COMPMID-414 - Port CLConvolutionLayer to support 8 bit fixed point - CLWeightsReshapeKernel

Change-Id: Ie32e6bdd557a8243eb9988aa7eab4e4ca2291e79
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/78701
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Moritz Pflanzer <moritz.pflanzer@arm.com>

3 files changed, 21 insertions, 69 deletions

diff --git a/arm_compute/core/CL/kernels/CLWeightsReshapeKernel.h b/arm_compute/core/CL/kernels/CLWeightsReshapeKernel.h
index 1dc8a8b80e..0d00f0e00e 100644
--- a/arm_compute/core/CL/kernels/CLWeightsReshapeKernel.h
+++ b/arm_compute/core/CL/kernels/CLWeightsReshapeKernel.h
@@ -31,11 +31,8 @@ namespace arm_compute
 class CLWeightsReshapeKernel : public ICLKernel
 {
 public:
-    /** Constructor.
-     *
-     * @param[in] is_shared Flag to indicate whether the weights are shared or not.
-     */
-    CLWeightsReshapeKernel(bool is_shared = false);
+    /** Constructor.*/
+    CLWeightsReshapeKernel();
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     CLWeightsReshapeKernel(const CLWeightsReshapeKernel &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
@@ -50,7 +47,7 @@ public:
     /** Set the input and output of the kernel.
      *
      * @param[in]  input  The input tensor to convert. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM] if shared,
-     *                    and 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM,  num_patches] if unshared. Data types supported: F16, F32
+     *                    and 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM,  num_patches] if unshared. Data types supported: QS8/F16/F32
      * @param[in]  biases The shared biases tensor to append.  Bias is 1D tensor with dimensions [OFM] if shared and 2D tensor with
      *                    dimensions [OFM, num_patches] if unshared. Data types supported: Same as @p input
      * @param[out] output The output tensor. Should be a 2D Tensor. Data types supported: Same as @p input
@@ -58,57 +55,12 @@ public:
     void configure(const ICLTensor *input, const ICLTensor *biases, ICLTensor *output);
 
     // Inherited methods overridden:
-    virtual void run(const Window &window, cl::CommandQueue &queue) = 0;
+    void run(const Window &window, cl::CommandQueue &queue) override;
 
-protected:
-    bool             _is_shared;
+private:
     const ICLTensor *_input;
     const ICLTensor *_biases;
     ICLTensor       *_output;
 };
-
-/** Interface for the weights reshape kernel used by convolution and fully connected layers.
- *
- * Rearranges each 3-dimensional kernel to a single row leading to a matrix with linearized kernels.
- * In combination with the @ref CLIm2ColKernel can transform a convolution into a matrix multiplication.
- *
- * For example assuming a 3D weight kernel of 3x3 dimensions and depth of 2 we have:
- * @f[
- * \left( \begin{array}{ccc}
- * a000 & a001 & a002 \\
- * a010 & a011 & a012 \\
- * a020 & a021 & a022 \\
- * \end{array} \right)
- * \left( \begin{array}{ccc}
- * a100 & a101 & a102 \\
- * a110 & a111 & a112 \\
- * a120 & a121 & a122 \\
- * \end{array} \right)
- * \rightarrow
- * \left( \begin{array}{ccccccccc}
- * a000 & a001 & a002 & a010 & a011 & a012 & a020 & a021 & a022 & a100 & a101 & a102 & a110 & a111 & a112 & a120 & a121 & a122 \\
- * \end{array} \right)
- * @f]
- */
-class CLConvolutionLayerWeightsReshapeKernel : public CLWeightsReshapeKernel
-{
-public:
-    /** Default constructor */
-    CLConvolutionLayerWeightsReshapeKernel();
-
-    // Inherited methods overridden:
-    void run(const Window &window, cl::CommandQueue &queue) override;
-};
-
-/** Interface for the weights reshape kernel used by locally connected layers. */
-class CLLocallyConnectedLayerWeightsReshapeKernel : public CLWeightsReshapeKernel
-{
-public:
-    /** Default constructor */
-    CLLocallyConnectedLayerWeightsReshapeKernel();
-
-    // Inherited methods overridden:
-    void run(const Window &window, cl::CommandQueue &queue) override;
-};
 }
 #endif /*__ARM_COMPUTE_CLWEIGHTSRESHAPEKERNEL_H__ */
diff --git a/arm_compute/runtime/CL/functions/CLConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLConvolutionLayer.h
index 6a40396f9a..8030b40a71 100644
--- a/arm_compute/runtime/CL/functions/CLConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLConvolutionLayer.h
@@ -53,7 +53,7 @@ public:
     CLConvolutionLayerReshapeWeights();
     /** Set the input and output tensors.
      *
-     * @param[in]  weights      Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: F32.
+     * @param[in]  weights      Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: QS8/F16/F32.
      * @param[in]  biases       Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p weights.
      * @param[out] output       Destination tensor. Data types supported: Same as @p weights.
      * @param[in]  transpose1xW True if the weights are to undergo a 1xW transposition after reshaping (in case of GEMM operation), false otherwise.
@@ -64,16 +64,16 @@ public:
     void run() override;
 
