COMPMID-471 Implement Deconvolution on OpenCL

Change-Id: Ie00c6b08a51d30c5ce2637d40ee3d165b8a68686
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/110311
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>

13 files changed, 338 insertions, 194 deletions

diff --git a/arm_compute/core/CL/CLKernels.h b/arm_compute/core/CL/CLKernels.h
index 9da0e5ab3a..64687fb26a 100644
--- a/arm_compute/core/CL/CLKernels.h
+++ b/arm_compute/core/CL/CLKernels.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -42,6 +42,7 @@
 #include "arm_compute/core/CL/kernels/CLCol2ImKernel.h"
 #include "arm_compute/core/CL/kernels/CLColorConvertKernel.h"
 #include "arm_compute/core/CL/kernels/CLConvolutionKernel.h"
+#include "arm_compute/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h"
 #include "arm_compute/core/CL/kernels/CLDepthConcatenateLayerKernel.h"
 #include "arm_compute/core/CL/kernels/CLDepthConvertLayerKernel.h"
 #include "arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3Kernel.h"
diff --git a/arm_compute/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h b/arm_compute/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h
new file mode 100644
index 0000000000..8867ca1c37
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h
@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 2017, 2018 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_CLDECONVOLUTIONLAYERUPSAMPLEKERNEL_H__
+#define __ARM_COMPUTE_CLDECONVOLUTIONLAYERUPSAMPLEKERNEL_H__
+
+#include "arm_compute/core/CL/ICLKernel.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** Interface for the Deconvolution layer kernel on OpenCL.
+ */
+class CLDeconvolutionLayerUpsampleKernel : public ICLKernel
+{
+public:
+    /** Constructor */
+    CLDeconvolutionLayerUpsampleKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLDeconvolutionLayerUpsampleKernel(const CLDeconvolutionLayerUpsampleKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLDeconvolutionLayerUpsampleKernel &operator=(const CLDeconvolutionLayerUpsampleKernel &) = delete;
+    /** Default Move Constructor. */
+    CLDeconvolutionLayerUpsampleKernel(CLDeconvolutionLayerUpsampleKernel &&) = default;
+    /** Default move assignment operator. */
+    CLDeconvolutionLayerUpsampleKernel &operator=(CLDeconvolutionLayerUpsampleKernel &&) = default;
+    /** Default destructor */
+    ~CLDeconvolutionLayerUpsampleKernel() = default;
+
+    /** Initialise the kernel's input and output.
+     *
+     * @param[in]  input        Source tensor. Data types supported: F32.
+     * @param[out] output       Destination tensor. Data types supported: F32. All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
+     * @param[in]  inner_border Top and right inner border sizes. These rows and columns will be filled with zero.
+     * @param[in]  info         Contains padding and stride information described in @ref PadStrideInfo.
+     */
+    void configure(const ICLTensor *input, ICLTensor *output, const BorderSize &inner_border, const PadStrideInfo &info);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLDeconvolutionLayerUpsample
+     *
+     * @param[in] input        Source tensor info. Data types supported: F32.
+     * @param[in] output       Destination tensor info. Data types supported: F32. All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
+     * @param[in] inner_border Top and right inner border sizes. These rows and columns will be filled with zero.
+     * @param[in] info         Contains padding and stride information described in @ref PadStrideInfo.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const BorderSize &inner_border, const PadStrideInfo &info);
+
+    // Inherited methods overridden:
+    void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    const ICLTensor *_input;
+    ICLTensor       *_output;
+    BorderSize       _inner_border;
+    PadStrideInfo    _info;
+};
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_CLDECONVOLUTIONLAYERUPSAMPLEKERNEL_H__ */
diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index 8a4cf7abeb..d5c4c340ee 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -43,7 +43,6 @@
 #include "arm_compute/core/NEON/kernels/NEColorConvertKernel.h"
 #include "arm_compute/core/NEON/kernels/NEConvolutionKernel.h"
 #include "arm_compute/core/NEON/kernels/NECumulativeDistributionKernel.h"
-#include "arm_compute/core/NEON/kernels/NEDeconvolutionLayerUpsampleKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthConcatenateLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthConvertLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEDeconvolutionLayerUpsampleKernel.h b/arm_compute/core/NEON/kernels/NEDeconvolutionLayerUpsampleKernel.h
deleted file mode 100644
index 707564683f..0000000000
--- a/arm_compute/core/NEON/kernels/NEDeconvolutionLayerUpsampleKernel.h
+++ /dev/null
@@ -1,72 +0,0 @@
-/*
- * 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_NEDECONVOLUTIONLAYERKERNEL_H__
-#define __ARM_COMPUTE_NEDECONVOLUTIONLAYERKERNEL_H__
-
-#include "arm_compute/core/NEON/INEKernel.h"
-#include "arm_compute/core/Types.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** NEON kernel to perform scaling on a tensor */
-class NEDeconvolutionLayerUpsampleKernel : public INEKernel
-{
-public:
-    /** Default constructor */
-    NEDeconvolutionLayerUpsampleKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEDeconvolutionLayerUpsampleKernel(const NEDeconvolutionLayerUpsampleKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEDeconvolutionLayerUpsampleKernel &operator=(const NEDeconvolutionLayerUpsampleKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEDeconvolutionLayerUpsampleKernel(NEDeconvolutionLayerUpsampleKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEDeconvolutionLayerUpsampleKernel &operator=(NEDeconvolutionLayerUpsampleKernel &&) = default;
-    /** Default destructor */
-    ~NEDeconvolutionLayerUpsampleKernel() = default;
-
-    /** Initialise the kernel's inputs, output and interpolation policy
-     *
-     * @param[in]  input   Source tensor. Data types supported: F32.
-     * @param[in]  offsets Offset to access the pixel with NEAREST interpolation or the top-left pixel with BILINEAR interpolation in the input tensor. Data type supported: S32.
-     * @param[out] output  Destination tensor. Data types supported: F32. All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
-     */
-    void configure(const ITensor *input, const ITensor *offsets, ITensor *output);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-    BorderSize border_size() const override;
-
-private:
-    /** Function to perform scale using nearest interpolation on the given window */
-    void scale_nearest(const Window &window);
-
-    const ITensor *_offsets;
-    const ITensor *_input;
-    ITensor       *_output;
-};
-} // arm_compute
-#endif /*__ARM_COMPUTE_NEDECONVOLUTIONLAYERKERNEL_H__ */
diff --git a/arm_compute/core/Utils.h b/arm_compute/core/Utils.h
index f78add13f9..51967b1762 100644
--- a/arm_compute/core/Utils.h
+++ b/arm_compute/core/Utils.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016, 2017, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -614,25 +614,23 @@ TensorShape deconvolution_output_shape(const std::pair<unsigned int, unsigned in
 
