Port NEGEMM to memory injecting interface (Part 1)

- Start porting NEGEMM to the new API
- Port NEGEMMInterleave4x4Kernel to the new API
- Port NEGEMMMatrixAdditionKernel to the new API
- Port NEGEMMTranspose1xWKernel to the new API
- Remove padding from NEGEMMMatrixAdditionKernel
- Remove unused INESimpleKernel and ICPPSimpleKernel

Partially resolves: COMPMID-4402

Change-Id: I63edadddfe00a54586e5384d6a0211db25ae9042
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5857
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

6 files changed, 17 insertions, 56 deletions

diff --git a/arm_compute/runtime/NEON/functions/NEConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEConvolutionLayer.h
index f19aa8008b..bb4c456787 100644
--- a/arm_compute/runtime/NEON/functions/NEConvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEConvolutionLayer.h
@@ -111,7 +111,7 @@ public:
      *                              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 NEWeightsReshapeKernel. If this is not part of the fully connected layer the weights
-     *                              tensor has also been transposed with NEGEMMTranspose1xWKernel. Data type supported: Same as @p input.
+     *                              tensor has also been transposed with cpu::kernels::CpuGemmTranspose1xWKernel. 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. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU supported.
      * @param[in]  enable_fast_math (Optional) Enable fast math computation. In case this flag were set, the function could dispatch the fastest implementation
@@ -133,7 +133,7 @@ public:
      *                             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 NEWeightsReshapeKernel. If this is not part of the fully connected layer the weights
-     *                             tensor has also been transposed with NEGEMMTranspose1xWKernel. Data type supported: Same as @p input.
+     *                             tensor has also been transposed with cpu::kernels::CpuGemmTranspose1xWKernel. 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.
      * @param[in] enable_fast_math (Optional) Enable fast math computation. In case this flag were set, the function could dispatch the fastest implementation
@@ -156,7 +156,7 @@ public:
      *                             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 NEWeightsReshapeKernel. If this is not part of the fully connected layer the weights
-     *                             tensor has also been transposed with NEGEMMTranspose1xWKernel. Data type supported: Same as @p input.
+     *                             tensor has also been transposed with cpu::kernels::CpuGemmTranspose1xWKernel. 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.
      * @param[in] enable_fast_math (Optional) Enable fast math computation. In case this flag were set, the function could dispatch the fastest implementation
diff --git a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
index d2cd60e576..22ec9e0fec 100644
--- a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
@@ -80,7 +80,7 @@ private:
  *  -# @ref NEIm2ColKernel (called when the input comes from a convolutional layer)
  *  -# @ref NETranspose (if @p are_weights_reshaped is set to false and transpose_weights is set to true ) (called once)
  *  -# @ref NEGEMMMatrixMultiplyKernel or @ref NEGEMMLowpMatrixMultiplyCore (if quantized asymmetric)
- *  -# @ref NEGEMMMatrixAdditionKernel or @ref NEGEMMLowpOutputStage (if quantized asymmetric) (if @p biases is not equal to nullptr)
+ *  -# @ref cpu::kernels::CpuGemmMatrixAdditionKernel or @ref NEGEMMLowpOutputStage (if quantized asymmetric) (if @p biases is not equal to nullptr)
  *
  * @note  The fully connected layer accepts "weights" tensors only with 2 dimensions.
  */
diff --git a/arm_compute/runtime/NEON/functions/NEGEMM.h b/arm_compute/runtime/NEON/functions/NEGEMM.h
index 6c5be0eb5e..c1ae11bcbf 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMM.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMM.h
@@ -29,40 +29,26 @@
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/IWeightsManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
-#include "arm_compute/runtime/NEON/functions/NEActivationLayer.h"
-#include "arm_compute/runtime/NEON/functions/NEArithmeticAddition.h"
-#include "arm_compute/runtime/Tensor.h"
-#include "src/core/helpers/MemoryHelpers.h"
 
 #include <memory>
 
 namespace arm_compute
 {
-// Forward declarations
-class NEGEMMInterleave4x4Kernel;
-class NEGEMMMatrixAdditionKernel;
-class NEGEMMMatrixMultiplyKernel;
-class NEGEMMTranspose1xWKernel;
-namespace cpu
-{
-class CpuGemmAssemblyDispatch;
-}
-
 /** Basic function to execute GEMM. This function calls the following kernels:
  *
  * If optimized assembly is available:
  *  -# @ref cpu::CpuGemmAssemblyDispatch
- *  -# @ref NEActivationLayer (if alpha != 1.0)
+ *  -# @ref cpu::CpuActivation (if alpha != 1.0)
  * Else:
- *  -# @ref NEGEMMInterleave4x4Kernel (if the output tensor is a matrix)
- *  -# @ref NEGEMMTranspose1xWKernel (if the output tensor is a matrix)
+ *  -# @ref cpu::kernels::CpuGemmInterleave4x4Kernel (if the output tensor is a matrix)
+ *  -# @ref cpu::kernels::CpuGemmTranspose1xWKernel (if the output tensor is a matrix)
  *  -# @ref NEGEMMMatrixMultiplyKernel
  * In both cases:
- *  -# @ref NEGEMMMatrixAdditionKernel (if c != nullptr and beta != 0.0 and is not reshaped once)
+ *  -# @ref cpu::kernels::CpuGemmMatrixAdditionKernel (if c != nullptr and beta != 0.0 and is not reshaped once)
  * Else:
- *  -# @ref NEArithmeticAddition (if c != nullptr and is reshaped once and not optimized assembly in place)
+ *  -# @ref cpu::CpuAdd (if c != nullptr and is reshaped once and not optimized assembly in place)
  *
- *  -# @ref NEActivationLayer (if activation is specified in GEMMInfo)
+ *  -# @ref cpu::CpuActivation (if activation is specified in GEMMInfo)
  */
 class NEGEMM : public IFunction
 {
@@ -117,33 +103,8 @@ public:
     void prepare() override;
 
