COMPMID-2063: New Winograd implementation
Refactoring of winograd code reducing the size of the binaries
about 8X.
Change-Id: If8845bda324573e1a5cf436f354ac8603e88a92e
Signed-off-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/959
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Anthony Barbier <Anthony.barbier@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
diff --git a/src/core/NEON/kernels/NEWinogradConvolutionLayerKernel.cpp b/src/core/NEON/kernels/NEWinogradConvolutionLayerKernel.cpp
index 3e76a08..263ded0 100644
--- a/src/core/NEON/kernels/NEWinogradConvolutionLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEWinogradConvolutionLayerKernel.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -238,8 +238,7 @@
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
NEWinogradLayerTransformWeightsKernel<T, OutputTileRows, OutputTileCols, KernelRows, KernelCols>::NEWinogradLayerTransformWeightsKernel()
- : _weights_hwio(nullptr), _output(nullptr), _matrix_stride(0), _num_output_channels(0), _num_input_channels(0)
-
+ : _transform(nullptr), _weights_hwio(nullptr), _output(nullptr), _matrix_stride(0), _num_output_channels(0), _num_input_channels(0)
{
}
@@ -263,11 +262,10 @@
_matrix_stride = matrix_stride;
_num_output_channels = num_output_channels;
_num_input_channels = num_input_channels;
+ _transform = arm_compute::support::cpp14::make_unique<WeightsTransform>(num_output_channels, num_input_channels);
- const int matrix_row_stride = roundup(num_output_channels, WinogradConv::N_BLOCK);
- WeightsTransform transform(nullptr, nullptr, matrix_stride, matrix_row_stride, num_output_channels, num_input_channels);
- Window win;
- auto win_last = transform.get_window();
+ Window win;
+ auto win_last = _transform->get_window();
win.set(Window::DimX, Window::Dimension(0, win_last, 1));
INEKernel::configure(win);
}
@@ -278,12 +276,14 @@
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+ const size_t fst = window.x().start();
+ const size_t lst = window.x().end();
+ _transform->set_weight_tensor(_weights_hwio->buffer());
+ const int matrix_row_stride = roundup(_num_output_channels, WinogradConv::N_BLOCK);
+ _transform->set_output_matrices(_output->buffer(), _matrix_stride, matrix_row_stride);
+ _transform->set_working_space(_output->buffer());
- const int matrix_row_stride = roundup(_num_output_channels, WinogradConv::N_BLOCK);
- WeightsTransform transform(reinterpret_cast<T *>(_weights_hwio->buffer()), reinterpret_cast<T *>(_output->buffer()), _matrix_stride, matrix_row_stride, _num_output_channels, _num_input_channels);
- const size_t fst = window.x().start();
- const size_t lst = window.x().end();
- transform.run(fst, lst);
+ _transform->run(fst, lst);
}
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
@@ -331,6 +331,12 @@
}
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
+unsigned int NEWinogradLayerTransformInputKernel<T, OutputTileRows, OutputTileCols, KernelRows, KernelCols>::get_working_space_size(unsigned int num_threads) const
+{
+ return _transform->get_working_space_size(num_threads) / sizeof(T);
+}
+
+template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
int NEWinogradLayerTransformInputKernel<T, OutputTileRows, OutputTileCols, KernelRows, KernelCols>::get_matrix_stride(
const KernelShape &kernel_shape, const Tensor4DShape &input_shape, const PaddingType padding_type) const
{
@@ -339,7 +345,8 @@
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
NEWinogradLayerTransformInputKernel<T, OutputTileRows, OutputTileCols, KernelRows, KernelCols>::NEWinogradLayerTransformInputKernel()
- : _input_nhwc(), _num_batches(0), _num_rows(0), _num_cols(0), _num_channels(0), _padding(), _output(nullptr), _matrix_stride(0)
+ : _transform(nullptr), _input_nhwc(nullptr), _num_batches(0), _num_rows(0), _num_cols(0), _num_channels(0), _padding(), _output(nullptr), _matrix_stride(0), _padding_top(), _padding_left(),
+ _padding_right(), _padding_bottom(), _workspace(nullptr)
{
}
@@ -352,7 +359,8 @@
const int num_channels, /* Number of channels in input tensor. */
const PaddingType padding, /* Padding type. */
ITensor *output, /* Base of output matrices. */
- const int matrix_stride) /* Stride between output matrices. */