 /** Returns expected width and height of the deconvolution's output tensor.
  *
- * @param[in] in_width      Width of input tensor (Number of columns)
- * @param[in] in_height     Height of input tensor (Number of rows)
- * @param[in] kernel_width  Kernel width.
- * @param[in] kernel_height Kernel height.
- * @param[in] padx          X axis padding.
- * @param[in] pady          Y axis padding.
- * @param[in] ax            The number of zeros added to right edge of the input.
- * @param[in] ay            The number of zeros added to top edge of the input.
- * @param[in] upscalex      How much to scale the X axis.
- * @param[in] upscaley      How much to scale the Y axis.
- * @param[in] round         Rounding policy to be used when computing the output's dimensions.
+ * @param[in] in_width           Width of input tensor (Number of columns)
+ * @param[in] in_height          Height of input tensor (Number of rows)
+ * @param[in] kernel_width       Kernel width.
+ * @param[in] kernel_height      Kernel height.
+ * @param[in] padx               X axis padding.
+ * @param[in] pady               Y axis padding.
+ * @param[in] inner_border_right The number of zeros added to right edge of the input.
+ * @param[in] inner_border_top   The number of zeros added to top edge of the input.
+ * @param[in] stride_x           X axis input stride.
+ * @param[in] stride_y           Y axis input stride.
  *
  * @return A pair with the new width in the first position and the new height in the second.
  */
-
 const std::pair<unsigned int, unsigned int> deconvolution_output_dimensions(unsigned int in_width, unsigned int in_height,
                                                                             unsigned int kernel_width, unsigned int kernel_height,
-                                                                            unsigned int padx, unsigned int pady, unsigned int ax, unsigned int ay,
-                                                                            float upscalex, float upscaley, DimensionRoundingType round);
+                                                                            unsigned int padx, unsigned int pady, unsigned int inner_border_right, unsigned int inner_border_top,
+                                                                            unsigned int stride_x, unsigned int stride_y);
 
