COMPMID-1043: Rework GCGEMMMatrixMultiplyKernel interface and allow auto initialization of the tensors

This patch also:
- removes support for already reshaped weights in GCConvolutionLayer
- makes GCConvolutionLayer similar to CLGEMMConvolutionLayer
- enables usage of the GCGEMM function in GCConvolution instead of calling the
  GEMM kernels directly

Change-Id: I3e4a64335555e86e18585d38d8fda4bfdb44e265
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/127696
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

1 files changed, 51 insertions, 23 deletions

diff --git a/arm_compute/runtime/GLES_COMPUTE/functions/GCConvolutionLayer.h b/arm_compute/runtime/GLES_COMPUTE/functions/GCConvolutionLayer.h
index 54b17b40bb..fa29f447c8 100644
--- a/arm_compute/runtime/GLES_COMPUTE/functions/GCConvolutionLayer.h
+++ b/arm_compute/runtime/GLES_COMPUTE/functions/GCConvolutionLayer.h
@@ -27,15 +27,13 @@
 
 #include "arm_compute/core/GLES_COMPUTE/kernels/GCCol2ImKernel.h"
 #include "arm_compute/core/GLES_COMPUTE/kernels/GCFillBorderKernel.h"
-#include "arm_compute/core/GLES_COMPUTE/kernels/GCGEMMInterleave4x4Kernel.h"
-#include "arm_compute/core/GLES_COMPUTE/kernels/GCGEMMMatrixMultiplyKernel.h"
-#include "arm_compute/core/GLES_COMPUTE/kernels/GCGEMMTranspose1xWKernel.h"
 #include "arm_compute/core/GLES_COMPUTE/kernels/GCIm2ColKernel.h"
 #include "arm_compute/core/GLES_COMPUTE/kernels/GCWeightsReshapeKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/GLES_COMPUTE/GCMemoryGroup.h"
 #include "arm_compute/runtime/GLES_COMPUTE/GCTensor.h"
 #include "arm_compute/runtime/GLES_COMPUTE/functions/GCActivationLayer.h"
+#include "arm_compute/runtime/GLES_COMPUTE/functions/GCGEMM.h"
 #include "arm_compute/runtime/IFunction.h"
 
 #include <memory>
@@ -46,7 +44,6 @@ class IGCTensor;
 
 /** Function to reshape and transpose the weights. This function calls the following kernels:
  * -# @ref GCWeightsReshapeKernel
- * -# @ref GCGEMMTranspose1xWKernel
  */
 class GCConvolutionLayerReshapeWeights : public IFunction
 {
@@ -55,22 +52,18 @@ public:
     GCConvolutionLayerReshapeWeights();
     /** 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: 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.
-     *                          Data types supported: Same as @p weights.
+     * @param[in]  weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM].
+     *                     Data type supported: 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.
      */
-    void configure(const IGCTensor *weights, const IGCTensor *biases, IGCTensor *output, bool transpose1xW);
+    void configure(const IGCTensor *weights, const IGCTensor *biases, IGCTensor *output);
     // Inherited methods overridden:
     void run() override;
 
 private:
-    GCWeightsReshapeKernel   _weights_reshape_kernel;
-    GCGEMMTranspose1xWKernel _weights_transposed_kernel;
-    GCTensor                 _weights_reshaped;
-    bool                     _transpose1xW;
+    GCWeightsReshapeKernel _weights_reshape_kernel;
+    GCTensor               _weights_reshaped;
 };
 
