commit 4406fd6cc4abded564d3791324e1f48bdfd34273
author Frank Lei <frank.lei@arm.com> Thu Feb 01 14:47:14 2018 +0800
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:47:40 2018 +0000
tree 22fe402fe9ac7ca338df49e9eccd6eb1587ae875
parent 898d399a0f62c15612a52df4bff5018e783214e4

APPBROWSER-391: Fix GLES COMPUTE alignment issues

APPBROWSER-402: Performance optimization for squeezenet/xray model

Change-Id: If31b186b99a6d6087164019fe94d3ac9279e3204
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/119526
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

src/core/GLES_COMPUTE/kernels/GCPoolingLayerKernel.cpp

diff --git a/src/core/GLES_COMPUTE/kernels/GCPoolingLayerKernel.cpp b/src/core/GLES_COMPUTE/kernels/GCPoolingLayerKernel.cpp
index c688cd4..3a0944c 100644
--- a/src/core/GLES_COMPUTE/kernels/GCPoolingLayerKernel.cpp
+++ b/src/core/GLES_COMPUTE/kernels/GCPoolingLayerKernel.cpp
@@ -198,11 +198,14 @@
         const int output_height         = output->dimension(1);
         const int output_padding_right  = ceil_to_multiple(output_width, num_elems_processed_per_iteration) - output_width;
         const int output_padding_bottom = ceil_to_multiple(output_height, 1) - output_height;
-        const int input_padding_right   = ceil_to_multiple(input_width + 2 * border_size.right, num_elems_processed_per_iteration) - (input_width + 2 * border_size.right);
-        const int input_padding_bottom  = ceil_to_multiple(input_height + 2 * border_size.bottom, 1) - (input_height + 2 * border_size.bottom);
+
+        const int input_total_width    = std::max(int(input->padding().left), int(pool_pad_x)) + input_width + std::max(int(input->padding().right), int(pool_pad_x));
+        const int input_padding_right  = ceil_to_multiple(input_total_width, num_elems_processed_per_iteration) - input_width - pool_pad_x;
+        const int input_total_height   = std::max(int(input->padding().top), int(pool_pad_y)) + input_height + std::max(int(input->padding().bottom), int(pool_pad_y));
+        const int input_padding_bottom = input_total_height - input_height - pool_pad_y;
 
         // Configure kernel window
-        AccessWindowStatic input_access(input, -pool_pad_x, -pool_pad_y, input_width + border_size.right + input_padding_right, input_height + border_size.bottom + input_padding_bottom);
+        AccessWindowStatic input_access(input, -pool_pad_x, -pool_pad_y, input_width + input_padding_right, input_height + input_padding_bottom);
         AccessWindowStatic output_access(output, 0, 0, output_width + output_padding_right, output_height + output_padding_bottom);
         bool               window_changed = update_window_and_padding(win, input_access, output_access);
         output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
@@ -340,13 +343,19 @@
 
     _kernel.use();
 
+    _output->set_needs_shifting(true);
+
     Window window_collapsed = window.collapse_if_possible(IGCKernel::window(), Window::DimZ);
-    Window slice            = window_collapsed.first_slice_window_3D();
+
+    Window slice         = window_collapsed.first_slice_window_3D();
+    Window slice_in_orig = window_collapsed.first_slice_window_3D();
+
+    slice.shift(Window::DimX, -(_output->info()->padding()).left);
 
     do
     {
         // Upsample input by pool size
-        Window in_slice(slice); // NOLINT
+        Window in_slice(slice_in_orig); // NOLINT
         in_slice.set(Window::DimX, Window::Dimension(in_slice.x().start() - pool_pad_x, in_slice.x().end() * pool_stride_x, pool_stride_x * _num_elems_processed_per_iteration));
         in_slice.set(Window::DimY, Window::Dimension(in_slice.y().start() - pool_pad_y, in_slice.y().end() * pool_stride_y, pool_stride_y));
 
@@ -358,5 +367,5 @@
         _kernel.update_shader_params();
         enqueue(*this, slice);
     }
-    while(window_collapsed.slide_window_slice_3D(slice));
+    while(window_collapsed.slide_window_slice_3D(slice) && window_collapsed.slide_window_slice_3D(slice_in_orig));
 }