 /** Returns expected width and height of output scaled tensor depending on dimensions rounding mode.
  *
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index f31eb3d336..c7667f2c7b 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -106,7 +106,8 @@ inline TensorShape compute_depthwise_convolution_shape(const ITensorInfo &input,
 
     unsigned int output_width  = 0;
     unsigned int output_height = 0;
-    std::tie(output_width, output_height) = scaled_dimensions(input_shape.x(), input_shape.y(), weights_shape.x(), weights_shape.y(), conv_info);
+    std::tie(output_width, output_height) = scaled_dimensions(input_shape.x(), input_shape.y(), weights_shape.x(),
+                                                              weights_shape.y(), conv_info);
 
     TensorShape output_shape{ input_shape };
     output_shape.set(0, output_width);
@@ -114,6 +115,16 @@ inline TensorShape compute_depthwise_convolution_shape(const ITensorInfo &input,
 
     return output_shape;
 }
+inline TensorShape compute_deconvolution_shape(const ITensorInfo &input, unsigned int sx, unsigned int sy, unsigned int inner_border_right, unsigned int inner_border_top, const PadStrideInfo &info)
+{
+    TensorShape        scale_out_shape(input.tensor_shape());
+    const unsigned int out_x = input.dimension(0) + (input.dimension(0) - 1) * (sx - 1) + inner_border_right + 2 * info.pad().first;
+    const unsigned int out_y = input.dimension(1) + (input.dimension(1) - 1) * (sy - 1) + inner_border_top + 2 * info.pad().second;
+    scale_out_shape.set(0, out_x);
+    scale_out_shape.set(1, out_y);
+
+    return scale_out_shape;
+}
 } // namespace shape_calculator
 } // namespace misc
 } // namespace arm_compute
diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h
index f6ecef7a51..1154ab79aa 100644
--- a/arm_compute/runtime/CL/CLFunctions.h
+++ b/arm_compute/runtime/CL/CLFunctions.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -42,6 +42,8 @@
 #include "arm_compute/runtime/CL/functions/CLColorConvert.h"
 #include "arm_compute/runtime/CL/functions/CLConvolution.h"
 #include "arm_compute/runtime/CL/functions/CLConvolutionLayer.h"
+#include "arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h"
+#include "arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h"
 #include "arm_compute/runtime/CL/functions/CLDepthConcatenateLayer.h"
 #include "arm_compute/runtime/CL/functions/CLDepthConvertLayer.h"
 #include "arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h"
diff --git a/arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h
new file mode 100644
index 0000000000..e98cc9b3d6
--- /dev/null
+++ b/arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h
@@ -0,0 +1,103 @@
+/*
+ * Copyright (c) 2017, 2018 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_CLDECONVOLUTIONLAYER_H__
+#define __ARM_COMPUTE_CLDECONVOLUTIONLAYER_H__
+
+#include "arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h"
+#include "arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h"
+
+#include "arm_compute/runtime/CL/CLMemoryGroup.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+class ICLTensor;
+/** Function to run the deconvolution layer.
+ *
+ * Deconvolution Layer is the backward pass of Convolution Layer. First we transform the input depending on the stride and pad info and then perform a 1x1
+ * convolution pass. Input stride defines how many zeroes we should put between each element of the input, pad is the amount of padding and finally a is a user
+ * specified value where a < stride - 1, that increases the padding top and right of the input image.
+ *
+ *  The relation between input to output is as follows:
+ *      width_output = round((width_input − 1) ∗ (stride_x - 1) − 2 ∗ padding_x + kernel_x + inner_border_right )
+ *      height_output = round((height_input − 1) ∗ (stride_y - 1) − 2 ∗ padding_y + kernel_y + inner_border_top )
+ *
+ *  where:
+ *      width_input is the size of the first input dimension.
+ *      height_input is the size of the second input dimension.
+ *      width_output is the size of the first output dimension.
+ *      height_output is the size of the second output dimension.
+ *      kernel_x and kernel_y are the convolution sizes in x and y.
+ *      inner_border_right and inner_border_top the number of zeros added to the right and top edges of the input.
+ *      stride_x and stride_y is the input stride of the first and second dimension.
+ *
+ */
+class CLDeconvolutionLayer : public IFunction
+{
+public:
+    /** Constructor */
+    CLDeconvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Set the input, weights, biases and output tensors.
+     *
+     * @param[in,out] input              Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32.
+     * @param[in]     weights            The 4d weights with dimensions [width, height, OFM, IFM]. Data type supported: Same as @p input.
+     * @param[in]     bias               (Optional) The biases have one dimension. Data type supported: Same as @p input.
+     * @param[out]    output             Output tensor. The output has the same number of dimensions as the @p input.
+     * @param[in]     info               Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
+     * @param[in]     inner_border_right The number of zeros added to right edge of the input.
+     * @param[in]     inner_border_top   The number of zeros added to top edge of the input.
+     *
+     */
+    void configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, const PadStrideInfo &info,