 private:
-    MemoryGroup                                   _memory_group;
-    IWeightsManager                              *_weights_manager;
-    std::unique_ptr<NEGEMMInterleave4x4Kernel>    _interleave_kernel;
-    std::unique_ptr<NEGEMMTranspose1xWKernel>     _transpose_kernel;
-    std::unique_ptr<NEGEMMMatrixMultiplyKernel>   _mm_kernel;
-    std::unique_ptr<cpu::CpuGemmAssemblyDispatch> _asm_glue;
-    std::unique_ptr<NEGEMMMatrixAdditionKernel>   _ma_kernel;
-    NEActivationLayer                             _alpha_scale_func;
-    NEArithmeticAddition                          _add_bias;
-    NEActivationLayer                             _activation_func;
-
-    Tensor         _tmp_a;
-    Tensor         _tmp_b;
-    Tensor         _tmp_d;
-    const ITensor *_original_b;
-    bool           _run_vector_matrix_multiplication;
-    bool           _run_alpha_scale;
-    bool           _run_addition;
-    bool           _run_bias_addition;
-    bool           _run_activation;
-    bool           _reshape_b_only_on_first_run;
-    bool           _is_prepared;
-
-    ITensorPack                      _asm_glue_run_pack;
-    ITensorPack                      _asm_glue_prep_pack;
-    WorkspaceData<Tensor>            _asm_glue_workspace;
-    experimental::MemoryRequirements _aux_mem_req;
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_NEGEMM_H */
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
index edb58e956a..d334d518e2 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
@@ -203,7 +203,7 @@ public:
      *                          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 NEWeightsReshapeKernel. If this is not part of the fully connected layer the weights
-     *                          tensor has also been transposed with NEGEMMTranspose1xWKernel. Data type supported: Same as @p input.
+     *                          tensor has also been transposed with cpu::kernels::CpuGemmTranspose1xWKernel. 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. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU supported.
      * @param[in]  num_groups   (Optional) Number of groups when performing a grouped convolution. num_groups != 1 is not supported
@@ -223,7 +223,7 @@ public:
      *                         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 NEWeightsReshapeKernel. If this is not part of the fully connected layer the weights
-     *                         tensor has also been transposed with NEGEMMTranspose1xWKernel. Data type supported: Same as @p input.
+     *                         tensor has also been transposed with cpu::kernels::CpuGemmTranspose1xWKernel. 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. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU supported.
      * @param[in] num_groups   (Optional) Number of groups when performing a grouped convolution. num_groups != 1 is not supported
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
index ff888760e1..60cfd8f91d 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
@@ -35,8 +35,8 @@ namespace arm_compute
 class ITensor;
 /** Basic function to execute GEMMLowpMatrixMultiplyCore. This function calls the following kernels if the DOT product instruction is not available:
  *
- *  -# @ref NEGEMMInterleave4x4Kernel
- *  -# @ref NEGEMMTranspose1xWKernel
+ *  -# @ref cpu::kernels::CpuGemmInterleave4x4Kernel
+ *  -# @ref cpu::kernels::CpuGemmTranspose1xWKernel
  *  -# @ref NEGEMMLowpMatrixMultiplyKernel
  *  -# @ref NEGEMMLowpOffsetContributionKernel
  *  -# @ref NEActivationLayer
diff --git a/arm_compute/runtime/NEON/functions/NELSTMLayer.h b/arm_compute/runtime/NEON/functions/NELSTMLayer.h
index 075fb4530a..4272215486 100644
--- a/arm_compute/runtime/NEON/functions/NELSTMLayer.h
+++ b/arm_compute/runtime/NEON/functions/NELSTMLayer.h
@@ -25,7 +25,7 @@
 #define ARM_COMPUTE_NELSTMLAYER_H
 
 #include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/NEON/functions/NEArithmeticAddition.h"
+#include "arm_compute/runtime/NEON/functions/NEActivationLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEArithmeticAddition.h"
 #include "arm_compute/runtime/NEON/functions/NEArithmeticSubtraction.h"
 #include "arm_compute/runtime/NEON/functions/NEConcatenateLayer.h"