Move NEGEMMAssemblyDispatch as an internal function

Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I89ee26c1595d510c5048904cae9422528b76cd45
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4662
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

5 files changed, 18 insertions, 138 deletions

diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 5ac94102fc..f35144481d 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -75,7 +75,6 @@
 #include "arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEFuseBatchNormalization.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMM.h"
-#include "arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMMConv2d.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h"
diff --git a/arm_compute/runtime/NEON/functions/NEGEMM.h b/arm_compute/runtime/NEON/functions/NEGEMM.h
index 645ab56417..124f027227 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMM.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMM.h
@@ -30,17 +30,19 @@
 #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/NEON/functions/NEGEMMAssemblyDispatch.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <memory>
 
 namespace arm_compute
 {
+// Forward declarations
 class NEGEMMInterleave4x4Kernel;
 class NEGEMMMatrixAdditionKernel;
 class NEGEMMMatrixMultiplyKernel;
 class NEGEMMTranspose1xWKernel;
+class NEGEMMAssemblyDispatch;
+
 /** Basic function to execute GEMM on NEON. This function calls the following NEON kernels:
  *
  * If optimized assembly is available:
@@ -112,7 +114,7 @@ private:
     std::unique_ptr<NEGEMMInterleave4x4Kernel>  _interleave_kernel;
     std::unique_ptr<NEGEMMTranspose1xWKernel>   _transpose_kernel;
     std::unique_ptr<NEGEMMMatrixMultiplyKernel> _mm_kernel;
-    NEGEMMAssemblyDispatch                      _asm_glue;
+    std::unique_ptr<NEGEMMAssemblyDispatch>     _asm_glue;
     std::unique_ptr<NEGEMMMatrixAdditionKernel> _ma_kernel;
     NEActivationLayer                           _alpha_scale_func;
     NEArithmeticAddition                        _add_bias;
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h b/arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h
deleted file mode 100644
index 8f9498d0f5..0000000000
--- a/arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h
+++ /dev/null
@@ -1,125 +0,0 @@
-/*
- * Copyright (c) 2018-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_NEGEMMASSEMBLYDISPATCH_H
-#define ARM_COMPUTE_NEGEMMASSEMBLYDISPATCH_H
-
-#include "arm_compute/runtime/IFunction.h"
-#include "arm_compute/runtime/IMemoryManager.h"
-#include "arm_compute/runtime/IWeightsManager.h"
-#include "arm_compute/runtime/MemoryGroup.h"
-#include "arm_compute/runtime/Tensor.h"
-
-namespace arm_compute
-{
-/* Convolution method supported by the assembly gemm interface */
-enum class AsmConvMethod
-{
-    Im2Col,
-    Indirect,
-    Conv
-};
-
-struct AsmGemmInfo
-{
-    AsmConvMethod           method{ AsmConvMethod::Im2Col };
-    PadStrideInfo           ps_info{};
-    ActivationLayerInfo     activation_info{};
-    GEMMLowpOutputStageInfo output_stage{};
-    bool                    negated_offsets{ true };
-    bool                    reinterpret_input_as_3d{ false };
-    bool                    depth_output_gemm3d{ false };
-    int64_t                 padding_top{ 0 };
-    int64_t                 padding_left{ 0 };
-    float                   padding_value{ 0.f };
-};
-
-/** Assembly kernel glue */
-class NEGEMMAssemblyDispatch : public IFunction
-{
-public:
-    /** Constructor */
-    NEGEMMAssemblyDispatch(std::shared_ptr<IMemoryManager> memory_manager = nullptr, IWeightsManager *weights_manager = nullptr);
-    /** Prevent instances of this class from being copy constructed */
-    NEGEMMAssemblyDispatch(const NEGEMMAssemblyDispatch &) = delete;
-    /** Prevent instances of this class from being copied */
-    NEGEMMAssemblyDispatch &operator=(const NEGEMMAssemblyDispatch &) = delete;
-    NEGEMMAssemblyDispatch(NEGEMMAssemblyDispatch &&)                 = default;
-    NEGEMMAssemblyDispatch &operator=(NEGEMMAssemblyDispatch &&) = default;
-    ~NEGEMMAssemblyDispatch()                                    = default;
-
-    class IFallback
-    {
-    public:
-        virtual void run()                 = 0;
-        virtual void prepare()             = 0;
-        virtual bool is_configured() const = 0;
-        virtual ~IFallback()               = default;
-    };
-
-public:
-    /** If supported create a Compute Library function else fallback to the arm_gemm function.
-     *
-     * @param[in]  a    Input tensor (Matrix A)
-     * @param[in]  b    Input tensor (Matrix B)
-     * @param[in]  c    Input tensor (Matrix C) used to pass the bias for quantized calculations
-     * @param[out] d    Output tensor to store the result of matrix multiplication. Data type supported: same as @p input0.
-     * @param[in]  info GEMM meta-data
-     */
-    void configure(const ITensor *a, const ITensor *b, const ITensor *c, ITensor *d, const AsmGemmInfo &info);
-
-    /** Indicates whether or not this function can be used to process the given parameters.
-     *
-     * @param[in] a    Input tensor info (Matrix A)
-     * @param[in] b    Input tensor info (Matrix B)
-     * @param[in] c    Input tensor info (Matrix C) used to pass the bias for quantized calculations
-     * @param[in] d    Output tensor to store the result of matrix multiplication. Data type supported: same as @p input0.
-     * @param[in] info GEMM meta-data
-     *
-     * @return a status.
-     */
-    static Status validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *d, const AsmGemmInfo &info);
-    /** Checks if activation is supported by the gemm assembly dispatcher
-     *
-     * @param[in] activation Activation to check
-     *
-     * @return True if activation is supported else false
-     */
-    static bool is_activation_supported(const ActivationLayerInfo &activation);
-    /** Was the function successfully configured ?
-     *
-     * @return True if the function is configured and ready to run
-     */
-    bool is_configured() const;
-
-    // Inherited methods overridden:
-    void prepare() override;
-    void run() override;
-
-private:
-    std::unique_ptr<IFallback> _arm_gemm;        /** Interface for the arm_gemm fallback */
-    MemoryGroup                _memory_group;    /**< Function memory group */
-    IWeightsManager           *_weights_manager; /**< Pointer to the weights manager */
-};
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_NEGEMMASSEMBLYDISPATCH_H */
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMConv2d.h b/arm_compute/runtime/NEON/functions/NEGEMMConv2d.h
index 7cae39397f..2b3c162eab 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMConv2d.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMConv2d.h
@@ -28,7 +28,6 @@
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/NEON/functions/NEActivationLayer.h"
-#include "arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h"
 #include "arm_compute/runtime/NEON/functions/NEPermute.h"
 #include "arm_compute/runtime/Tensor.h"
 
