From 72686fa6ee0f04d458ed2274b4d34917628ef14d Mon Sep 17 00:00:00 2001 From: Pablo Tello Date: Mon, 3 Sep 2018 11:40:33 +0100 Subject: COMPMID-1550: Winograd integrate RSH changes. Refactors the transforms to make use of partial specialization. Change-Id: Idff68d22817a00a7ee9eef5351a5a9fd33147540 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/146635 Tested-by: Jenkins Reviewed-by: Georgios Pinitas --- .../convolution/winograd/transforms/input.hpp | 152 ++++++++++++--------- 1 file changed, 89 insertions(+), 63 deletions(-) (limited to 'arm_compute/core/NEON/kernels/convolution/winograd/transforms') diff --git a/arm_compute/core/NEON/kernels/convolution/winograd/transforms/input.hpp b/arm_compute/core/NEON/kernels/convolution/winograd/transforms/input.hpp index 369c2ff48f..473a13c3b0 100644 --- a/arm_compute/core/NEON/kernels/convolution/winograd/transforms/input.hpp +++ b/arm_compute/core/NEON/kernels/convolution/winograd/transforms/input.hpp @@ -29,10 +29,8 @@ namespace winograd { /***************************************************************************/ /* Instance-less API */ - template - template - void WinogradGEMM::InputTransform::execute( + template + void InputTransformImpl::execute( const T* const input, /** Input tensor data */ const int n_batches, /** Number of batches in input tensor. */ const int in_batch_stride, /** Stride between batches of the input. */ @@ -50,26 +48,9 @@ namespace winograd const int matrix_row_stride /** Stride within matrices. */ ) { - // If an Nx1 kernel then transpose and redirect to the 1xN implementation - if (kernel_cols == 1) - { - WinogradGEMM:: - template InputTransform::execute( - input, - n_batches, in_batch_stride, - n_cols, in_col_stride, - n_rows, in_row_stride, - n_channels, padding, - tile_N, tile_M, - output, matrix_stride, matrix_batch_stride, matrix_row_stride - ); - return; - } - // Compute the padding required on each edge of the image - const int pad_top = (padding == PADDING_SAME) ? (kernel_rows - 1) / 2 : 0; - const int pad_left = (padding == PADDING_SAME) ? (kernel_cols - 1) / 2 : 0; - const int tile_overlap = kernel_rows - 1; + const int pad_top = (padding == PADDING_SAME) ? (KernelRows - 1) / 2 : 0; + const int pad_left = (padding == PADDING_SAME) ? (KernelCols - 1) / 2 : 0; // Compute striding values (assuming NHWC ordered data) const int output_col_stride = matrix_row_stride; @@ -85,19 +66,19 @@ namespace winograd // Loop over rows of tiles for (int tile_i = 0; tile_i < tile_M; tile_i++) { + // Padding (top + bottom) for the row + const int row_top = tile_i*(InnerTileRows - overlap_rows) - pad_top; + const int row_bottom = row_top + InnerTileRows; + const int row_pad_top = std::max(0, pad_top - tile_i*(InnerTileRows - overlap_rows)); + const int row_pad_bottom = (row_bottom <= n_rows) ? 0 : row_bottom - n_rows; + // Pointer to the row - const int row_offset = (tile_i == 0) ? 0 : pad_top; + const int row_offset = std::min(0, row_pad_top - pad_top); const T* const input_base_row = ( - input_base_batch + ((inner_tile_rows - (kernel_rows - 1))*tile_i - row_offset)*in_row_stride + input_base_batch + ((InnerTileRows - overlap_rows)*tile_i + row_offset)*in_row_stride ); T* const outptr_base_row = outptr_base_batch + tile_i*output_row_stride; - // Padding (top + bottom) for the row - const int row_top = tile_i*(inner_tile_rows - tile_overlap) - pad_top; - const int row_bottom = row_top + inner_tile_rows; - const int row_pad_top = (tile_i == 0) ? pad_top : 0; - const int row_pad_bottom = (row_bottom <= n_rows) ? 0 : row_bottom - n_rows; - // Process the row process_tile_row( tile_N, n_channels, @@ -109,10 +90,40 @@ namespace winograd } } - template - template - void WinogradGEMM::InputTransform::process_tile_row( + + template + void InputTransformImpl::execute( + const T* const input, /** Input tensor data */ + const int n_batches, /** Number of batches in input tensor. */ + const int in_batch_stride, /** Stride between batches of the input. */ + const int n_rows, /** Number of rows in input tensor. */ + const int in_row_stride, /** Stride between rows of the input. */ + const int n_cols, /** Number of columns in input tensor. */ + const int in_col_stride, /** Stride between columns of the input. */ + const int n_channels, /** Number of channels in input tensor. */ + const PaddingType padding, /** Padding type. */ + const int tile_M, + const int tile_N, + T* const output, /** Base of output matrices. */ + const int matrix_stride, /** Stride between output matrices. */ + const int matrix_batch_stride, /** Stride between batches within the matrix. */ + const int matrix_row_stride /** Stride within matrices. */ + ) + { + // If an Nx1 kernel then transpose and redirect to the 1xN implementation + InputTransformImpl<1, KernelRows, 