 /** Basic function to compute the convolution layer. This function calls the following GLES kernels:
@@ -86,7 +79,14 @@ class GCConvolutionLayer : public IFunction
 public:
     /** Default constructor */
     GCConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
-
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    GCConvolutionLayer(const GCConvolutionLayer &) = delete;
+    /** Default move constructor */
+    GCConvolutionLayer(GCConvolutionLayer &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    GCConvolutionLayer &operator=(const GCConvolutionLayer &) = delete;
+    /** Default move assignment operator */
+    GCConvolutionLayer &operator=(GCConvolutionLayer &&) = default;
     /** Set the input and output tensors.
      *
      * @param[in]  input        Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
@@ -105,6 +105,26 @@ public:
      */
     void configure(const IGCTensor *input, const IGCTensor *weights, const IGCTensor *biases, IGCTensor *output, const PadStrideInfo &conv_info,
                    const WeightsInfo &weights_info = WeightsInfo(), const Size2D &dilation = Size2D(1U, 1U), const ActivationLayerInfo &act_info = ActivationLayerInfo());
+    /** Static function to check if given info will lead to a valid configuration of @ref GCConvolutionLayer.
+     *
+     * @param[in]  input        Source tensor. 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: QS8/QASYMM8/QS16/F16/F32.
+     * @param[in]  weights      Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: Same as @p input.
+     * @param[in]  biases       Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
+     *                          Data type supported: Should match @p input data type, except for input of QASYMM8 type where biases should be of S32 type.
+     * @param[out] output       Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
+     *                          Data types supported: Same as @p input.
+     * @param[in]  conv_info    Contains padding and stride information described in @ref PadStrideInfo.
+     * @param[in]  weights_info Specifies if the weights tensor has been reshaped with GCWeightsReshapeKernel. If this is not part of the fully connected layer the weights
+     *                          tensor has also been transposed with GCGEMMTranspose1xWKernel. Data type supported: Same as @p input.
+     * @param[in]  dilation     (Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
+     * @param[in]  act_info     (Optional) Activation layer information in case of a fused activation.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
+                           const WeightsInfo &weights_info = WeightsInfo(), const Size2D &dilation = Size2D(1U, 1U), const ActivationLayerInfo &act_info = ActivationLayerInfo());
 
     // Inherited methods overridden:
     void run() override;
@@ -115,20 +135,30 @@ private:
      * @param input                     Input tensor. Data types supported: F16/F32.
      * @param weights                   Weights tensor. Data type supported: Same as @p input.
      * @param output                    Output tensor. Data types supported: Same as @p input,
-     * @param is_interleaved_transposed Flag that signals if matrix is interleaved transposed
      */
-    void configure_mm(const IGCTensor *input, const IGCTensor *weights, IGCTensor *output, bool is_interleaved_transposed = true);
+    void configure_mm(const IGCTensor *input, const IGCTensor *weights, IGCTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref GCGEMMConvolutionLayer matrix multiply routines
+     *
+     * @param[in] input   Input tensor. Data types supported: QS8/QASYMM8/QS16/F16/F32.
+     * @param[in] weights Weights tensor. Data type supported: Same as @p input.
+     * @param[in] output  Output tensor. Data types supported: Same as @p input,
+     *                                      except for input of QASYMM8 type where output should be of S32 type.
+     *
+     * @return a status
+     */
+    static Status validate_mm(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output);
 
 private:
     GCMemoryGroup                    _memory_group;
     GCConvolutionLayerReshapeWeights _reshape_weights;
     GCIm2ColKernel                   _input_im2col_kernel;
-    GCGEMMInterleave4x4Kernel        _input_interleave_kernel;
-    GCGEMMMatrixMultiplyKernel       _mm_kernel;
+    GCGEMM                           _mm_gemm;
     GCCol2ImKernel                   _output_col2im_kernel;
     GCFillBorderKernel               _fill_border;
     GCActivationLayer                _activationlayer_function;
 
+    const IGCTensor *_original_weights;
+
     GCTensor _input_im2col_reshaped;
     GCTensor _input_interleaved_reshaped;
     GCTensor _weights_reshaped;
@@ -136,9 +166,7 @@ private:
     GCTensor _gemm_output;
     GCTensor _tmp_output;
 
-    bool _append_bias;
-    bool _is_fully_connected_convolution;
-    bool _are_weights_reshaped;
+    bool _is_first_run;
     bool _is_activationlayer_enabled;
 };
 }