@@ -37,6 +36,8 @@ namespace arm_compute
 {
 // Forward declarations
 class ITensor;
+class NEGEMMAssemblyDispatch;
+
 /** Basic function to compute the convolution layer. This function calls the following NEON kernels/functions:
  *
  * Supports only NHWC data layout
@@ -60,6 +61,8 @@ public:
     NEGEMMConv2d &operator=(const NEGEMMConv2d &) = delete;
     /** Default move assignment operator */
     NEGEMMConv2d &operator=(NEGEMMConv2d &&) = default;
+    /** Destructor */
+    ~NEGEMMConv2d();
     /** Set the input and output tensors.
      *
      * @param[in]  input   Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
@@ -96,13 +99,13 @@ public:
     void prepare() override;
 
 private:
-    NEGEMMAssemblyDispatch _gemm_asm_func;
-    NEActivationLayer      _activation_func;
-    NEPermute              _weights_permute_func;
-    const ITensor         *_original_weights;
-    Tensor                 _permuted_weights;
-    bool                   _is_prepared;
-    bool                   _run_activation;
+    std::unique_ptr<NEGEMMAssemblyDispatch> _gemm_asm_func;
+    NEActivationLayer                       _activation_func;
+    NEPermute                               _weights_permute_func;
+    const ITensor                          *_original_weights;
+    Tensor                                  _permuted_weights;
+    bool                                    _is_prepared;
+    bool                                    _run_activation;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_NEGEMMCONV2D_H */
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
index cb1d6bd782..8eea9d7d24 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
@@ -27,8 +27,8 @@
 #include "NEActivationLayer.h"
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/IWeightsManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
-#include "arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <memory>
@@ -45,6 +45,7 @@ class NEGEMMLowpOffsetContributionOutputStageKernel;
 class NEGEMMLowpMatrixAReductionKernel;
 class NEGEMMLowpMatrixBReductionKernel;
 class NEGEMMTranspose1xWKernel;
+class NEGEMMAssemblyDispatch;
 
 /** Basic function to execute GEMMLowpMatrixMultiplyCore on NEON. This function calls the following NEON kernels if the DOT product instruction is not available:
  *
@@ -115,7 +116,7 @@ public:
 private:
     MemoryGroup                                                    _memory_group;
     IWeightsManager                                               *_weights_manager;
-    NEGEMMAssemblyDispatch                                         _asm_glue;
+    std::unique_ptr<NEGEMMAssemblyDispatch>                        _asm_glue;
     std::unique_ptr<NEGEMMLowpMatrixMultiplyKernel>                _mm_kernel;
     std::unique_ptr<NEGEMMInterleave4x4Kernel>                     _mtx_a_reshape_kernel;
     std::unique_ptr<NEGEMMTranspose1xWKernel>                      _mtx_b_reshape_kernel;