Port ClTranspose to new API

Partially Resolves: COMPMID-4277 (1/2)

Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I704c2303135cbe1ba46d2fd5642c84c562204bc7
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5194
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

4 files changed, 142 insertions, 99 deletions

diff --git a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
index 3f17e4a921..a640e344d4 100644
--- a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -36,15 +36,27 @@
 
 namespace arm_compute
 {
-/** Basic function to reshape the weights of Fully Connected layer with OpenCL. This function calls the following kernels:
+/** Function to reshape the weights of Fully Connected layer with OpenCL by transposing input tensor. This function calls the following kernel:
  *
- *  -# @ref CLTransposeKernel
+ *  -# @ref opencl::kernels::ClTransposeKernel
  *
  * @note  The fully connected layer accepts "weights" tensors only with 2 dimensions.
  */
-class CLFullyConnectedLayerReshapeWeights : public ICLSimpleFunction
+class CLFullyConnectedLayerReshapeWeights : public IFunction
 {
 public:
+    /** Constructor */
+    CLFullyConnectedLayerReshapeWeights();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLFullyConnectedLayerReshapeWeights(const CLFullyConnectedLayerReshapeWeights &) = delete;
+    /** Default move constructor */
+    CLFullyConnectedLayerReshapeWeights(CLFullyConnectedLayerReshapeWeights &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLFullyConnectedLayerReshapeWeights &operator=(const CLFullyConnectedLayerReshapeWeights &) = delete;
+    /** Default move assignment operator */
+    CLFullyConnectedLayerReshapeWeights &operator=(CLFullyConnectedLayerReshapeWeights &&) = default;
+    /** Default destructor */
+    ~CLFullyConnectedLayerReshapeWeights();
     /** Set the input and output tensors.
      *
      * @param[in]  input  Weights tensor. The weights must be 2 dimensional. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
@@ -66,6 +78,13 @@ public:
      * @return a status
      */
     static Status validate(const ITensorInfo *input, const ITensorInfo *output);
+
+    // Inherited methods overridden
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 
 namespace weights_transformations
diff --git a/arm_compute/runtime/CL/functions/CLLSTMLayer.h b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
index 20b068316c..38a24d030b 100644
--- a/arm_compute/runtime/CL/functions/CLLSTMLayer.h
+++ b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
@@ -46,8 +46,14 @@
 namespace arm_compute
 {
 class CLCompileContext;
-class CLTransposeKernel;
 class ICLTensor;
+namespace opencl
+{
+namespace kernels
+{
+class ClTransposeKernel;
+}
+}
 
 /** This function performs a single time step in a Long Short-Term Memory (LSTM) layer.
  *
@@ -212,90 +218,91 @@ public:
     void prepare() override;
 