+                   unsigned int inner_border_right, unsigned int inner_border_top);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLDeconvolutionLayer
+     *
+     * @param[in] input              Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32.
+     * @param[in] weights            The 4d weights info with dimensions [width, height, OFM, IFM]. Data type supported: Same as @p input.
+     * @param[in] bias               (Optional) The biases have one dimension. Data type supported: Same as @p input.
+     * @param[in] output             Output tensor info. The output has the same number of dimensions as the @p input.
+     * @param[in] info               Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
+     * @param[in] inner_border_right The number of zeros added to right edge of the input.
+     * @param[in] inner_border_top   The number of zeros added to top edge of the input.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, ITensorInfo *output, const PadStrideInfo &info,
+                           unsigned int inner_border_right, unsigned int inner_border_top);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    CLMemoryGroup                _memory_group;
+    CLDeconvolutionLayerUpsample _scale_f;
+    CLDirectConvolutionLayer     _conv_f;
+    CLTensor                     _scaled_output;
+};
+}
+#endif /* __ARM_COMPUTE_CLDECONVOLUTIONLAYER_H__ */
\ No newline at end of file
diff --git a/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h b/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h
new file mode 100644
index 0000000000..74ee4efb2c
--- /dev/null
+++ b/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h
@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2017, 2018 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_CLDECONVOLUTIONLAYERUPSAMPLE_H__
+#define __ARM_COMPUTE_CLDECONVOLUTIONLAYERUPSAMPLE_H__
+
+#include "arm_compute/runtime/IFunction.h"
+
+#include "arm_compute/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/CLMemoryGroup.h"
+#include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** Basic function to run @ref CLDeconvolutionLayerUpsampleKernel */
+class CLDeconvolutionLayerUpsample : public IFunction
+{
+public:
+    /** Default constructor */
+    CLDeconvolutionLayerUpsample();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLDeconvolutionLayerUpsample(const CLDeconvolutionLayerUpsample &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLDeconvolutionLayerUpsample &operator=(const CLDeconvolutionLayerUpsample &) = delete;
+    /** Allow instances of this class to be moved */
+    CLDeconvolutionLayerUpsample(CLDeconvolutionLayerUpsample &&) = default;
+    /** Allow instances of this class to be moved */
+    CLDeconvolutionLayerUpsample &operator=(CLDeconvolutionLayerUpsample &&) = default;
+    /** Default destructor */
+    virtual ~CLDeconvolutionLayerUpsample() = default;
+
+    /** Initialize the function's source, destination, interpolation type and border_mode.
+     *
+     * @param[in, out] input        Source tensor. Data type supported: F32.
+     * @param[out]     output       Destination tensor. Data type supported: F32.
+     * @param[in]      inner_border The number of zeros added to right and top edges of the input.
+     * @param[in]      info         Contains padding and policies to be used in the deconvolution.
+     */
+    void configure(ICLTensor *input, ICLTensor *output, const BorderSize &inner_border,
+                   const PadStrideInfo &info);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLDeconvolutionLayerUpsample
+     *
+     * @param[in] input        Source tensor info. Data type supported: F32.
+     * @param[in] output       Destination tensor info. Data type supported: F32.
+     * @param[in] inner_border The number of zeros added to right and top edges of the input.
+     * @param[in] info         Contains padding and policies to be used in the deconvolution.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const BorderSize &inner_border,
+                           const PadStrideInfo &info);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    CLDeconvolutionLayerUpsampleKernel _upsample;
+    ICLTensor                         *_output;
+};
+}
+#endif /* __ARM_COMPUTE_CLDECONVOLUTIONLAYERUPSAMPLE_H__ */
diff --git a/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h
index f31a45be90..205c90c478 100644
--- a/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -40,6 +40,7 @@ class ICLTensor;
 class CLDirectConvolutionLayer : public IFunction
 {
 public:
+    /** Default constructor */
     CLDirectConvolutionLayer();
     /** Set the input and output tensors.
      *
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 08852cf368..d09fcb280c 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -44,7 +44,6 @@
 #include "arm_compute/runtime/NEON/functions/NEConvolution.h"
 #include "arm_compute/runtime/NEON/functions/NEConvolutionLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h"
-#include "arm_compute/runtime/NEON/functions/NEDeconvolutionLayerUpsample.h"
 #include "arm_compute/runtime/NEON/functions/NEDepthConcatenateLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDepthConvertLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDepthwiseConvolutionLayer.h"
diff --git a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
index 8757bc63aa..091a928db6 100644
--- a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017, 2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,7 +24,6 @@
 #ifndef __ARM_COMPUTE_NEDECONVOLUTIONLAYER_H__
 #define __ARM_COMPUTE_NEDECONVOLUTIONLAYER_H__
 
