COMPMID-2449: Implement NEUnPoolLayer

Change-Id: I5677c87bba97dd395a3e13dbce34a3dd2c437033
Signed-off-by: morgolock <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3289
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

5 files changed, 207 insertions, 0 deletions

diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index 38701f434a..dfe0ccaafc 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -106,6 +106,7 @@
 #include "arm_compute/core/NEON/kernels/NELKTrackerKernel.h"
 #include "arm_compute/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.h"
 #include "arm_compute/core/NEON/kernels/NEMagnitudePhaseKernel.h"
+#include "arm_compute/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEMeanStdDevKernel.h"
 #include "arm_compute/core/NEON/kernels/NEMeanStdDevNormalizationKernel.h"
 #include "arm_compute/core/NEON/kernels/NEMedian3x3Kernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h b/arm_compute/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h
new file mode 100644
index 0000000000..269317b6c1
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h
@@ -0,0 +1,97 @@
+/*
+ * Copyright (c) 2020 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_NEMAXUNPOOLINGLAYERKERNEL_H
+#define ARM_COMPUTE_NEMAXUNPOOLINGLAYERKERNEL_H
+
+#include "arm_compute/core/NEON/INEKernel.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Interface for the pooling layer kernel */
+class NEMaxUnpoolingLayerKernel : public INEKernel
+{
+public:
+    const char *name() const override
+    {
+        return "NEMaxUnpoolingLayerKernel";
+    }
+    /** Default constructor */
+    NEMaxUnpoolingLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEMaxUnpoolingLayerKernel(const NEMaxUnpoolingLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEMaxUnpoolingLayerKernel &operator=(const NEMaxUnpoolingLayerKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    NEMaxUnpoolingLayerKernel(NEMaxUnpoolingLayerKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    NEMaxUnpoolingLayerKernel &operator=(NEMaxUnpoolingLayerKernel &&) = default;
+    /** Default destructor */
+    ~NEMaxUnpoolingLayerKernel() = default;
+    /** Set the input and output tensors.
+     *
+     * @note Output shape must be equal to the shape of the original input to pool.
+     *
+     * @param[in]  input     Source tensor. Data types supported: F16/F32.
+     * @param[out] indices   The indices of the maximal values. Data type supported: U32.
+     * @param[out] output    Destination tensor. Data types supported: Same as @p input.
+     * @param[in]  pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
+     */
+    void configure(const ITensor *input, const ITensor *indices, ITensor *output, const PoolingLayerInfo &pool_info);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEMaxUnpoolingLayerKernel
+     *
+     * @param[in] input     Source tensor. Data types supported: F16/F32.
+     * @param[in] output    Destination tensor. Data types supported: Same as @p input.
+     * @param[in] indices   The indices of the maximal values. Data type supported: U32.
+     * @param[in] pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info);
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+
+private:
+    /** Function to perform 2x2 pooling and compute the pooling indices. The indices can be used for max unpool.
+     *
+     * @param[in] window_input Input region on which to execute the kernel.
+     */
+    template <typename T>
+    void unpooling2(const Window &window_input);
+
+    using UnpoolingFunction = void (NEMaxUnpoolingLayerKernel::*)(const Window &window);
+
+private:
+    UnpoolingFunction _func;
+    const ITensor    *_input;
+    ITensor          *_output;
+    const ITensor    *_indices;
+    PoolingLayerInfo  _pool_info;
+    DataLayout        _data_layout;
+    unsigned int      _num_elems_processed_per_iteration;
+};
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_NEMAXUNPOOLINGLAYERKERNEL_H */
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index dfccec8b37..bc85c6986f 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -809,6 +809,37 @@ inline TensorShape compute_pool_shape(const ITensorInfo &input, PoolingLayerInfo
     return output_shape;
 }
 