 private:
-    CLConvolutionLayerWeightsReshapeKernel _weights_reshape_kernel;
-    CLGEMMTranspose1xWKernel               _weights_transposed_kernel;
-    CLTensor                               _weights_reshaped;
-    bool                                   _transpose1xW;
+    CLWeightsReshapeKernel   _weights_reshape_kernel;
+    CLGEMMTranspose1xWKernel _weights_transposed_kernel;
+    CLTensor                 _weights_reshaped;
+    bool                     _transpose1xW;
 };
 
 /** Basic function to compute the convolution layer. This function calls the following OpenCL kernels:
  *
- * -# @ref CLConvolutionLayerWeightsReshapeKernel (executed only once for each configuration)
- * -# @ref CLGEMMTranspose1xWKernel               (executed only once for each configuration)
+ * -# @ref CLWeightsReshapeKernel (executed only once for each configuration)
+ * -# @ref CLGEMMTranspose1xWKernel (executed only once for each configuration)
  * -# @ref CLIm2ColKernel
  * -# @ref CLGEMMInterleave4x4Kernel
  * -# @ref CLGEMMMatrixMultiplyKernel
diff --git a/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
index b4e469196e..5f4f1ba1d7 100644
--- a/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
@@ -39,7 +39,7 @@ class ICLTensor;
 
 /** Basic function to compute the locally connected layer. This function calls the following OpenCL kernels:
  *
- * -# @ref CLLocallyConnectedLayerWeightsReshapeKernel (executed only once for each configuration)
+ * -# @ref CLWeightsReshapeKernel (executed only once for each configuration)
  * -# @ref CLIm2ColKernel
  * -# @ref CLLocallyConnectedMatrixMultiplyKernel
  * -# @ref CLCol2ImKernel
@@ -66,14 +66,14 @@ public:
     void run() override;
 
 private:
-    CLIm2ColKernel                              _input_im2col_kernel;
-    CLLocallyConnectedLayerWeightsReshapeKernel _weights_reshape_kernel;
-    CLLocallyConnectedMatrixMultiplyKernel      _mm_kernel;
-    CLCol2ImKernel                              _output_col2im_kernel;
-    CLTensor                                    _input_im2col_reshaped;
-    CLTensor                                    _weights_reshaped;
-    CLTensor                                    _gemm_output;
-    bool                                        _is_first_run;
+    CLIm2ColKernel                         _input_im2col_kernel;
+    CLWeightsReshapeKernel                 _weights_reshape_kernel;
+    CLLocallyConnectedMatrixMultiplyKernel _mm_kernel;
+    CLCol2ImKernel                         _output_col2im_kernel;
+    CLTensor                               _input_im2col_reshaped;
+    CLTensor                               _weights_reshaped;
+    CLTensor                               _gemm_output;
+    bool                                   _is_first_run;
 };
 }
 #endif /* __ARM_COMPUTE_CLLOCALLYCONNECTEDLAYER_H__ */