COMPMID-1933: Implement NEHeightConcatenateLayer.

Added support to concactenate tensors along the Y axis in NEConcatenateLayer.

Change-Id: Ib714bfcf9954cc35918efa7d52fc9164bb08bdf6
Signed-off-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/841
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

4 files changed, 109 insertions, 16 deletions

diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index 5b1b701a9d..1ce3821e81 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -89,6 +89,7 @@
 #include "arm_compute/core/NEON/kernels/NEHOGDescriptorKernel.h"
 #include "arm_compute/core/NEON/kernels/NEHOGDetectorKernel.h"
 #include "arm_compute/core/NEON/kernels/NEHarrisCornersKernel.h"
+#include "arm_compute/core/NEON/kernels/NEHeightConcatenateLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEHistogramKernel.h"
 #include "arm_compute/core/NEON/kernels/NEIm2ColKernel.h"
 #include "arm_compute/core/NEON/kernels/NEIntegralImageKernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEHeightConcatenateLayerKernel.h b/arm_compute/core/NEON/kernels/NEHeightConcatenateLayerKernel.h
new file mode 100644
index 0000000000..540de68cb6
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEHeightConcatenateLayerKernel.h
@@ -0,0 +1,84 @@
+/*
+ * Copyright (c) 2019 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_NEHEIGHTCONCATENATELAYERKERNEL_H__
+#define __ARM_COMPUTE_NEHEIGHTCONCATENATELAYERKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Interface for the height concatenate kernel.
+ *  The input tensor will be concatenated into the output tensor.
+ */
+class NEHeightConcatenateLayerKernel : public INEKernel
+{
+public:
+    const char *name() const override
+    {
+        return "NEHeightConcatenateLayerKernel";
+    }
+    /** Default constructor */
+    NEHeightConcatenateLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEHeightConcatenateLayerKernel(const NEHeightConcatenateLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEHeightConcatenateLayerKernel &operator=(const NEHeightConcatenateLayerKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    NEHeightConcatenateLayerKernel(NEHeightConcatenateLayerKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    NEHeightConcatenateLayerKernel &operator=(NEHeightConcatenateLayerKernel &&) = default;
+    /** Default destructor */
+    ~NEHeightConcatenateLayerKernel() = default;
+    /** Initialise the kernel's inputs and output
+     *
+     * @param[in]     input         Input tensor. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+     * @param[in]     height_offset The starting offset on the Y axis for the output tensor.
+     * @param[in,out] output        Output tensor. Data types supported: Same as @p input.
+     *
+     */
+    void configure(const ITensor *input, unsigned int height_offset, ITensor *output);
+    /**  Static function to check if given info will lead to a valid configuration of @ref NEHeightConcatenateLayerKernel
+     *
+     * @param[in] input         Input tensor info. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
+     * @param[in] height_offset The starting offset on the Y axis for the output tensor.
+     * @param[in] output        Output tensor info. Data types supported: Same as @p input.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, unsigned int height_offset, const ITensorInfo *output);
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+
+private:
+    const ITensor *_input;
+    ITensor       *_output;
+    unsigned int   _height_offset;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NEHEIGHTCONCATENATELAYERKERNEL_H__ */
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index 9d36405041..6782cda7fe 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -1222,30 +1222,30 @@ inline TensorShape calculate_depth_concatenate_shape(const std::vector<T *> &inp
     return out_shape;
 }
 
-/** Calculate the width concatenate output shape of a vector of tensors
+/** Calculate the concatenate output shape of the concatenate operation along a single axis
  *
- * @param[in] inputs_vector Vector containing the shapes of the inputs
+ * @param[in] input Vector containing the shapes of the inputs
+ * @param[in] axis  Axis along which to concatenate the input tensors
  *
  * @return the calculated shape
  */
 template <typename T>
-inline TensorShape calculate_width_concatenate_shape(const std::vector<T *> &inputs_vector)
+inline TensorShape calculate_concatenate_shape(const std::vector<T *> &input, size_t axis)
 {
-    TensorShape out_shape = extract_shape(inputs_vector[0]);
+    TensorShape out_shape = extract_shape(input[0]);
 
