COMPMID-2179 New generic depthwise convolution for NEON F32 NHWC

Change-Id: I2b883785c0500d4bdb6ee4700382ee058be2cd36
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1538
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

2 files changed, 110 insertions, 0 deletions

diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index e41f299611..8fddec2c5f 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -53,6 +53,7 @@
 #include "arm_compute/core/NEON/kernels/NEDepthConvertLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthToSpaceLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.h"
+#include "arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthwiseIm2ColKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthwiseVectorToTensorKernel.h"
 #include "arm_compute/core/NEON/kernels/NEDepthwiseWeightsReshapeKernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayerKernel.h b/arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayerKernel.h
new file mode 100644
index 0000000000..63635b3a6c
--- /dev/null
+++ b/arm_compute/core/NEON/kernels/NEDepthwiseConvolutionLayerKernel.h
@@ -0,0 +1,109 @@
+/*
+ * 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_NEDEPTHWISECONVOLUTIONKERNEL_H__
+#define __ARM_COMPUTE_NEDEPTHWISECONVOLUTIONKERNEL_H__
+
+#include "arm_compute/core/NEON/INEKernel.h"
+
+namespace arm_compute
+{
+// Forward declarations
+class ITensor;
+
+/** Interface for the kernel to run a depthwise convolution on a tensor. */
+class NEDepthwiseConvolutionLayerKernel : public INEKernel
+{
+public:
+    const char *name() const override
+    {
+        return "NEDepthwiseConvolutionLayerKernel";
+    }
+    /** Default constructor */
+    NEDepthwiseConvolutionLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEDepthwiseConvolutionLayerKernel(const NEDepthwiseConvolutionLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEDepthwiseConvolutionLayerKernel &operator=(const NEDepthwiseConvolutionLayerKernel &) = delete;
+    /** Default Move Constructor. */
+    NEDepthwiseConvolutionLayerKernel(NEDepthwiseConvolutionLayerKernel &&) = default;
+    /** Default move assignment operator */
+    NEDepthwiseConvolutionLayerKernel &operator=(NEDepthwiseConvolutionLayerKernel &&) = default;
+    /** Initialize the function's source, destination and parameters.
+     *
+     * @note Supported data layouts: NHWC
+     *
+     * @param[in]  input            Source tensor. DataType supported: F32.
+     * @param[in]  weights          Weights tensor. This is a 3D tensor with dimensions [IFM, W, H]. Data type supported: Same as @p input.
+     * @param[in]  biases           Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. Data type supported: Same as @p input.
+     * @param[out] output           Destination tensor. Data type supported: Same as @p input.
+     * @param[in]  conv_info        Padding and stride information to use for the convolution.
+     * @param[in]  depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
+     * @param[in]  dilation         (Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
+     *
+     */
+    void configure(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier = 1,
+                   const Size2D &dilation = Size2D(1U, 1U));
+    /** Static function to check if given info will lead to a valid configuration of @ref NEDepthwiseConvolutionLayerKernel
+     *
+     * @note Supported data layouts: NHWC
+     *
+     * @param[in] input            Source tensor info. DataType supported: F32.
+     * @param[in] weights          Weights tensor info. This is a 3D tensor with dimensions [IFM, W, H]. Data type supported: Same as @p input.
+     * @param[in] biases           Biases tensor info. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. Data type supported: Same as @p input.
+     * @param[in] output           Destination tensor info. Data type supported: Same as @p input.
+     * @param[in] conv_info        Padding and stride information to use for the convolution.
+     * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
+     * @param[in] dilation         (Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier = 1,
+                           const Size2D &dilation = Size2D(1U, 1U));
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+    BorderSize border_size() const override;
+
+private:
+    template <typename T, int S, bool has_biases>
+    void run_depthwise(const Window &window);
+
+    /** Common signature for all the specialised depthwise convolution functions
+     *
+     * @param[in] window Region on which to execute the kernel.
+     */
+    using DepthwiseFunctionPtr = void (NEDepthwiseConvolutionLayerKernel::*)(const Window &window);
+
+    DepthwiseFunctionPtr _func;
+    BorderSize           _border_size;
+    const ITensor       *_input;
+    const ITensor       *_weights;
+    const ITensor       *_biases;
+    ITensor             *_output;
+    PadStrideInfo        _conv_info;
+    unsigned int         _depth_multiplier;
+    Size2D               _dilation;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NEDEPTHWISECONVOLUTIONKERNEL_H__ */