
+}
+TEST_SUITE_END()
+TEST_SUITE_END()
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/DepthwiseSeparableConvolutionLayerFixture.h b/tests/validation/fixtures/DepthwiseSeparableConvolutionLayerFixture.h
index 112bd19848..9b9c0c5af7 100644
--- a/tests/validation/fixtures/DepthwiseSeparableConvolutionLayerFixture.h
+++ b/tests/validation/fixtures/DepthwiseSeparableConvolutionLayerFixture.h
@@ -88,7 +88,7 @@ protected:
         TensorType dst               = create_tensor<TensorType>(output_shape, DataType::F32);
 
         // Create Depthwise Separable Convolution Layer configure function
-        CLDepthwiseSeparableConvolutionLayer depthwise_separable_convolution_layer;
+        FunctionType depthwise_separable_convolution_layer;
         depthwise_separable_convolution_layer.configure(&src, &depthwise_weights, &depthwise_biases, &depthwise_out, &pointwise_weights, &pointwise_biases, &dst, pad_stride_depthwise_info,
                                                         pad_stride_pointwise_info);