commit 215b4ea6c9dee480a22070d5873b0b8cb52531a0
author Gian Marco Iodice <gianmarco.iodice@arm.com> Thu Jun 28 16:29:29 2018 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:54:54 2018 +0000
tree 398e552c4d01c0b84d03a873098a9183ba8f82e4
parent ad486e21e5870f41774f30825c270762e08ae71e

COMPMID-1277 - Optimizing CLIm2ColKernel for NHWC.

This patch includes:

- Im2Col optimizations for NHWC using a new data layout
- Refactoring of CLIm2ColKernel adding validation method and auto-init
- Removed im2col_reduced from CLIm2ColKernel and created a new kernel CLFlattenLayerKernel

Change-Id: I1620640b6796baa268324b33ae92cdd8de53e27c
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/141241
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>

arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h

diff --git a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
index e8fe8e4..450cd83 100644
--- a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
@@ -27,11 +27,11 @@
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 #include "arm_compute/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.h"
-#include "arm_compute/core/CL/kernels/CLIm2ColKernel.h"
 #include "arm_compute/core/CL/kernels/CLTransposeKernel.h"
 #include "arm_compute/runtime/CL/CLMemoryGroup.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h"
+#include "arm_compute/runtime/CL/functions/CLFlattenLayer.h"
 #include "arm_compute/runtime/CL/functions/CLGEMM.h"
 #include "arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h"
 #include "arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h"
@@ -130,14 +130,14 @@
     void configure_mm(const ICLTensor *input, const ICLTensor *weights, ICLTensor *output);
 
     CLMemoryGroup                                       _memory_group;
-    CLIm2ColKernel                                      _im2col_kernel;
     CLConvertFullyConnectedWeights                      _convert_weights;
+    CLFlattenLayer                                      _flatten_layer;
     CLFullyConnectedLayerReshapeWeights                 _reshape_weights_kernel;
     CLGEMM                                              _mm_gemm;
     CLGEMMLowpMatrixMultiplyCore                        _mm_gemmlowp;
     CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint _gemmlowp_output_stage;
     CLGEMMMatrixAccumulateBiasesKernel                  _accumulate_biases_kernel;
-    CLTensor                                            _im2col_output;
+    CLTensor                                            _flatten_output;
     CLTensor                                            _gemmlowp_output;
     CLTensor                                            _converted_weights_output;
     CLTensor                                            _reshape_weights_output;