COMPMID-926 Add depth multiplier support to NEON/CL/GLES depthwise convolution

Change-Id: I03f32c62350e5ea43e77bb15fc5a832d83719e3b
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/126657
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

4 files changed, 46 insertions, 41 deletions

diff --git a/arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.h b/arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.h
index 0f3f4bfc76..f80985a936 100644
--- a/arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.h
+++ b/arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.h
@@ -39,15 +39,16 @@ public:
     CLDepthwiseConvolutionLayer3x3NCHWKernel();
     /** Initialize the function's source, destination, conv and border_size.
      *
-     * @param[in]  input     Source tensor. DataType supported: QASYMM8/F16/F32.
-     * @param[in]  weights   Weights tensor. A 3D tensor with dimensions [3, 3, IFM]. Data type supported: Same as @p input.
-     * @param[in]  biases    (Optional) 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]  act_info  (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU for QASYMM8 supported.
+     * @param[in]  input            Source tensor. DataType supported: QASYMM8/F16/F32.
+     * @param[in]  weights          Weights tensor. A 3D tensor with dimensions [3, 3, IFM]. Data type supported: Same as @p input.
+     * @param[in]  biases           (Optional) 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]  act_info         (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU for QASYMM8 supported.
      */
-    void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info,
+    void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier,
                    ActivationLayerInfo act_info) override;
 
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h b/arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h
index 4ecc07af6a..62b9a8682e 100644
--- a/arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h
+++ b/arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h
@@ -40,29 +40,31 @@ public:
     /** Default move assignment operator. */
     /** Initialize the function's source, destination, conv and border_size.
      *
-     * @param[in]  input     Source tensor. DataType supported: QASYMM8.
-     * @param[in]  weights   Weights tensor. A 3D tensor with dimensions [IFM, 3, 3]. Data type supported: Same as @p input.
-     * @param[in]  biases    (Optional) 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]  act_info  (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported.
+     * @param[in]  input            Source tensor. DataType supported: QASYMM8.
+     * @param[in]  weights          Weights tensor. A 3D tensor with dimensions [IFM, 3, 3]. Data type supported: Same as @p input.
+     * @param[in]  biases           (Optional) 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]  act_info         (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported.
      */
-    void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info,
+    void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier,
                    ActivationLayerInfo act_info) override;
     /** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseConvolutionLayer3x3NHWCKernel
      *
-     * @param[in] input     Source tensor. DataType supported: QASYMM8.
-     * @param[in] weights   Weights tensor. A 3D tensor with dimensions [IFM, 3, 3]. Data type supported: Same as @p input.
-     * @param[in] biases    (Optional) Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed.
-     *                      Data type supported: Same as @p input.
-     * @param[in] 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] act_info  (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported.
+     * @param[in] input            Source tensor. DataType supported: QASYMM8.
+     * @param[in] weights          Weights tensor. A 3D tensor with dimensions [IFM, 3, 3]. Data type supported: Same as @p input.
+     * @param[in] biases           (Optional) Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed.
+     *                             Data type supported: Same as @p input.
+     * @param[in] 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] act_info         (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
+    static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier,
                            ActivationLayerInfo act_info = ActivationLayerInfo());
 
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLDepthwiseIm2ColKernel.h b/arm_compute/core/CL/kernels/CLDepthwiseIm2ColKernel.h
index 86b8b90e94..b8343da50a 100644
--- a/arm_compute/core/CL/kernels/CLDepthwiseIm2ColKernel.h
+++ b/arm_compute/core/CL/kernels/CLDepthwiseIm2ColKernel.h
@@ -33,7 +33,7 @@ class ICLTensor;
 
 /** Interface for the depthwise im2col reshape kernel.
  *  This kernel reshape the input low 3 dimensions to a new 3D shape  where the output's first dimension is
- *  the linear patch size (FILTER_WIDTH * FILTER_HEIGHT) and second dimension is number of patches in per image and third dimension unchanged .
+ *  the linear patch size (FILTER_WIDTH * FILTER_HEIGHT) and second dimension is number of patches per image and third dimension unchanged .
  **/
 class CLDepthwiseIm2ColKernel : public ICLKernel
 {
@@ -50,15 +50,16 @@ public:
     CLDepthwiseIm2ColKernel &operator=(CLDepthwiseIm2ColKernel &&) = default;
     /** Set the input and output of the kernel.
      *
-     * @param[in]  input       The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
-     *                         while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8/F32
-     * @param[out] output      The output tensor. First 3 lower dimensions represent a transform of each 3D input,
-     *                         while every dimension above 3 represents a batch. Data types supported: Same as @p input
-     * @param[in]  kernel_dims The kernel dimensions (width and height).
-     * @param[in]  conv_info   Contains padding and stride information described in @ref PadStrideInfo.
-     * @param[in]  has_bias    Boolean that specifies if the depthwise convolution has bias.
+     * @param[in]  input            The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
+     *                              while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8/F32
+     * @param[out] output           The output tensor. First 3 lower dimensions represent a transform of each 3D input,
+     *                              while every dimension above 3 represents a batch. Data types supported: Same as @p input
+     * @param[in]  kernel_dims      The kernel dimensions (width and height).
+     * @param[in]  conv_info        Contains padding and stride information described in @ref PadStrideInfo.
+     * @param[in]  has_bias         Boolean that specifies if the depthwise convolution has bias.
+     * @param[in]  depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1.
      */
-    void configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias = false);
+    void configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias = false, unsigned int depth_multiplier = 1);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/ICLDepthwiseConvolutionLayer3x3Kernel.h b/arm_compute/core/CL/kernels/ICLDepthwiseConvolutionLayer3x3Kernel.h
index f02ba331b5..15233c5c32 100644
--- a/arm_compute/core/CL/kernels/ICLDepthwiseConvolutionLayer3x3Kernel.h
+++ b/arm_compute/core/CL/kernels/ICLDepthwiseConvolutionLayer3x3Kernel.h
@@ -50,15 +50,16 @@ public:
     ICLDepthwiseConvolutionLayer3x3Kernel &operator=(ICLDepthwiseConvolutionLayer3x3Kernel &&) = default;
     /** Initialize the function's source, destination, conv and border_size.
      *
-     * @param[in]  input     Source tensor. DataType supported: QASYMM8/F16/F32.
-     * @param[in]  weights   Weights tensor. A 3D tensor with dimensions [3, 3, IFM]. Data type supported: Same as @p input.
-     * @param[in]  biases    (Optional) 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]  act_info  (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU for QASYMM8 supported.
+     * @param[in]  input            Source tensor. DataType supported: QASYMM8/F16/F32.
+     * @param[in]  weights          Weights tensor. A 3D tensor with dimensions [3, 3, IFM]. Data type supported: Same as @p input.
+     * @param[in]  biases           (Optional) 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]  act_info         (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU for QASYMM8 supported.
      */
-    virtual void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info,
+    virtual void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier = 1,
                            ActivationLayerInfo act_info = ActivationLayerInfo()) = 0;
 
 protected: