COMPMID-1772: Implement PadV2 for NEON

Change-Id: Ia4604524a034c46b004fd850183480c5fbfd8cb3
Reviewed-on: https://review.mlplatform.org/437
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>

3 files changed, 196 insertions, 6 deletions

diff --git a/src/core/NEON/kernels/NEMemsetKernel.cpp b/src/core/NEON/kernels/NEMemsetKernel.cpp
new file mode 100644
index 0000000000..2b57b1595b
--- /dev/null
+++ b/src/core/NEON/kernels/NEMemsetKernel.cpp
@@ -0,0 +1,82 @@
+/*
+ * Copyright (c) 2018-2019 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/core/NEON/kernels/NEMemsetKernel.h"
+
+#include "arm_compute/core/AccessWindowStatic.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/ITensor.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/core/Window.h"
+
+namespace arm_compute
+{
+NEMemsetKernel::NEMemsetKernel()
+    : _tensor(nullptr), _constant_value()
+{
+}
+
+void NEMemsetKernel::configure(ITensor *tensor, const PixelValue &constant_value)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(tensor);
+    _tensor         = tensor;
+    _constant_value = constant_value;
+
+    // Configure kernel window
+    Window win = calculate_max_window(*tensor->info(), Steps());
+    INEKernel::configure(win);
+}
+
+void NEMemsetKernel::run(const Window &window, const ThreadInfo &info)
+{
+    ARM_COMPUTE_UNUSED(info);
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
+
+    // Collapse all the batches on the third dimension
+    bool   has_collapsed = true;
+    Window collapsed     = window.collapse_if_possible(window, Window::DimZ, &has_collapsed);
+    ARM_COMPUTE_ERROR_ON(!has_collapsed);
+
+    uint8_t *const start_valid_region = _tensor->ptr_to_element(_tensor->info()->valid_region().anchor);
+    const auto     window_width       = static_cast<int>(collapsed.x().end()) - static_cast<int>(collapsed.x().start());
+    const size_t   element_size       = _tensor->info()->element_size();
+
+    // Unroll X dimension
+    collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
+
+    Iterator tensor_it(_tensor, collapsed);
+    execute_window_loop(collapsed, [&](const Coordinates & id)
+    {
+        uint8_t *base_addr = start_valid_region + tensor_it.offset();
+        // Set memory
+        for(int i = 0; i < window_width; ++i)
+        {
+            std::memcpy(base_addr + i * element_size, &_constant_value.value, element_size);
+        }
+
+    },
+    tensor_it);
+}
+} // namespace arm_compute
diff --git a/src/runtime/CL/functions/CLPadLayer.cpp b/src/runtime/CL/functions/CLPadLayer.cpp
index de43c7dca2..3aa1b1e1a0 100644
--- a/src/runtime/CL/functions/CLPadLayer.cpp
+++ b/src/runtime/CL/functions/CLPadLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -34,21 +34,21 @@ CLPadLayer::CLPadLayer()
 {
 }
 
-void CLPadLayer::configure(ICLTensor *input, ICLTensor *output, const PaddingList &padding)
+void CLPadLayer::configure(ICLTensor *input, ICLTensor *output, const PaddingList &padding, PixelValue constant_value)
 {
     // Copy the input to the output
     _copy_kernel.configure(input, output, padding);
 
     // Set the pages of the output to zero
-    _memset_kernel.configure(output, PixelValue());
+    _memset_kernel.configure(output, constant_value);
 
     // Fill padding on the first two dimensions with zeros
-    _fillborder_kernel.configure(input, input->info()->padding(), BorderMode::CONSTANT);
+    _fillborder_kernel.configure(input, input->info()->padding(), BorderMode::CONSTANT, constant_value);
 }
 
-Status CLPadLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding)
+Status CLPadLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding, PixelValue constant_value)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(CLMemsetKernel::validate(input, PixelValue()));
+    ARM_COMPUTE_RETURN_ON_ERROR(CLMemsetKernel::validate(input, constant_value));
     ARM_COMPUTE_RETURN_ON_ERROR(CLCopyKernel::validate(input, output, padding));
 
     return Status{};
diff --git a/src/runtime/NEON/functions/NEPadLayer.cpp b/src/runtime/NEON/functions/NEPadLayer.cpp
new file mode 100644
index 0000000000..f5c2718cec
--- /dev/null
+++ b/src/runtime/NEON/functions/NEPadLayer.cpp
@@ -0,0 +1,108 @@
+/*
+ * Copyright (c) 2018-2019 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/NEPadLayer.h"
+
+#include "arm_compute/runtime/NEON/NEScheduler.h"
+
+#include "arm_compute/core/ITensor.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
+#include "support/ToolchainSupport.h"
+
+namespace arm_compute
+{
+namespace
+{
+TensorInfo get_expected_output_tensorinfo(const ITensorInfo &input, const PaddingList &paddings)
+{
+    const TensorShape expected_output_shape = arm_compute::misc::shape_calculator::compute_padded_shape(input.tensor_shape(), paddings);
+    const TensorInfo  expected_output_info  = input.clone()->set_tensor_shape(expected_output_shape);
+    return expected_output_info;
+}
+
+Status validate_arguments(const ITensorInfo &input, ITensorInfo &output, const PaddingList &paddings)
+{
+    const TensorInfo expected_output_info = get_expected_output_tensorinfo(input, paddings);
+    auto_init_if_empty(output, expected_output_info);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(&output, &expected_output_info);
+
+    return Status{};
+}
+
+Coordinates get_subtensor_coords(const PaddingList &paddings)
+{
+    Coordinates coords;
+    for(unsigned int i = 0; i < paddings.size(); ++i)
+    {
+        coords.set(i, paddings[i].first);
+    }
+
+    return coords;
+}
+} // namespace
+
+NEPadLayer::NEPadLayer()
+    : _memset_kernel(), _copy_kernel(), _output_subtensor()
+{
+}
+
+void NEPadLayer::configure(ITensor *input, ITensor *output, const PaddingList &padding, PixelValue constant_value)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_THROW_ON_ERROR(NEPadLayer::validate(input->info(), output->info(), padding, constant_value));
+
+    // Auto-init
+    auto_init_if_empty(*output->info(), get_expected_output_tensorinfo(*input->info(), padding));
+
+    // Create SubTensor (Can use sub-tensor as the kernels to be executed do not require padding)
+    _output_subtensor = SubTensor(output, input->info()->tensor_shape(), get_subtensor_coords(padding), true);
+
+    // Set the pages of the output to the specified value
+    _memset_kernel.configure(output, constant_value);
+
+    // Copy the input to the output
+    _copy_kernel.configure(input, &_output_subtensor);
+}
+
+Status NEPadLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding, PixelValue constant_value)
+{
+    ARM_COMPUTE_UNUSED(constant_value);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+
+    auto output_clone = output->clone();
+
+    SubTensorInfo output_subtensor_info(output_clone.get(), input->tensor_shape(), get_subtensor_coords(padding), true);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*input, *output_clone, padding));
+    ARM_COMPUTE_RETURN_ON_ERROR(NECopyKernel::validate(input, &output_subtensor_info));
+
+    return Status{};
+}
+
+void NEPadLayer::run()
+{
+    NEScheduler::get().schedule(&_memset_kernel, Window::DimY);
+    NEScheduler::get().schedule(&_copy_kernel, Window::DimY);
+}
+} // namespace arm_compute