+ const int matrix_stride, /* Stride between output matrices. */
+ ITensor *workspace)
{
_input_nhwc = input_nhwc;
_num_batches = num_batches;
@@ -362,9 +370,28 @@
_padding = padding;
_output = output;
_matrix_stride = matrix_stride;
- InputTransform transform(nullptr, num_batches, num_rows, num_cols, num_channels, padding, nullptr, matrix_stride, num_channels);
- Window win;
- auto win_last = transform.get_window();
+ _workspace = workspace;
+
+ _padding_top = (padding == PADDING_SAME) ? (KernelRows - 1) / 2 : 0;
+ _padding_left = (padding == PADDING_SAME) ? (KernelCols - 1) / 2 : 0;
+ _padding_bottom = (padding == PADDING_SAME) ? iceildiv(KernelRows - 1, 2) : 0;
+ _padding_right = (padding == PADDING_SAME) ? iceildiv(KernelCols - 1, 2) : 0;
+
+ _transform = arm_compute::support::cpp14::make_unique<InputTransform>(
+ KernelRows,
+ KernelCols,
+ num_batches,
+ num_rows,
+ num_cols,
+ num_channels,
+ _padding_top, /**< Padding to apply to the top of the image. */
+ _padding_left, /**< Padding to apply to the left of the image. */
+ _padding_bottom, /**< Padding to apply to the bottom of the image. */
+ _padding_right /**< Padding to apply to the right of the image. */
+ );
+
+ Window win;
+ auto win_last = _transform->get_window();
win.set(Window::DimX, Window::Dimension(0, win_last, 1));
INEKernel::configure(win);
}
@@ -374,22 +401,25 @@
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(_workspace);
- const int element_size_in_bytes = _input_nhwc->info()->element_size();
- const int input_col_stride = _input_nhwc->info()->strides_in_bytes().y() / element_size_in_bytes;
- const int input_row_stride = _input_nhwc->info()->strides_in_bytes().z() / element_size_in_bytes;
- const int input_batch_stride = _input_nhwc->info()->strides_in_bytes()[3] / element_size_in_bytes;
- const auto input_nhwc_ptr = reinterpret_cast<const T *>(_input_nhwc->buffer() + _input_nhwc->info()->offset_first_element_in_bytes());
- auto output_ptr = reinterpret_cast<T *>(_output->buffer() + _output->info()->offset_first_element_in_bytes());
- InputTransform input_transform(input_nhwc_ptr,
- _num_batches, _num_rows, _num_cols, _num_channels, _padding,
- output_ptr,
- _matrix_stride, _num_channels, input_batch_stride, input_row_stride, input_col_stride);
+ const int element_size_in_bytes = _input_nhwc->info()->element_size();
+ const int input_col_stride = _input_nhwc->info()->strides_in_bytes().y() / element_size_in_bytes;
+ const int input_row_stride = _input_nhwc->info()->strides_in_bytes().z() / element_size_in_bytes;
+ const int input_batch_stride = _input_nhwc->info()->strides_in_bytes()[3] / element_size_in_bytes;
+ const auto input_nhwc_ptr = reinterpret_cast<const T *>(_input_nhwc->buffer() + _input_nhwc->info()->offset_first_element_in_bytes());
+ auto output_ptr = reinterpret_cast<T *>(_output->buffer() + _output->info()->offset_first_element_in_bytes());
+ ARM_COMPUTE_ERROR_ON_NULLPTR(output_ptr);
+
+ _transform->set_input_tensor(input_nhwc_ptr, input_batch_stride, input_row_stride, input_col_stride);
+ _transform->set_output_matrices(output_ptr, _matrix_stride, _num_channels);
+
+ _transform->set_working_space(_workspace->buffer());
// The code below cannot be moved to configure because biases hasn't been allocated at that point
const size_t fst = window.x().start();
const size_t lst = window.x().end();
- input_transform.run(fst, lst);
+ _transform->run(fst, lst, info.thread_id);
}
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
@@ -435,11 +465,18 @@
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
NEWinogradLayerTransformOutputKernel<T, OutputTileRows, OutputTileCols, KernelRows, KernelCols>::NEWinogradLayerTransformOutputKernel()
- : _biases(nullptr), _output_workspace(nullptr), _matrix_stride(0), _matrix_row_stride(0), _output_nhwc(nullptr), _num_batches(0), _num_rows(0), _num_cols(0), _num_channels(0)
+ : _transform(nullptr), _biases(nullptr), _transformed_output(nullptr), _workspace(nullptr), _matrix_stride(0), _matrix_row_stride(0), _output_nhwc(nullptr), _num_batches(0), _num_rows(0),
+ _num_cols(0), _num_channels(0)
{
}
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
