COMPMID-2257: Implement NEGenerateProposals.

Change-Id: I8d751f8b09f842a214c305a4530a71d82f16db8f
Signed-off-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1943
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>

1 files changed, 96 insertions, 20 deletions

diff --git a/src/core/NEON/kernels/NECopyKernel.cpp b/src/core/NEON/kernels/NECopyKernel.cpp
index 4722c05507..83f3dded4f 100644
--- a/src/core/NEON/kernels/NECopyKernel.cpp
+++ b/src/core/NEON/kernels/NECopyKernel.cpp
@@ -29,28 +29,88 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
 
-using namespace arm_compute;
+namespace arm_compute
+{
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding = PaddingList())
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON(padding.size() > 4);
+
+    // Validate output if initialized
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(misc::shape_calculator::compute_padded_shape(input->tensor_shape(), padding), output->tensor_shape());
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    }
+
+    return Status{};
+}
+
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
+{
+    // Output auto inizialitation if not yet initialized
+    auto_init_if_empty(*output, *input);
+    return std::make_pair(Status{}, calculate_max_window(*output));
+}
+
+std::pair<Status, Window> validate_and_configure_window_with_padding(ITensorInfo *input, ITensorInfo *output, const PaddingList &padding)
+{
+    const TensorShape input_shape  = input->tensor_shape();
+    const TensorShape padded_shape = misc::shape_calculator::compute_padded_shape(input_shape, padding);
+    auto_init_if_empty(*output, input->clone()->set_tensor_shape(padded_shape));
+    // Configure window
+    const Window win = calculate_max_window(*output, output->dimension(0));
+    return std::make_pair(Status{}, win);
+}
+
+} // namespace
 
 NECopyKernel::NECopyKernel()
-    : _input(nullptr), _output(nullptr)
+    : _input(nullptr), _output(nullptr), _padding()
 {
 }
 
-void NECopyKernel::configure(const ITensor *input, ITensor *output)
+void NECopyKernel::configure(const ITensor *input, ITensor *output, const PaddingList &padding)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), padding));
+
+    _input   = input;
+    _output  = output;
+    _padding = padding;
 
-    _input  = input;
-    _output = output;
+    std::pair<Status, Window> win_config;
 
-    INEKernel::configure(calculate_max_window(*output->info()));
+    if(padding.empty())
+    {
+        win_config = validate_and_configure_window(input->info(), output->info());
+    }
+    else
+    {
+        win_config = validate_and_configure_window_with_padding(input->info(), output->info(), padding);
+    }
+
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    INEKernel::configure(win_config.second);
 }
 
-Status NECopyKernel::validate(const arm_compute::ITensorInfo *input, const arm_compute::ITensorInfo *output)
+Status NECopyKernel::validate(const arm_compute::ITensorInfo *input, const arm_compute::ITensorInfo *output, const PaddingList &padding)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, padding));
+
+    if(padding.empty())
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
+    }
+    else
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_with_padding(input->clone().get(), output->clone().get(), padding).first);
+    }
+
     return Status{};
 }
 
@@ -60,22 +120,38 @@ void NECopyKernel::run(const Window &window, const ThreadInfo &info)
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
 
-    Window output_window{ window };
-    output_window.set(Window::DimX, Window::Dimension(output_window.x().start(), output_window.x().end(), _input->info()->dimension(0)));
-
-    Window out_slice = output_window.first_slice_window_1D();
+    if(_padding.empty())
+    {
+        Window output_window{ window };
+        output_window.set(Window::DimX, Window::Dimension(output_window.x().start(), output_window.x().end(), _input->info()->dimension(0)));
+        Window out_slice = output_window.first_slice_window_1D();
+        do
+        {
+            Iterator input_it(_input, out_slice);
+            Iterator output_it(_output, out_slice);
 
-    do
+            execute_window_loop(out_slice, [&](const Coordinates &)
+            {
+                memcpy(output_it.ptr(), input_it.ptr(), _output->info()->dimension(0) * _output->info()->element_size());
+            },
+            input_it, output_it);
+        }
+        while(output_window.slide_window_slice_1D(out_slice));
+    }
+    else
     {
-        Iterator input_it(_input, out_slice);
-        Iterator output_it(_output, out_slice);
+        Window input_window{ window };
+        input_window.set(Window::DimX, Window::Dimension(0, window.x().end() - _padding[0].first, _input->info()->dimension(0)));
 
-        execute_window_loop(out_slice, [&](const Coordinates &)
+        Iterator     input_it(_input, input_window);
+        Iterator     output_it(_output, window);
+        const size_t row_size_in_bytes = _input->info()->dimension(0) * _input->info()->element_size();
+        execute_window_loop(window, [&](const Coordinates &)
         {
-            memcpy(output_it.ptr(), input_it.ptr(), _output->info()->dimension(0) * _output->info()->element_size());
+            auto dst_ptr = output_it.ptr() + _padding[0].first * _output->info()->element_size();
+            std::memcpy(dst_ptr, input_it.ptr(), row_size_in_bytes);
         },
         input_it, output_it);
-
     }
-    while(output_window.slide_window_slice_1D(out_slice));
 }
+} // namespace arm_compute