aboutsummaryrefslogtreecommitdiff
path: root/arm_compute/core/NEON/kernels/winograd
diff options
context:
space:
mode:
Diffstat (limited to 'arm_compute/core/NEON/kernels/winograd')
-rw-r--r--arm_compute/core/NEON/kernels/winograd/perf.h23
-rw-r--r--arm_compute/core/NEON/kernels/winograd/winograd_layer.hpp88
2 files changed, 83 insertions, 28 deletions
diff --git a/arm_compute/core/NEON/kernels/winograd/perf.h b/arm_compute/core/NEON/kernels/winograd/perf.h
index 0cdf742a25..3c0d36646d 100644
--- a/arm_compute/core/NEON/kernels/winograd/perf.h
+++ b/arm_compute/core/NEON/kernels/winograd/perf.h
@@ -1,3 +1,26 @@
+/*
+ * Copyright (c) 2018 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.
+ */
#pragma once
/* Prototypes from perf.c */
diff --git a/arm_compute/core/NEON/kernels/winograd/winograd_layer.hpp b/arm_compute/core/NEON/kernels/winograd/winograd_layer.hpp
index a3b3db42dd..4559312df4 100644
--- a/arm_compute/core/NEON/kernels/winograd/winograd_layer.hpp
+++ b/arm_compute/core/NEON/kernels/winograd/winograd_layer.hpp
@@ -74,55 +74,87 @@ class WinogradConvolutionLayer
/** Determine how much memory (in units of TIn) to allocate for the
* transformed weights.
+ *
+ * @param[in] n_output_channels Number of output feature maps.
+ * @param[in] n_input_channels Number of input feature maps.
*/
static unsigned int get_weight_storage_size(
- const int n_output_channels, /** Number of output feature maps. */
- const int n_input_channels /** Number of input feature maps. */
+ const int n_output_channels,
+ const int n_input_channels
);
/** Determine how much memory (in units of TIn) to allocate for the
* transformed input.
+ *
+ * @param[in] n_batches Number of batches in the input tensor.
+ * @param[in] n_channels Number of feature maps in the input tensor.
+ * @param[in] n_rows Number of rows in each feature map.
+ * @param[in] n_cols Number of columns in each feature map.
+ * @param[in] same_padding Use "SAME" padding, otherwise use "VALID".
*/
static unsigned int get_input_storage_size(
- const int n_batches, /** Number of batches in the input tensor. */
- const int n_channels, /** Number of feature maps in the input tensor. */
- const int n_rows, /** Number of rows in each feature map. */
- const int n_cols, /** Number of columns in each feature map. */
- const bool same_padding /** Use "SAME" padding, otherwise use "VALID". */
+ const int n_batches,
+ const int n_channels,
+ const int n_rows,
+ const int n_cols,
+ const bool same_padding
);
/** Determine how much memory (in units of TOut) to allocate for the
* (Winograd domain) output.
+ *
+ * @param[in] n_batches Number of batches in the output tensor.
+ * @param[in] n_rows Number of rows in each feature map of the input tensor.
+ * @param[in] n_cols Number of columns in each feature map of the input tensor.
+ * @param[in] n_output_channels Number of feature maps in the output tensor.
+ * @param[in] same_padding Use "SAME" padding, otherwise use "VALID".
*/
static unsigned int get_output_storage_size(
- const int n_batches, /** Number of batches in the output tensor. */
- const int n_rows, /** Number of rows in each feature map of the input tensor. */
- const int n_cols, /** Number of columns in each feature map of the input tensor. */
- const int n_output_channels, /** Number of feature maps in the output tensor. */
- const bool same_padding /** Use "SAME" padding, otherwise use "VALID". */
+ const int n_batches,
+ const int n_rows,
+ const int n_cols,
+ const int n_output_channels,
+ const bool same_padding
);
- /** Get the shape (rows, cols) of a feature map of the output tensor. */
+ /** Get the shape (rows, cols) of a feature map of the output tensor.
+ *
+ * @param[in] n_input_rows Number of rows in the input feature map.
+ * @param[in] n_input_cols Number of columns in the input feature map.
+ * @param[in] same_padding Use "SAME" padding, otherwise use "VALID".
+ */
static std::pair<int, int> get_output_feature_map_shape(
- const int n_input_rows, /** Number of rows in the input feature map. */
- const int n_input_cols, /** Number of columns in the input feature map. */
- const bool same_padding /** Use "SAME" padding, otherwise use "VALID". */
+ const int n_input_rows,
+ const int n_input_cols,
+ const bool same_padding
);
/** Create a new Winograd convolution layer.
+ * @param[in] n_batches Number of batches in the input and output tensors.
+ * @param[in] n_input_channels Number of feature maps in a batch of the input tensor.
+ * @param[in] n_input_rows Number of rows in a feature map of the input tensor.
+ * @param[in] n_input_cols Number of columns in a feature map of the input tensor.
+ * @param[in] n_output_channels Number of feature maps in the output tensor.
+ * @param[in] same_padding Use "SAME" padding, otherwise use "VALID".
+ * @param[in] weights Pointer to weight tensor in spatial domain. Must be ordered as "Height x Rows x Input Feature Maps x Output Feature Maps.
+ * @param[out] weights_storage Pointer to storage for weight tensor in the Winograd domain. Must be at least the size returned by `get_weight_storage_size
+ * @param[in] input Pointer to NHWC ordered input tensor, in the spatial domain.
+ * @param[out] winograd_input Pointer to working space for the input tensor in the Winograd domain. Must be at least the size returned by `get_input_storage_size`.
+ * @param[out] output Pointer to NHWC ordered output tensor, in the spatial domain.
+ * @param[out] winograd_output Pointer to working space for the output tensor in the Winograd domain. Must be at least the size returned by `get_output_storage_size`.
*/
WinogradConvolutionLayer(
- const int n_batches, /** Number of batches in the input and output tensors. */
- const int n_input_channels, /** Number of feature maps in a batch of the input tensor. */
- const int n_input_rows, /** Number of rows in a feature map of the input tensor. */
- const int n_input_cols, /** Number of columns in a feature map of the input tensor. */
- const int n_output_channels, /** Number of feature maps in the output tensor. */
- const bool same_padding, /** Use "SAME" padding, otherwise use "VALID". */
- const TIn* const weights, /** Pointer to weight tensor in spatial domain. Must be ordered as "Height x Rows x Input Feature Maps x Output Feature Maps. */
- TIn* const weights_storage, /** Pointer to storage for weight tensor in the Winograd domain. Must be at least the size returned by `get_weight_storage_size`. */
- const TIn* const input, /** Pointer to NHWC ordered input tensor, in the spatial domain. */
- TIn* const winograd_input, /** Pointer to working space for the input tensor in the Winograd domain. Must be at least the size returned by `get_input_storage_size`. */
- TOut* const output, /** Pointer to NHWC ordered output tensor, in the spatial domain. */
- TOut* const winograd_output /** Pointer to working space for the output tensor in the Winograd domain. Must be at least the size returned by `get_output_storage_size`. */
+ const int n_batches,
+ const int n_input_channels,
+ const int n_input_rows,
+ const int n_input_cols,
+ const int n_output_channels,
+ const bool same_padding,
+ const TIn* const weights,
+ TIn* const weights_storage,
+ const TIn* const input,
+ TIn* const winograd_input,
+ TOut* const output,
+ TOut* const winograd_output
);
};