+/** Calculate the output unpool shape of a tensor
+ *
+ * @param[in] input     Input tensor info
+ * @param[in] pool_info Pooling layer info
+ *
+ * @return the calculated shape
+ */
+inline TensorShape compute_unpool_shape(const ITensorInfo &input, PoolingLayerInfo pool_info)
+{
+    const unsigned int idx_width   = get_data_layout_dimension_index(input.data_layout(), DataLayoutDimension::WIDTH);
+    const unsigned int idx_height  = get_data_layout_dimension_index(input.data_layout(), DataLayoutDimension::HEIGHT);
+    const TensorShape  input_shape = input.tensor_shape();
+    ARM_COMPUTE_ERROR_ON(input_shape[idx_height] <= 1 || input_shape[idx_width] <= 1);
+    const PadStrideInfo pad_stride_info = pool_info.pad_stride_info;
+    const unsigned int  stride_x        = pad_stride_info.stride().first;
+    const unsigned int  stride_y        = pad_stride_info.stride().second;
+
+    const int pad_left   = pad_stride_info.pad_left();
+    const int pad_top    = pad_stride_info.pad_top();
+    const int pad_right  = pad_stride_info.pad_right();
+    const int pad_bottom = pad_stride_info.pad_bottom();
+
+    TensorShape        output_shape = input_shape;
+    const unsigned int out_width    = (input_shape[idx_width] - 1) * stride_x - pad_left - pad_right + pool_info.pool_size.width;
+    const unsigned int out_height   = (input_shape[idx_height] - 1) * stride_y - pad_top - pad_bottom + pool_info.pool_size.height;
+
+    output_shape.set(idx_width, out_width);
+    output_shape.set(idx_height, out_height);
+    return output_shape;
+}
+
 /** Calculate the output roi align shape of a tensor
  *
  * @param[in] input     Input tensor info
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index de364fa9af..19137b8830 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -105,6 +105,7 @@
 #include "arm_compute/runtime/NEON/functions/NELaplacianReconstruct.h"
 #include "arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEMagnitude.h"
+#include "arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEMeanStdDev.h"
 #include "arm_compute/runtime/NEON/functions/NEMeanStdDevNormalizationLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEMedian3x3.h"
diff --git a/arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h b/arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h
new file mode 100644
index 0000000000..f93e4caf93
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h
@@ -0,0 +1,77 @@
+/*
+ * Copyright (c) 2020 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_NEMAXUNPOOLINGLAYER_H
+#define ARM_COMPUTE_NEMAXUNPOOLINGLAYER_H
+
+#include "arm_compute/runtime/IFunction.h"
+
+#include "arm_compute/core/NEON/kernels/NEMaxUnpoolingLayerKernel.h"
+#include "arm_compute/core/NEON/kernels/NEMemsetKernel.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Function to perform MaxUnpooling. This function calls the following NEON kernels:
+ *
+ * -# @ref NEMemsetKernel
+ * -# @ref NEMaxUnpoolingLayerKernel
+ */
+class NEMaxUnpoolingLayer : public IFunction
+{
+public:
+    /** Constructor */
+    NEMaxUnpoolingLayer();
+    /** Set the input and output tensors.
+     *
+     * @note F16 is supported for pool sizes 2 and 3 only
+     *
+     * @param[in, out] input     Source tensor. (Written to only when padding != 0) Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
+     * @param[out]     output    Destination tensor. Data types supported: Same as @p input.
+     * @param[out]     indices   The indices of the maximal values. Data type supported: U32.
+     * @param[in]      pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
+     */
+    void configure(ITensor *input, ITensor *indices, ITensor *output, const PoolingLayerInfo &pool_info);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEMaxUnpoolingLayer
+     *
+     * @note F16 is supported for pool sizes 2 and 3 only
+     *
+     * @param[in] input     Source tensor. (Written to only when padding != 0) Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
+     * @param[in] indices   The indices of the maximal values. Data type supported: U32.
+     * @param[in] output    Destination tensor. Data types supported: Same as @p input.
+     * @param[in] pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    NEMemsetKernel            _memset_kernel;
+    NEMaxUnpoolingLayerKernel _unpooling_layer_kernel;
+};
+}
+#endif /* ARM_COMPUTE_NEMAXUNPOOLINGLAYER_H */