aboutsummaryrefslogtreecommitdiff
path: root/src/runtime/NEON/functions/NEScale.cpp
diff options
context:
space:
mode:
authorAnthony Barbier <anthony.barbier@arm.com>2017-09-04 18:44:23 +0100
committerAnthony Barbier <anthony.barbier@arm.com>2018-09-17 13:03:09 +0100
commit6ff3b19ee6120edf015fad8caab2991faa3070af (patch)
treea7a6dcd16dfd56d79fa1b56a313caeebcc939b68 /src/runtime/NEON/functions/NEScale.cpp
downloadComputeLibrary-6ff3b19ee6120edf015fad8caab2991faa3070af.tar.gz
COMPMID-344 Updated doxygen
Change-Id: I32f7b84daa560e460b77216add529c8fa8b327ae
Diffstat (limited to 'src/runtime/NEON/functions/NEScale.cpp')
-rw-r--r--src/runtime/NEON/functions/NEScale.cpp171
1 files changed, 171 insertions, 0 deletions
diff --git a/src/runtime/NEON/functions/NEScale.cpp b/src/runtime/NEON/functions/NEScale.cpp
new file mode 100644
index 0000000000..b70f626df0
--- /dev/null
+++ b/src/runtime/NEON/functions/NEScale.cpp
@@ -0,0 +1,171 @@
+/*
+ * Copyright (c) 2016, 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/NEON/functions/NEScale.h"
+
+#include "arm_compute/core/Coordinates.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/NEScaleKernel.h"
+#include "arm_compute/core/PixelValue.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Window.h"
+#include "arm_compute/runtime/TensorAllocator.h"
+
+#include <cmath>
+#include <cstddef>
+#include <utility>
+
+using namespace arm_compute;
+
+namespace
+{
+void precompute_dx_dy_offsets(ITensor *dx, ITensor *dy, ITensor *offsets, float wr, float hr, size_t input_element_size)
+{
+ ARM_COMPUTE_ERROR_ON(nullptr == offsets);
+
+ Window win;
+ win.set(Window::DimX, Window::Dimension(0, offsets->info()->dimension(0), 1));
+ win.set(Window::DimY, Window::Dimension(0, offsets->info()->dimension(1), 1));
+
+ if(dx != nullptr && dy != nullptr)
+ {
+ // Pre-compute the offset and pixel's distance for BILINEAR interpolation
+ Iterator offsets_it(offsets, win);
+ Iterator dx_it(dx, win);
+ Iterator dy_it(dy, win);
+
+ execute_window_loop(win, [&](const Coordinates & id)
+ {
+ const float in_x = (id.x() + 0.5f) * wr - 0.5f;
+ const float in_y = (id.y() + 0.5f) * hr - 0.5f;
+ const int in_xi = std::floor(in_x);
+ const int in_yi = std::floor(in_y);
+
+ *reinterpret_cast<int32_t *>(offsets_it.ptr()) = in_xi * input_element_size;
+ *reinterpret_cast<float *>(dx_it.ptr()) = in_x - in_xi;
+ *reinterpret_cast<float *>(dy_it.ptr()) = in_y - in_yi;
+ },
+ offsets_it, dx_it, dy_it);
+ }
+ else
+ {
+ // Pre-compute the offset for NEAREST interpolation
+ Iterator offsets_it(offsets, win);
+
+ execute_window_loop(win, [&](const Coordinates & id)
+ {
+ const size_t in_xi = (id.x() + 0.5f) * wr;
+
+ *reinterpret_cast<int32_t *>(offsets_it.ptr()) = in_xi * input_element_size;
+ },
+ offsets_it);
+ }
+}
+} // namespace
+
+NEScale::NEScale()
+ : _offsets(), _dx(), _dy()
+{
+}
+
+void NEScale::configure(ITensor *input, ITensor *output, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value)
+{
+ ARM_COMPUTE_ERROR_ON(nullptr == input);
+ ARM_COMPUTE_ERROR_ON(nullptr == output);
+
+ for(size_t i = 2; i < Coordinates::num_max_dimensions; ++i)
+ {
+ ARM_COMPUTE_ERROR_ON(input->info()->dimension(i) != output->info()->dimension(i));
+ }
+
+ // Get the tensor shape
+ const TensorShape shape(output->info()->dimension(0), output->info()->dimension(1));
+
+ // Compute the ratio between source width/height and destination width/height
+ const auto wr = static_cast<float>(input->info()->dimension(0)) / static_cast<float>(output->info()->dimension(0));
+ const auto hr = static_cast<float>(input->info()->dimension(1)) / static_cast<float>(output->info()->dimension(1));
+
+ // Get the element size of the input image
+ const size_t input_element_size = input->info()->element_size();
+
+ // Area interpolation behaves as Nearest Neighbour in case of up-sampling
+ if(policy == InterpolationPolicy::AREA && wr <= 1.f && hr <= 1.f)
+ {
+ policy = InterpolationPolicy::NEAREST_NEIGHBOR;
+ }
+
+ auto k = arm_compute::cpp14::make_unique<NEScaleKernel>();
+
+ // Check if the border mode is UNDEFINED
+ const bool border_undefined = border_mode == BorderMode::UNDEFINED;
+
+ switch(policy)
+ {
+ case InterpolationPolicy::NEAREST_NEIGHBOR:
+ {
+ TensorInfo tensor_info_offsets(shape, Format::S32);
+ _offsets.allocator()->init(tensor_info_offsets);
+
+ k->configure(input, nullptr, nullptr, &_offsets, output, policy, border_undefined);
+
+ // Allocate once the configure methods have been called
+ _offsets.allocator()->allocate();
+
+ // Pre-compute offsets for nearest interpolation
+ precompute_dx_dy_offsets(nullptr, nullptr, &_offsets, wr, hr, input_element_size);
+ break;
+ }
+ case InterpolationPolicy::BILINEAR:
+ {
+ TensorInfo tensor_info_offsets(shape, Format::S32);
+ TensorInfo tensor_info_dxdy(shape, Format::F32);
+
+ _offsets.allocator()->init(tensor_info_offsets);
+ _dx.allocator()->init(tensor_info_dxdy);
+ _dy.allocator()->init(tensor_info_dxdy);
+
+ k->configure(input, &_dx, &_dy, &_offsets, output, policy, border_undefined);
+
+ // Allocate once the configure methods have been called
+ _offsets.allocator()->allocate();
+ _dx.allocator()->allocate();
+ _dy.allocator()->allocate();
+
+ // Pre-compute dx, dy and offsets for bilinear interpolation
+ precompute_dx_dy_offsets(&_dx, &_dy, &_offsets, wr, hr, input_element_size);
+ break;
+ }
+ case InterpolationPolicy::AREA:
+ {
+ k->configure(input, nullptr, nullptr, nullptr, output, policy, border_undefined);
+ break;
+ }
+ default:
+ ARM_COMPUTE_ERROR("Unsupported interpolation mode");
+ }
+
+ _kernel = std::move(k);
+ _border_handler.configure(input, _kernel->border_size(), border_mode, PixelValue(constant_border_value));
+}