 private:
-    MemoryGroup                        _memory_group;
-    CLFullyConnectedLayer              _fully_connected_input_gate;
-    CLArithmeticAddition               _accum_input_gate1;
-    CLArithmeticSubtraction            _subtract_input_gate;
-    CLPixelWiseMultiplication          _pixelwise_mul_input_gate;
-    CLActivationLayer                  _activation_input_gate;
-    CLFullyConnectedLayer              _fully_connected_forget_gate;
-    CLArithmeticAddition               _accum_forget_gate1;
-    CLPixelWiseMultiplication          _pixelwise_mul_forget_gate;
-    CLActivationLayer                  _activation_forget_gate;
-    CLFullyConnectedLayer              _fully_connected_cell_state;
-    CLGEMM                             _gemm_cell_state1;
-    std::unique_ptr<CLTransposeKernel> _transpose_cell_state;
-    CLArithmeticAddition               _accum_cell_state1;
-    CLArithmeticAddition               _accum_cell_state2;
-    CLPixelWiseMultiplication          _pixelwise_mul_cell_state1;
-    CLActivationLayer                  _activation_cell_state;
-    CLActivationLayer                  _cell_clip;
-    CLPixelWiseMultiplication          _pixelwise_mul_cell_state2;
-    CLFullyConnectedLayer              _fully_connected_output;
-    CLPixelWiseMultiplication          _pixelwise_mul_output_state1;
-    CLArithmeticAddition               _accum_output1;
-    CLActivationLayer                  _activation_output;
-    CLActivationLayer                  _activation_output_state;
-    CLPixelWiseMultiplication          _pixelwise_mul_output_state2;
-    CLFullyConnectedLayer              _fully_connected_output_state;
-    CLActivationLayer                  _projection_clip;
-    CLCopy                             _copy_cell_state;
-    CLCopy                             _copy_output;
-    CLConcatenateLayer                 _concat_scratch_buffer;
-    CLConcatenateLayer                 _concat_inputs_forget_gate;
-    CLConcatenateLayer                 _concat_weights_forget_gate;
-    CLConcatenateLayer                 _concat_weights_input_gate;
-    CLConcatenateLayer                 _concat_weights_output;
-    CLFill                             _ones_fill;
-    CLMeanStdDevNormalizationLayer     _mean_std_norm_input_gate;
-    CLPixelWiseMultiplication          _pixelwise_mul_input_gate_coeff;
-    CLArithmeticAddition               _accum_input_gate_bias;
-    CLMeanStdDevNormalizationLayer     _mean_std_norm_forget_gate;
-    CLPixelWiseMultiplication          _pixelwise_mul_forget_gate_coeff;
-    CLArithmeticAddition               _accum_forget_gate_bias;
-    CLMeanStdDevNormalizationLayer     _mean_std_norm_cell_gate;
-    CLPixelWiseMultiplication          _pixelwise_mul_cell_gate_coeff;
-    CLArithmeticAddition               _accum_cell_gate_bias;
-    CLMeanStdDevNormalizationLayer     _mean_std_norm_output_gate;
-    CLPixelWiseMultiplication          _pixelwise_mul_output_gate_coeff;
-    CLArithmeticAddition               _accum_output_gate_bias;
-    CLTensor                           _input_gate_out1;
-    CLTensor                           _input_gate_out2;
-    CLTensor                           _input_gate_out3;
-    CLTensor                           _input_gate_out4;
-    CLTensor                           _forget_gate_out1;
-    CLTensor                           _forget_gate_out2;
-    CLTensor                           _forget_gate_out3;
-    CLTensor                           _forget_gate_out4;
-    CLTensor                           _forget_gate_out5;
-    CLTensor                           _forget_gate_out6;
-    CLTensor                           _cell_state_out1;
-    CLTensor                           _cell_state_out2;
-    CLTensor                           _cell_state_out3;
-    CLTensor                           _cell_state_out4;
-    CLTensor                           _cell_state_out5;
-    CLTensor                           _output1;
-    CLTensor                           _output2;
-    CLTensor                           _output3;
-    CLTensor                           _output4;
-    CLTensor                           _cell_state_activation;
-    CLTensor                           _output_state1;
-    CLTensor                           _ones;
-    CLTensor                           _input_layer_norm_out1;
-    CLTensor                           _input_layer_norm_out2;
-    CLTensor                           _forget_layer_norm_out1;
-    CLTensor                           _forget_layer_norm_out2;
-    CLTensor                           _cell_layer_norm_out1;
-    CLTensor                           _cell_layer_norm_out2;
-    CLTensor                           _output_layer_norm_out1;
-    CLTensor                           _output_layer_norm_out2;
-    bool                               _run_peephole_opt;
-    bool                               _run_cifg_opt;
-    bool                               _perform_cell_clipping;
-    bool                               _has_projection_weights;
-    bool                               _perform_projection_clipping;
-    bool                               _is_prepared;
-    bool                               _is_layer_norm_lstm;
+    MemoryGroup                                         _memory_group;
+    CLFullyConnectedLayer                               _fully_connected_input_gate;
+    CLArithmeticAddition                                _accum_input_gate1;
+    CLArithmeticSubtraction                             _subtract_input_gate;
+    CLPixelWiseMultiplication                           _pixelwise_mul_input_gate;
+    CLActivationLayer                                   _activation_input_gate;
+    CLFullyConnectedLayer                               _fully_connected_forget_gate;
+    CLArithmeticAddition                                _accum_forget_gate1;
+    CLPixelWiseMultiplication                           _pixelwise_mul_forget_gate;
+    CLActivationLayer                                   _activation_forget_gate;
+    CLFullyConnectedLayer                               _fully_connected_cell_state;
+    CLGEMM                                              _gemm_cell_state1;
+    std::unique_ptr<opencl::kernels::ClTransposeKernel> _transpose_cell_state;
+    CLArithmeticAddition                                _accum_cell_state1;
+    CLArithmeticAddition                                _accum_cell_state2;
+    CLPixelWiseMultiplication                           _pixelwise_mul_cell_state1;
+    CLActivationLayer                                   _activation_cell_state;