-#include "arm_compute/runtime/NEON/functions/NEDeconvolutionLayerUpsample.h"
 #include "arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h"
 
 #include "arm_compute/core/Types.h"
@@ -39,13 +38,13 @@ namespace arm_compute
 {
 /** Function to run the deconvolution layer.
  *
- *  The operation is similar to convolution but it's implemented by up-sampling the inputs with zeros insertions between the inputs and convolving
- *  the kernels on the up-sampled result.
+ * Deconvolution Layer is the backward pass of Convolution Layer. First we transform the input depending on the stride and pad info and then perfrom a 1x1
+ * convolution pass. Input stride defines how many zeroes we should put between each element of the input, pad is the amount of padding and finaly a is a user
+ * specified value where a < stride - 1 that increases the padding top and right of the input image.
  *
- *  Before the Deconvolution is done, up-scaling the first 2D with zeros is performed. The relation between input to
- *  output is as follows:
- *      width_output = round((width_input − 1) ∗ upscale_x − 2 ∗ padding_x + kernel_x + a_x )
- *      height_output = round((height_input − 1) ∗ upscale_y − 2 ∗ padding_y + kernel_y + a_y )
+ *  The relation between input to output is as follows:
+ *       width_output = round((width_input − 1) ∗ (stride_x - 1) − 2 ∗ padding_x + kernel_x + inner_border_right )
+ *       height_output = round((height_input − 1) ∗ (stride_y - 1) − 2 ∗ padding_y + kernel_y + inner_border_top )
  *
  *  where
  *      width is the size of the first input dimension.
@@ -53,44 +52,54 @@ namespace arm_compute
  *      width_output is the size of the first output dimension.
  *      height_output is the size of the second output dimension.
  *      kernel_x and kernel_y are the convolution sizes in x and y.
- *      ax and ay the number of zeros added to the top and right edges of the input.
- *      upscale_x and upscale_y how much to scale the X and Y axis.
+ *      inner_border_right and inner_border_top the number of zeros added to the top and right edges of the input.
+ *      stride_x and stride_y is the input stride of the first and second dimension.
  *
  *  This function calls the following NEON kernels:
  *
- * -# @ref NEDeconvolutionLayerUpsampleKernel
  * -# @ref NEDirectConvolutionLayer
  *
  */
 class NEDeconvolutionLayer : public IFunction
 {
 public:
-    /** Constructor */
+    /** Default constructor */
     NEDeconvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEDeconvolutionLayer(const NEDeconvolutionLayer &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEDeconvolutionLayer &operator=(const NEDeconvolutionLayer &) = delete;
+    /** Allow instances of this class to be moved */
+    NEDeconvolutionLayer(NEDeconvolutionLayer &&) = default;
+    /** Allow instances of this class to be moved */
+    NEDeconvolutionLayer &operator=(NEDeconvolutionLayer &&) = default;
+    /** Default destructor */
+    virtual ~NEDeconvolutionLayer() = default;
     /** Set the input, weights, biases and output tensors.
      *
-     * @param[in,out] input    Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32.
-     * @param[in]     weights  The 4d weights with dimensions [width, height, OFM, IFM]. Data type supported: Same as @p input.
-     * @param[in]     bias     Optional, ignored if NULL. The biases have one dimension. Data type supported: Same as @p input.
-     * @param[out]    output   Output tensor. The output has the same number of dimensions as the @p input.
-     * @param[in]     info     Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
-     * @param[in]     ax       The number of zeros added to right edge of the input.
-     * @param[in]     ay       The number of zeros added to top edge of the input.
-     * @param[in]     upscalex How much to scale the X axis.
-     * @param[in]     upscaley How much to scale the Y axis.
+     * @param[in,out] input              Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32.
+     * @param[in]     weights            The 4d weights with dimensions [width, height, OFM, IFM]. Data type supported: Same as @p input.
+     * @param[in]     bias               Optional, ignored if NULL. The biases have one dimension. Data type supported: Same as @p input.
+     * @param[out]    output             Output tensor. The output has the same number of dimensions as the @p input.
+     * @param[in]     info               Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
+     * @param[in]     inner_border_right The number of zeros added to right edge of the input.
+     * @param[in]     inner_border_top   The number of zeros added to top edge of the input.
      *
      */
     void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &info,
-                   unsigned int ax, unsigned int ay, float upscalex, float upscaley);
+                   unsigned int inner_border_right, unsigned int inner_border_top);
 
