commit 9be0c5a71cd7c1f08b65c48e53d083c59dd093a3
author Georgios Pinitas <georgios.pinitas@arm.com> Mon Feb 19 12:46:29 2018 +0000
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:47:18 2018 +0000
tree f9029878deaa3714dbded630bf3c48eb9af06623
parent 287b570b86ba40a801136aded140b83435ca9314

COMPMID-936: Convolution failure in NEON Convolution Layer.

Change-Id: I68a98eff57c8db719a501b68541666e8bc5f2081
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/121180
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp

diff --git a/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp b/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp
index 92383d9..dad4fee 100644
--- a/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp
+++ b/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp
@@ -272,8 +272,8 @@
                                     -conv_pad_top,
                                     (num_x_steps - 1) * input_num_elems_processed + num_elems_read_per_iteration,
                                     _input->info()->tensor_shape().y() + conv_pad_bottom);
-    AccessWindowStatic weights_access(_weights->info(), 0, 0, _weights->info()->dimension(0), _weights->info()->dimension(1));
-    AccessWindowStatic output_access(_output->info(), 0, 0, num_x_steps * _num_elems_written_per_iteration, output_shape.y());
+    AccessWindowStatic     weights_access(_weights->info(), 0, 0, _weights->info()->dimension(0), _weights->info()->dimension(1));
+    AccessWindowHorizontal output_access(_output->info(), 0, _num_elems_written_per_iteration);
 
     update_window_and_padding(win, input_access, weights_access, output_access);
     output_access.set_valid_region(win, ValidRegion(Coordinates(), _output->info()->tensor_shape()));

src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp

diff --git a/src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp b/src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp
index 3f33c43..08d8f8c 100644
--- a/src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp
+++ b/src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp
@@ -451,8 +451,10 @@
             break;
         }
         case DataType::S32:
+        {
             _func = (bias == nullptr) ? &output_stage<int32_t, uint8_t, false, false> : &output_stage<int32_t, uint8_t, false, true>;
             break;
+        }
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
         case DataType::F16:
         {