-    size_t width = 0;
-    for(const auto &tensor : inputs_vector)
+    size_t new_size = 0;
+    for(const auto &tensor : input)
     {
         ARM_COMPUTE_ERROR_ON(tensor == nullptr);
         const TensorShape shape = extract_shape(tensor);
-        width += shape.x();
+        new_size += shape[axis];
     }
 
-    out_shape.set(0, width);
+    out_shape.set(axis, new_size);
 
     return out_shape;
 }
-
 /** Calculate the stack output shape of a tensor
  *
  * @param[in] a           Input tensor info
diff --git a/arm_compute/runtime/NEON/functions/NEConcatenateLayer.h b/arm_compute/runtime/NEON/functions/NEConcatenateLayer.h
index 2cdc720fb6..b70d6ebc7c 100644
--- a/arm_compute/runtime/NEON/functions/NEConcatenateLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEConcatenateLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,6 +26,7 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
+#include "arm_compute/core/NEON/kernels/NEHeightConcatenateLayerKernel.h"
 #include "arm_compute/core/Types.h"
 
 #include <memory>
@@ -41,6 +42,7 @@ class Status;
 /** Basic function to execute concatenate tensors along a given axis. This function calls the following kernels:
  *
  * -# @ref NEWidthConcatenateLayer (if underlying concatenation axis is 0).
+ * -# @ref NEHeightConcatenateLayerKernel (if underlying concatenation axis is 1).
  * -# @ref NEDepthConcatenateLayer (if underlying concatenation axis is 2).
  */
 class NEConcatenateLayer : public IFunction
@@ -51,21 +53,21 @@ public:
     /** Initialise the kernel's inputs vector and output.
      *
      * @note Input and output tensor dimensions preconditions defer depending on the concatenation axis.
-     * @note Preconditions can be found respectively at @ref NEWidthConcatenateLayer and @ref NEDepthConcatenateLayer.
+     * @note Preconditions can be found respectively at @ref NEWidthConcatenateLayer, @ref NEHeightConcatenateLayerKernel and @ref NEDepthConcatenateLayer.
      *
      * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: QASYMM8/F16/F32.
      * @param[out]    output        Output tensor. Data types supported: Same as @p input.
-     * @param[in]     axis          Concatenation axis. Supported underlying concatenation axis are 0 and 2.
+     * @param[in]     axis          Concatenation axis. Supported underlying concatenation axis are 0, 1 and 2.
      */
     void configure(const std::vector<ITensor *> &inputs_vector, ITensor *output, DataLayoutDimension axis);
     /** Static function to check if given info will lead to a valid configuration of @ref NEConcatenateLayer
      *
      * @note Input and output tensor dimensions preconditions defer depending on the concatenation axis.
-     * @note Preconditions can be found respectively at @ref NEWidthConcatenateLayer and @ref NEDepthConcatenateLayer.
+     * @note Preconditions can be found respectively at @ref NEWidthConcatenateLayer, @ref NEHeightConcatenateLayerKernel and @ref NEDepthConcatenateLayer.
      *
      * @param[in] inputs_vector The vectors containing all the tensors info to concatenate. Data types supported: QASYMM8/F16/F32.
      * @param[in] output        Output tensor info. Data types supported: Same as @p input.
-     * @param[in] axis          Concatenation axis. Supported underlying concatenation axis are 0 and 2.
+     * @param[in] axis          Concatenation axis. Supported underlying concatenation axis are 0, 1 and 2.
      *
      * @return a status
      */
@@ -75,7 +77,13 @@ public:
     void run() override;
 
 private:
-    std::unique_ptr<IFunction> _concat_function;
+    void configure_h_concatenate(std::vector<ITensor *> inputs_vector, ITensor *output);
+    static Status validate_h_concatenate(const std::vector<ITensorInfo *> &inputs_vector, const ITensorInfo *output);
+
+    std::unique_ptr<IFunction>                        _concat_function;
+    std::unique_ptr<NEHeightConcatenateLayerKernel[]> _hconcat_kernels;
+    unsigned int                                      _num_inputs;
+    unsigned int                                      _axis;
 };
-}
+} // namespace arm_compute
 #endif /* __ARM_COMPUTE_NECONCATENATELAYER_H__ */