1, InnerTileRows, T>::execute( + input, + n_batches, in_batch_stride, + n_cols, in_col_stride, + n_rows, in_row_stride, + n_channels, padding, + tile_N, tile_M, + output, matrix_stride, matrix_batch_stride, matrix_row_stride + ); + } + + template + void InputTransformImpl::process_tile_row( const int tile_N, int n_channels, const T* const input_base, @@ -127,33 +138,25 @@ namespace winograd const int n_cols ) { - if (kernel_cols == 1) - { - // If an Nx1 implementation then this should never be reached. - return; - } - - constexpr int tile_overlap = kernel_cols - 1; - // Loop over columns of tiles for (int tile_j = 0; tile_j < tile_N; tile_j++) { // Padding (left + right) for the tile - const int t_pad_left = (tile_j == 0) ? row_pad_left : 0; - const int t_start = tile_j*(inner_tile_cols - tile_overlap) - row_pad_left; - const int t_end = t_start + inner_tile_cols; + const int t_start = tile_j*(InnerTileCols - overlap_cols) - row_pad_left; + const int t_end = t_start + InnerTileCols; + const int t_pad_left = std::max(0, row_pad_left - tile_j*(InnerTileCols - overlap_cols)); const int t_pad_right = (t_end <= n_cols) ? 0 : t_end - n_cols; // Get pointers into the inputs and outputs - const int col_offset = (tile_j == 0) ? 0 : row_pad_left; + const int col_offset = std::min(0, t_pad_left - row_pad_left); const T* const input_base_col = ( - input_base + ((inner_tile_cols - tile_overlap)*tile_j - col_offset)*input_col_stride + input_base + ((InnerTileCols - overlap_cols)*tile_j + col_offset)*input_col_stride ); T* const outptr = matrix_base + tile_j*matrix_row_stride; // Apply the specific tile processing function - const int f_pad_top = pad_top ? 1 : 0; - const int f_pad_left = t_pad_left ? 1 : 0; + const int f_pad_top = iceildiv(pad_top, 2); + const int f_pad_left = iceildiv(t_pad_left, 2); tile_fns[f_pad_top][f_pad_left][pad_bottom][t_pad_right]( n_channels, input_base_col, @@ -166,9 +169,8 @@ namespace winograd } /***************************************************************************/ - template - template - WinogradGEMM::InputTransform::InputTransform( + template + InputTransform::InputTransform( const T* const input, /** Input tensor data */ const int n_batches, /** Number of batches in input tensor. */ const int n_rows, /** Number of rows in input tensor. */ @@ -184,10 +186,10 @@ namespace winograd ) : _inptr(input), _outptr(output), _n_batches(n_batches), _n_rows(n_rows), _n_cols(n_cols), _n_channels(n_channels), _matrix_stride(matrix_stride), _matrix_row_stride(matrix_row_stride), - _tiles_M(iceildiv((padding == PADDING_SAME) ? n_rows : n_rows - kr + 1, - output_tile_rows)), - _tiles_N(iceildiv((padding == PADDING_SAME) ? n_cols : n_cols - kc + 1, - output_tile_cols)), + _tiles_M(iceildiv((padding == PADDING_SAME) ? n_rows : n_rows - KernelRows + 1, + InnerTileRows - KernelRows + 1)), + _tiles_N(iceildiv((padding == PADDING_SAME) ? n_cols : n_cols - KernelCols + 1, + InnerTileCols - KernelCols + 1)), _in_col_stride(in_col_stride ? in_col_stride : n_channels), _in_row_stride(in_row_stride ? in_row_stride : n_cols * _in_col_stride), _in_batch_stride(in_batch_stride ? in_batch_stride : n_rows * _in_row_stride), @@ -195,18 +197,16 @@ namespace winograd { } - template - template - unsigned int WinogradGEMM::InputTransform::get_window() const + template + unsigned int InputTransform::get_window() const { // The final window includes the tail, all other windows will be a multiple // of the window block in size. return iceildiv(_n_channels, WINDOW_BLOCK); } - template - template - void WinogradGEMM::InputTransform::run( + template + void InputTransform::run( const unsigned int start, const unsigned int stop ) { @@ -238,4 +238,30 @@ namespace winograd _matrix_row_stride ); } + + template + void InputTransform::execute( + const T* const input, /** Input tensor data */ + const int n_batches, /** Number of batches in input tensor. */ + const int in_batch_stride, /** Stride between batches of the input. */ + const int n_rows, /** Number of rows in input tensor. */ + const int in_row_stride, /** Stride between rows of the input. */ + const int n_cols, /** Number of columns in input tensor. */ + const int in_col_stride, /** Stride between columns of the input. */ + const int n_channels, /** Number of channels in input tensor. */ + const PaddingType padding, /** Padding type. */ + const int tile_M, + const int tile_N, + T* const output, /** Base of output matrices. */ + const int matrix_stride, /** Stride between output matrices. */ + const int matrix_batch_stride, /** Stride between batches within the matrix. */ + const int matrix_row_stride /** Stride within matrices. */ + ) + { + Transform::execute( + input, n_batches, in_batch_stride, n_rows, in_row_stride, n_cols, + in_col_stride, n_channels, padding, tile_M, tile_N, output, + matrix_stride, matrix_batch_stride, matrix_row_stride + ); + } } -- cgit v1.2.1