commit 2e5fd637205770ec5e11096e6e19b8efc67d544e
author SiCongLi <sicong.li@arm.com> Mon Mar 02 15:39:15 2020 +0000
committer SiCong Li <sicong.li@arm.com> Wed Mar 25 11:23:19 2020 +0000
tree c7f5b53d6ec54cf39fcc638bad9d054c3e04bf43
parent 51a9558d484e812707a776efceef18b93698cec9

COMPMID-3098 Fuse Relu and Bounded Relu with FullyConnected NEON

Change-Id: Id28062445590d6c06b35f7d7434eb38393ae94a7
Signed-off-by: SiCongLi <sicong.li@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2875
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

arm_compute/core/Types.h

diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index b640987..711b68f 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -2140,6 +2140,14 @@
     {
         return _activation_info;
     }
+    /** Set activation layer info
+     *
+     * @param[in] activation_info ActivationLayerInfo object to set
+     */
+    void set_activation_info(const ActivationLayerInfo &activation_info)
+    {
+        _activation_info = activation_info;
+    }
 
 private:
     bool                    _is_a_reshaped;

arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h

diff --git a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
index db09da4..b14650c 100644
--- a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
@@ -162,9 +162,9 @@
     void prepare() override;
 
 private:
-    void configure_fc_fc(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output);
-    void configure_conv_fc(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output);
-    void configure_mm(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output);
+    void configure_fc_fc(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const ActivationLayerInfo &act);
+    void configure_conv_fc(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const ActivationLayerInfo &act);
+    void configure_mm(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const ActivationLayerInfo &act);
 
     MemoryGroup                                                         _memory_group;
     IWeightsManager                                                    *_weights_manager;
@@ -182,7 +182,7 @@
     bool                                                                _are_weights_converted;
     bool                                                                _are_weights_reshaped;
     bool                                                                _is_fc_after_conv;
-    bool                                                                _is_quantized;
+    bool                                                                _is_quantized_asymmetric;
     bool                                                                _is_prepared;
 };
 } // namespace arm_compute

arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h

diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
index 508159e..74dedcf 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,9 +28,12 @@
 #include "arm_compute/core/NEON/INEKernel.h"
 #include "arm_compute/core/NEON/kernels/NEConvertQuantizedSignednessKernel.h"
 #include "arm_compute/core/NEON/kernels/NEConvertQuantizedSignednessKernel.h"
+#include "arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h"
+#include "arm_compute/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionOutputStageKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMLowpReductionKernel.h"
+#include "arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h"
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
@@ -60,7 +63,7 @@
 {
 public:
     /** Constructor */
-    NEGEMMLowpMatrixMultiplyCore(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    NEGEMMLowpMatrixMultiplyCore(std::shared_ptr<IMemoryManager> memory_manager = nullptr, IWeightsManager *weights_manager = nullptr);
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     NEGEMMLowpMatrixMultiplyCore(const NEGEMMLowpMatrixMultiplyCore &) = delete;
     /** Default move constructor */
@@ -109,10 +112,11 @@
 
 private:
     MemoryGroup                                   _memory_group;
+    IWeightsManager                              *_weights_manager;
     NEGEMMAssemblyDispatch                        _asm_glue;
-    std::unique_ptr<INEKernel>                    _mm_kernel;
-    std::unique_ptr<INEKernel>                    _mtx_a_reshape_kernel;
-    std::unique_ptr<INEKernel>                    _mtx_b_reshape_kernel;
+    NEGEMMLowpMatrixMultiplyKernel                _mm_kernel;
+    NEGEMMInterleave4x4Kernel                     _mtx_a_reshape_kernel;
+    NEGEMMTranspose1xWKernel                      _mtx_b_reshape_kernel;
     NEGEMMLowpMatrixAReductionKernel              _mtx_a_reduction_kernel;
     NEGEMMLowpMatrixBReductionKernel              _mtx_b_reduction_kernel;
     NEGEMMLowpOffsetContributionKernel            _offset_contribution_kernel;