commit 8938bd3f40ea62ff56d6ed4e2db0a8aee34dd64a
author Kaizen <kaizen@arm.com> Thu Sep 28 14:38:23 2017 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Thu Sep 28 16:31:13 2017 +0100
tree c234331232f227e0cdfb567a54ecaa5460aaa064
parent f4a254c2745aeaab6f7276a675147d707002fe7a

arm_compute v17.09

Change-Id: I4bf8f4e6e5f84ce0d5b6f5ba570d276879f42a81

src/runtime/NEON/functions/NEConvolution.cpp

diff --git a/src/runtime/NEON/functions/NEConvolution.cpp b/src/runtime/NEON/functions/NEConvolution.cpp
index 3f39ae2..f10ffa6 100644
--- a/src/runtime/NEON/functions/NEConvolution.cpp
+++ b/src/runtime/NEON/functions/NEConvolution.cpp
@@ -24,7 +24,6 @@
 #include "arm_compute/runtime/NEON/functions/NEConvolution.h"
 
 #include "arm_compute/core/Error.h"
-#include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
 #include "arm_compute/core/NEON/kernels/NEConvolutionKernel.h"
 #include "arm_compute/core/PixelValue.h"
@@ -33,6 +32,7 @@
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/runtime/NEON/NEScheduler.h"
 #include "arm_compute/runtime/TensorAllocator.h"
+#include "support/ToolchainSupport.h"
 
 #include <array>
 #include <utility>
@@ -41,15 +41,15 @@
 
 void NEConvolution3x3::configure(ITensor *input, ITensor *output, const int16_t *conv, uint32_t scale, BorderMode border_mode, uint8_t constant_border_value)
 {
-    auto k = arm_compute::cpp14::make_unique<NEConvolution3x3Kernel>();
+    auto k = arm_compute::support::cpp14::make_unique<NEConvolution3x3Kernel>();
     k->configure(input, output, conv, scale, border_mode == BorderMode::UNDEFINED);
     _kernel = std::move(k);
     _border_handler.configure(input, _kernel->border_size(), border_mode, PixelValue(constant_border_value));
 }
 
 template <unsigned int matrix_size>
-NEConvolutionSquare<matrix_size>::NEConvolutionSquare()
-    : _tmp(), _is_separable(false), _kernel_hor(), _kernel_vert(), _kernel(), _border_handler()
+NEConvolutionSquare<matrix_size>::NEConvolutionSquare(std::shared_ptr<IMemoryManager> memory_manager)
+    : _memory_group(std::move(memory_manager)), _tmp(), _is_separable(false), _kernel_hor(), _kernel_vert(), _kernel(), _border_handler()
 {
 }
 
@@ -72,6 +72,10 @@
 
         _tmp.allocator()->init(TensorInfo(input->info()->tensor_shape(), 1, intermediate_type));
 
+        // Manage intermediate buffers
+        _memory_group.manage(&_tmp);
+
+        // Calculate scale
         if(scale == 0)
         {
             scale = calculate_matrix_scale(conv, matrix_size);
@@ -94,12 +98,16 @@
 template <unsigned int matrix_size>
 void                   NEConvolutionSquare<matrix_size>::run()
 {
-    _border_handler.run(_border_handler.window());
+    NEScheduler::get().schedule(&_border_handler, Window::DimZ);
 
     if(_is_separable)
     {
+        _memory_group.acquire();
+
         NEScheduler::get().schedule(&_kernel_hor, Window::DimY);
         NEScheduler::get().schedule(&_kernel_vert, Window::DimY);
+
+        _memory_group.release();
     }
     else
     {
@@ -113,7 +121,7 @@
 
 void NEConvolutionRectangle::configure(ITensor *input, ITensor *output, const int16_t *conv, uint32_t rows, uint32_t cols, uint32_t scale, BorderMode border_mode, uint8_t constant_border_value)
 {
-    auto k = arm_compute::cpp14::make_unique<NEConvolutionRectangleKernel>();
+    auto k = arm_compute::support::cpp14::make_unique<NEConvolutionRectangleKernel>();
     k->configure(input, output, conv, rows, cols, scale, border_mode == BorderMode::UNDEFINED);
     _kernel = std::move(k);
     _border_handler.configure(input, _kernel->border_size(), border_mode, PixelValue(constant_border_value));