MLCE-129: NEPad 30x slower than TensorFlow's implementation

Change-Id: I44770e6a3134c70c4bd58f890d06cb43c9bd8bff
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1853
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 114 insertions, 0 deletions

diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index f86a1d0507..8d8f7439a5 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -114,6 +114,7 @@
 #include "arm_compute/core/NEON/kernels/NENonLinearFilterKernel.h"
 #include "arm_compute/core/NEON/kernels/NENonMaximaSuppression3x3Kernel.h"
 #include "arm_compute/core/NEON/kernels/NENormalizationLayerKernel.h"
+#include "arm_compute/core/NEON/kernels/NEPadLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEPermuteKernel.h"
 #include "arm_compute/core/NEON/kernels/NEPixelWiseMultiplicationKernel.h"
 #include "arm_compute/core/NEON/kernels/NEPoolingLayerKernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEPadLayerKernel.h b/arm_compute/core/NEON/kernels/NEPadLayerKernel.h
new file mode 100644
index 0000000000..0a57b97997
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEPadLayerKernel.h
@@ -0,0 +1,113 @@
+/*
+ * 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_NEPADLAYERKERNEL_H__
+#define __ARM_COMPUTE_NEPADLAYERKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** NEON kernel to add padding to a tensor
+ *
+ * Add padding given padding information
+ */
+class NEPadLayerKernel : public INEKernel
+{
+public:
+    const char *name() const override
+    {
+        return "NEPadLayerKernel";
+    }
+    /** Default constructor */
+    NEPadLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEPadLayerKernel(const NEPadLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEPadLayerKernel &operator=(const NEPadLayerKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    NEPadLayerKernel(NEPadLayerKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    NEPadLayerKernel &operator=(NEPadLayerKernel &&) = default;
+    /** Default destructor */
+    ~NEPadLayerKernel() = default;
+
+    /** Initialize the function
+     *
+     * @param[in]  input          Source tensor. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[out] output         Output tensor. Data type supported: same as @p input
+     * @param[in]  padding        The padding for each spatial dimension of the input tensor. The pair padding[i]
+     *                            specifies the front and the end padding in the i-th dimension.
+     * @param[in]  constant_value (Optional) Constant value to be used for the padding
+     * @param[in]  mode           (Optional) Controls whether the padding should be filled with @p constant_value using CONSTANT.
+     *                           Only CONSTANT padding mode is currently supported
+     */
+    void configure(ITensor *input, ITensor *output, const PaddingList &padding, const PixelValue constant_value = PixelValue(), const PaddingMode mode = PaddingMode::CONSTANT);
+    /**  Static function to check if given info will lead to a valid configuration of @ref NEPadLayer.
+     *
+     * @param[in] input          Source tensor info. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32.
+     * @param[in] output         Output tensor info. Data type supported: same as @p input
+     * @param[in] padding        The padding for each spatial dimension of the input tensor. The pair padding[i]
+     *                           specifies the front and the end padding in the i-th dimension.
+     * @param[in] constant_value (Optional) Constant value to be used for the padding
+     * @param[in] mode           (Optional) Controls whether the padding should be filled with @p constant_value using CONSTANT.
+     *                           Only CONSTANT padding mode is currently supported
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding, const PixelValue constant_value = PixelValue(), const PaddingMode mode = PaddingMode::CONSTANT);
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+
+private:
+    /** Template function to run the padding function with constant padding
+     *
+     * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
+     */
+    template <typename T>
+    void run_pad_constant(const Window &window);
+
+    /** Function to run the padding function with constant padding for 3D input and 1D, 2D, 3D padding
+     *
+     * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
+     */
+    void run_pad_constant_uint8_3Dinput_3Dpad(const Window &window);
+
+    /** Common signature for all the specialised permute functions
+     *
+     * @param[in] window Region on which to execute the kernel.
+     */
+    using PadFunctionPtr = void (NEPadLayerKernel::*)(const Window &window);
+
+    PadFunctionPtr _func;
+    const ITensor *_input;
+    ITensor       *_output;
+    PaddingList    _padding;
+    PixelValue     _constant_value;
+    PaddingMode    _mode;
+};
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_NEPADLAYERKERNEL_H__ */