     // Inherited methods overridden:
     void run() override;
 
 private:
-    MemoryGroup                  _memory_group;
-    NEDeconvolutionLayerUpsample _scale_f;
-    NEDirectConvolutionLayer     _conv_f;
-    Tensor                       _scaled_output;
+    MemoryGroup              _memory_group;
+    NEDirectConvolutionLayer _conv_f;
+    Tensor                   _scaled_output;
+    ITensor                 *_input;
+    PadStrideInfo            _info;
+    std::pair<unsigned int, unsigned int> _inner_border;
 };
 } // arm_compute
 #endif /* __ARM_COMPUTE_NEDECONVOLUTIONLAYER_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayerUpsample.h b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayerUpsample.h
deleted file mode 100644
index d2ac12a58a..0000000000
--- a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayerUpsample.h
+++ /dev/null
@@ -1,72 +0,0 @@
-/*
- * Copyright (c) 2016, 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_NEDECONVOLUTIONUPSAMPLE_H__
-#define __ARM_COMPUTE_NEDECONVOLUTIONUPSAMPLE_H__
-
-#include "arm_compute/core/NEON/kernels/NEDeconvolutionLayerUpsampleKernel.h"
-#include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
-#include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/IFunction.h"
-#include "arm_compute/runtime/IMemoryManager.h"
-#include "arm_compute/runtime/MemoryGroup.h"
-#include "arm_compute/runtime/Tensor.h"
-
-#include <cstdint>
-#include <memory>
-
-namespace arm_compute
-{
-class ITensor;
-
-/** Basic function to run @ref NEDeconvolutionLayerUpsampleKernel */
-class NEDeconvolutionLayerUpsample : public IFunction
-{
-public:
-    /** Constructor
-     *
-     * Initialize NEDeconvolutionLayerUpsample
-     */
-    NEDeconvolutionLayerUpsample(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
-    /** Initialize the function's source, destination, interpolation type and border_mode.
-     *
-     * @param[in, out] input  Source tensor. Data type supported: F32.
-     * @param[out]     output Destination tensor. Data type supported: F32.
-     * @param[in]      a      Top and right inner border sizes. These rows and columns will be filled with zero.
-     * @param[in]      iz     The number of zeros to be inserted between each input sample
-     * @param[in]      info   Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
-     */
-    void configure(ITensor *input, ITensor *output, const std::pair<unsigned int, unsigned int> &a,
-                   const std::pair<unsigned int, unsigned int> &iz, const PadStrideInfo &info);
-
-    // Inherited methods overridden:
-    void run() override;
-
-private:
-    MemoryGroup                        _memory_group;
-    Tensor                             _offsets;
-    NEFillBorderKernel                 _border_handler;
-    NEDeconvolutionLayerUpsampleKernel _upsample;
-};
-} // arm_compute
-#endif /*__ARM_COMPUTE_NEDECONVOLUTIONUPSAMPLE_H__ */