+    CLActivationLayer                                   _cell_clip;
+    CLPixelWiseMultiplication                           _pixelwise_mul_cell_state2;
+    CLFullyConnectedLayer                               _fully_connected_output;
+    CLPixelWiseMultiplication                           _pixelwise_mul_output_state1;
+    CLArithmeticAddition                                _accum_output1;
+    CLActivationLayer                                   _activation_output;
+    CLActivationLayer                                   _activation_output_state;
+    CLPixelWiseMultiplication                           _pixelwise_mul_output_state2;
+    CLFullyConnectedLayer                               _fully_connected_output_state;
+    CLActivationLayer                                   _projection_clip;
+    CLCopy                                              _copy_cell_state;
+    CLCopy                                              _copy_output;
+    CLConcatenateLayer                                  _concat_scratch_buffer;
+    CLConcatenateLayer                                  _concat_inputs_forget_gate;
+    CLConcatenateLayer                                  _concat_weights_forget_gate;
+    CLConcatenateLayer                                  _concat_weights_input_gate;
+    CLConcatenateLayer                                  _concat_weights_output;
+    CLFill                                              _ones_fill;
+    CLMeanStdDevNormalizationLayer                      _mean_std_norm_input_gate;
+    CLPixelWiseMultiplication                           _pixelwise_mul_input_gate_coeff;
+    CLArithmeticAddition                                _accum_input_gate_bias;
+    CLMeanStdDevNormalizationLayer                      _mean_std_norm_forget_gate;
+    CLPixelWiseMultiplication                           _pixelwise_mul_forget_gate_coeff;
+    CLArithmeticAddition                                _accum_forget_gate_bias;
+    CLMeanStdDevNormalizationLayer                      _mean_std_norm_cell_gate;
+    CLPixelWiseMultiplication                           _pixelwise_mul_cell_gate_coeff;
+    CLArithmeticAddition                                _accum_cell_gate_bias;
+    CLMeanStdDevNormalizationLayer                      _mean_std_norm_output_gate;
+    CLPixelWiseMultiplication                           _pixelwise_mul_output_gate_coeff;
+    CLArithmeticAddition                                _accum_output_gate_bias;
+    CLTensor                                            _input_gate_out1;
+    CLTensor                                            _input_gate_out2;
+    CLTensor                                            _input_gate_out3;
+    CLTensor                                            _input_gate_out4;
+    CLTensor                                            _forget_gate_out1;
+    CLTensor                                            _forget_gate_out2;
+    CLTensor                                            _forget_gate_out3;
+    CLTensor                                            _forget_gate_out4;
+    CLTensor                                            _forget_gate_out5;
+    CLTensor                                            _forget_gate_out6;
+    CLTensor                                            _cell_state_out1;
+    CLTensor                                            _cell_state_out2;
+    CLTensor                                            _cell_state_out3;
+    CLTensor                                            _cell_state_out4;
+    CLTensor                                            _cell_state_out5;
+    CLTensor                                            _output1;
+    CLTensor                                            _output2;
+    CLTensor                                            _output3;
+    CLTensor                                            _output4;
+    CLTensor                                            _cell_state_activation;
+    CLTensor                                            _output_state1;
+    CLTensor                                            _ones;
+    CLTensor                                            _input_layer_norm_out1;
+    CLTensor                                            _input_layer_norm_out2;
+    CLTensor                                            _forget_layer_norm_out1;
+    CLTensor                                            _forget_layer_norm_out2;
+    CLTensor                                            _cell_layer_norm_out1;
+    CLTensor                                            _cell_layer_norm_out2;
+    CLTensor                                            _output_layer_norm_out1;
+    CLTensor                                            _output_layer_norm_out2;
+    bool                                                _run_peephole_opt;
+    bool                                                _run_cifg_opt;
+    bool                                                _perform_cell_clipping;
+    bool                                                _has_projection_weights;
+    bool                                                _perform_projection_clipping;
+    bool                                                _is_prepared;
+    bool                                                _is_layer_norm_lstm;
+    const ICLTensor                                    *_recurrent_to_cell_weights{ nullptr };
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLLSTMLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLPermute.h b/arm_compute/runtime/CL/functions/CLPermute.h
index bcd9566fbf..e59cca2a67 100644
--- a/arm_compute/runtime/CL/functions/CLPermute.h
+++ b/arm_compute/runtime/CL/functions/CLPermute.h
@@ -46,11 +46,11 @@ public:
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     CLPermute(const CLPermute &) = delete;
     /** Default move constructor */
-    CLPermute(CLPermute &&);
+    CLPermute(CLPermute &&) = default;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     CLPermute &operator=(const CLPermute &) = delete;
     /** Default move assignment operator */
-    CLPermute &operator=(CLPermute &&);
+    CLPermute &operator=(CLPermute &&) = default;
     /** Set the input and output tensors.
      *
      * @note Arbitrary permutation vectors are supported with rank not greater than 4
diff --git a/arm_compute/runtime/CL/functions/CLTranspose.h b/arm_compute/runtime/CL/functions/CLTranspose.h
index 2b7a03f23f..43cebeba90 100644
--- a/arm_compute/runtime/CL/functions/CLTranspose.h
+++ b/arm_compute/runtime/CL/functions/CLTranspose.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,7 +25,9 @@
 #define ARM_COMPUTE_CLTRANSPOSE_H
 
