commit 8b72199f25487040713d1668c998fdde3707413c
author Georgios Pinitas <georgios.pinitas@arm.com> Mon Oct 28 16:24:28 2019 +0000
committer Georgios Pinitas <georgios.pinitas@arm.com> Tue Oct 29 10:44:09 2019 +0000
tree a898e56a58d3fc83953ef365ee71371a9245d6f5
parent fde091434901fa3d7b504842efa373270a325c51

COMPMID-1889: Fuse bias addition and output stage in CLFCLayer.

Delegates bias addition and output stage calculation in
CLGEMMLowp/CLGEMM respectively.

Change-Id: Ifd1f75b34eae766d3df80f07aec35fab45733e0b
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2175
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>
Comments-Addressed: Arm Jenkins <bsgcomp@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 9512b22..7f87253 100644
--- a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
@@ -174,9 +174,9 @@
     void prepare() override;
 
 private:
-    void configure_fc_fc(const ICLTensor *input, const ICLTensor *weights, ICLTensor *output, bool retain_internal_weights);
-    void configure_conv_fc(const ICLTensor *input, const ICLTensor *weights, ICLTensor *output, bool retain_internal_weights);
-    void configure_mm(const ICLTensor *input, const ICLTensor *weights, ICLTensor *output, bool retain_internal_weights);
+    void configure_fc_fc(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, bool retain_internal_weights);
+    void configure_conv_fc(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, bool retain_internal_weights);
+    void configure_mm(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, bool retain_internal_weights);
 
     MemoryGroup                                                         _memory_group;
     IWeightsManager                                                    *_weights_manager;
@@ -187,16 +187,12 @@
     CLFullyConnectedLayerReshapeWeights                                 _reshape_weights_function;
     CLGEMM                                                              _mm_gemm;
     CLGEMMLowpMatrixMultiplyCore                                        _mm_gemmlowp;
-    CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint                 _gemmlowp_output_stage;
-    CLGEMMMatrixAccumulateBiasesKernel                                  _accumulate_biases_kernel; // TODO(COMPMID-1889): Use CLGEMM to add bias in CLFullyConnectedLayer
     CLTensor                                                            _flatten_output;
-    CLTensor                                                            _gemmlowp_output;
     CLTensor                                                            _converted_weights_output;
     CLTensor                                                            _reshape_weights_output;
     bool                                                                _are_weights_converted;
     bool                                                                _are_weights_reshaped;
     bool                                                                _is_fc_after_conv;
-    bool                                                                _accumulate_biases;
     bool                                                                _is_quantized;
     bool                                                                _is_prepared;
     const ICLTensor                                                    *_original_weights;