commit a437638028c216a03572593b9d4e3532df2a308d
author Anthony Barbier <anthony.barbier@arm.com> Wed Apr 12 15:12:46 2017 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Wed Apr 12 17:42:01 2017 +0100
tree b2aeef6797fdc620419431b4b3d3c4f46febf69e
parent 68a98dc29106a4c8c34e8cac542fe0bdad4ad531

arm_compute v17.04

src/runtime/NEON/functions/NESoftmaxLayer.cpp

diff --git a/src/runtime/NEON/functions/NESoftmaxLayer.cpp b/src/runtime/NEON/functions/NESoftmaxLayer.cpp
index 3949529..55d4d3a 100644
--- a/src/runtime/NEON/functions/NESoftmaxLayer.cpp
+++ b/src/runtime/NEON/functions/NESoftmaxLayer.cpp
@@ -32,7 +32,7 @@
 using namespace arm_compute;
 
 NESoftmaxLayer::NESoftmaxLayer()
-    : _max_kernel(), _shift_exp_sum_kernel(), _norm_kernel(), _fill_border_kernel(), _max(), _sum(), _tmp()
+    : _max_kernel(), _shift_exp_sum_kernel(), _norm_kernel(), _fill_border_kernel(), _fill_border_kernel_sum(), _max(), _sum(), _tmp()
 {
 }
 
@@ -42,31 +42,34 @@
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32);
 
     // Create intermediate tensors shapes
-    TensorInfo tensor_info_tmp(input->info()->tensor_shape(), input->info()->num_channels(), input->info()->data_type());
-    tensor_info_tmp.auto_padding();
-    _tmp.allocator()->init(tensor_info_tmp);
-    _tmp.allocator()->allocate();
+    _tmp.allocator()->init(TensorInfo(input->info()->tensor_shape(), input->info()->num_channels(), input->info()->data_type()));
 
     TensorShape shape = input->info()->tensor_shape();
     shape.set(0, 1);
     TensorInfo tensor_info_max_sum(shape, input->info()->num_channels(), input->info()->data_type());
-    tensor_info_max_sum.auto_padding();
     _max.allocator()->init(tensor_info_max_sum);
-    _max.allocator()->allocate();
     _sum.allocator()->init(tensor_info_max_sum);
-    _sum.allocator()->allocate();
 
     // Configure Kernels
-    _fill_border_kernel.configure(input, 3, BorderMode::CONSTANT, PixelValue(-FLT_MAX));
     _max_kernel.configure(input, &_max);
     _shift_exp_sum_kernel.configure(input, &_max, &_tmp, &_sum);
     _norm_kernel.configure(&_tmp, &_sum, output);
+    _fill_border_kernel.configure(input, _max_kernel.border_size(), BorderMode::CONSTANT, PixelValue(-FLT_MAX));
+    // Fill the border around tmp buffer with sensible negative value.
+    // This avoids exp(-FLT_MAX) which will lead to -inf and destroy the calculation of sum when input is not a multiple of processed elements
+    _fill_border_kernel_sum.configure(input, _shift_exp_sum_kernel.border_size(), BorderMode::CONSTANT, PixelValue(-50.f));
+
+    // Allocate intermediate tensors
+    _tmp.allocator()->allocate();
+    _max.allocator()->allocate();
+    _sum.allocator()->allocate();
 }
 
 void NESoftmaxLayer::run()
 {
     NEScheduler::get().multithread(&_fill_border_kernel);
     NEScheduler::get().multithread(&_max_kernel);
+    NEScheduler::get().multithread(&_fill_border_kernel_sum);
     NEScheduler::get().multithread(&_shift_exp_sum_kernel);
     NEScheduler::get().multithread(&_norm_kernel);
 }