 #include "arm_compute/core/Error.h"
-#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
+#include "arm_compute/runtime/IFunction.h"
+
+#include <memory>
 
 namespace arm_compute
 {
@@ -33,14 +35,22 @@ class CLCompileContext;
 class ICLTensor;
 class ITensorInfo;
 
-/** Basic function to transpose a matrix on OpenCL. This function calls the following OpenCL kernel:
- *
- *  -# @ref CLTransposeKernel
- *
- */
-class CLTranspose : public ICLSimpleFunction
+/** Basic function to execute an @ref opencl::kernels::ClTransposeKernel. */
+class CLTranspose : public IFunction
 {
 public:
+    /** Constructor */
+    CLTranspose();
+    /** Destructor */
+    ~CLTranspose();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLTranspose(const CLTranspose &) = delete;
+    /** Default move constructor */
+    CLTranspose(CLTranspose &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLTranspose &operator=(const CLTranspose &) = delete;
+    /** Default move assignment operator */
+    CLTranspose &operator=(CLTranspose &&) = default;
     /** Initialise the kernel's inputs and output
      *
      * @param[in]  input  Input tensor. Data types supported: All.
@@ -62,6 +72,13 @@ public:
      * @return a status
      */
     static Status validate(const ITensorInfo *input, const ITensorInfo *output);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 }