+unsigned int NEWinogradLayerTransformOutputKernel<T, OutputTileRows, OutputTileCols, KernelRows, KernelCols>::get_working_space_size(unsigned int num_threads) const
+{
+ return _transform->get_working_space_size(num_threads) / sizeof(T);
+}
+
+template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
int NEWinogradLayerTransformOutputKernel<T, OutputTileRows, OutputTileCols, KernelRows, KernelCols>::get_matrix_stride(
const KernelShape &kernel_shape, const Tensor4DShape &input_shape, const PaddingType padding_type) const
{
@@ -455,28 +492,29 @@
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>
void NEWinogradLayerTransformOutputKernel<T, OutputTileRows, OutputTileCols, KernelRows, KernelCols>::configure(
const ITensor *biases,
- const ITensor *output_workingspace,
+ const ITensor *transformed_output,
const int matrix_stride,
ITensor *output_nhwc,
const int num_batches,
const int num_rows,
const int num_cols,
- const int num_channels)
+ const int num_channels,
+ ITensor *workspace)
{
- _biases = biases;
- _output_workspace = output_workingspace;
- _matrix_stride = matrix_stride;
- _matrix_row_stride = roundup(num_channels, WinogradConv::N_BLOCK);
- _output_nhwc = output_nhwc;
- _num_batches = num_batches;
- _num_rows = num_rows;
- _num_cols = num_cols;
- _num_channels = num_channels;
+ _biases = biases;
+ _workspace = workspace;
+ _transformed_output = transformed_output;
+ _matrix_stride = matrix_stride;
+ _matrix_row_stride = roundup(num_channels, WinogradConv::N_BLOCK);
+ _output_nhwc = output_nhwc;
+ _num_batches = num_batches;
+ _num_rows = num_rows;
+ _num_cols = num_cols;
+ _num_channels = num_channels;
// We don't have the biases buffer at this stage as it hasn't been allocated, we pass in nullptr OutputTransform is only used here to compute the window
- OutputTransform output_transform(nullptr, _matrix_stride, _matrix_row_stride, nullptr, nullptr, _num_batches, _num_rows, _num_cols, _num_channels);
-
+ _transform = arm_compute::support::cpp14::make_unique<OutputTransform>(num_batches, num_rows, num_cols, num_channels);
Window win;
- auto win_last = output_transform.get_window();
+ auto win_last = _transform->get_window();
win.set(Window::DimX, Window::Dimension(0, win_last, 1));
_output_nhwc->info()->set_valid_region(ValidRegion(Coordinates(), _output_nhwc->info()->tensor_shape()));
@@ -488,22 +526,22 @@
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
- ARM_COMPUTE_ERROR_ON_NULLPTR(_output_workspace);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(_workspace);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(_transformed_output);
ARM_COMPUTE_ERROR_ON_NULLPTR(_output_nhwc);
- const int out_batch_stride = 0;
+ const int out_batch_stride = _output_nhwc->info()->strides_in_bytes()[3] / sizeof(T);
const int out_row_stride = _output_nhwc->info()->strides_in_bytes()[2] / sizeof(T);
const int out_col_stride = _output_nhwc->info()->strides_in_bytes()[1] / sizeof(T);
- OutputTransform output_transform(reinterpret_cast<T *>(_output_workspace->buffer()), _matrix_stride, _matrix_row_stride,
- (_biases ? reinterpret_cast<T *>(_biases->buffer() + _biases->info()->offset_first_element_in_bytes()) : nullptr),
- reinterpret_cast<T *>(_output_nhwc->buffer() + _output_nhwc->info()->offset_first_element_in_bytes()),
- _num_batches, _num_rows, _num_cols, _num_channels, out_batch_stride, out_row_stride, out_col_stride);
-
+ _transform->set_input_matrices(_transformed_output->buffer(), _matrix_stride, _matrix_row_stride);
+ _transform->set_bias((_biases ? reinterpret_cast<T *>(_biases->buffer() + _biases->info()->offset_first_element_in_bytes()) : nullptr));
+ _transform->set_output_tensor(_output_nhwc->buffer() + _output_nhwc->info()->offset_first_element_in_bytes(), out_batch_stride, out_row_stride, out_col_stride);
+ _transform->set_working_space(_workspace->buffer());
// The code below cannot be moved to configure because biases hasn't been allocated at that point
const size_t fst = window.x().start();
const size_t lst = window.x().end();
- output_transform.run(fst, lst);
+ _transform->run(fst, lst, info.thread_id);
}
template <typename T, int OutputTileRows, int OutputTileCols, int KernelRows, int KernelCols>