1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318
|
/*
* Copyright (c) 2018-2021 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.
*/
#ifndef ARM_COMPUTE_WINDOW_ITERATOR_H
#define ARM_COMPUTE_WINDOW_ITERATOR_H
#include "arm_compute/core/Coordinates.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/ITensor.h"
#include "arm_compute/core/Window.h"
namespace arm_compute
{
/** Convert an offset in window steps into absolute coordinates.
*
* @param[in] w Window @p offset is related to.
* @param[in] offset Offset inside the window expressed in number of window steps.
*
* @return Absolute coordinates.
*/
inline Coordinates convert_window_coord_to_position(const Window &w, const Coordinates &offset)
{
Coordinates position;
for (unsigned int i = 0; i < Coordinates::num_max_dimensions; ++i)
{
position.set(i, w[i].start() + offset[i] * w[i].step());
}
return position;
}
/** Tensor accessors to make it easier to interface with arm_gemm */
template <typename T>
class TensorAccessor
{
public:
/** Constructor:
*
* @param[in] tensor Source tensor, must be allocated.
*/
TensorAccessor(const ITensor &tensor)
: _first(tensor.ptr_to_element(Coordinates())), _strides(tensor.info()->strides_in_bytes())
{
}
/** Get the stride of the dimension dim expressed in number of Ts.
*
* @param[in] dim Dimension of the wanted stride.
*
* @return Stride in number of Ts.
*/
inline size_t stride(size_t dim) const
{
ARM_COMPUTE_ERROR_ON(_strides[dim] % sizeof(T) != 0);
return _strides[dim] / sizeof(T);
}
/** Manually set the stride of a dimension
*
* @param[in] dim Dimension of the stride to set.
* @param[in] size Value to set the stride to (in bytes).
*/
void set_stride(size_t dim, size_t size)
{
_strides[dim] = size;
}
/** Manually set the strides
*
* @param[in] strides Strides to set
*/
void set_strides(const Strides &strides)
{
_strides = strides;
}
/** Returns a pointer to the element at coordinates (x,y,z,w)
*
* @param[in] x X coordinates
* @param[in] y (optional) Y coordinates
* @param[in] z (optional) Z coordinates
* @param[in] w (optional) W coordinates
*/
inline T *get_ptr(unsigned int x, unsigned int y = 0, unsigned int z = 0, unsigned int w = 0)
{
return reinterpret_cast<T *>(_first + x * _strides[0] + y * _strides[1] + z * _strides[2] + w * _strides[3]);
}
/** Returns a pointer to the element at coordinates (x,y,z,w)
*
* @param[in] x X coordinates
* @param[in] y (optional) Y coordinates
* @param[in] z (optional) Z coordinates
* @param[in] w (optional) W coordinates
*/
inline T *operator()(unsigned int x, unsigned int y = 0, unsigned int z = 0, unsigned int w = 0)
{
return get_ptr(x, y, z, w);
}
/** Returns a pointer to the first element of the tensor
*
* @return Pointer to the first element.
*/
inline T *first_element()
{
return reinterpret_cast<T *>(_first);
}
/** Returns a pointer to the first element of the tensor
*
* @return Pointer to the first element.
*/
inline T *operator()()
{
return first_element();
}
private:
uint8_t *_first; /**< Pointer to the first element of the tensor.*/
Strides _strides; /**< Strides in bytes of the tensor */
};
/** Iterate over a portion of a Window */
template <typename L>
class WindowIterator
{
public:
/** Construct a WindowIterator object
*
* @param[in] w Window to use for the iteration
* @param[in] start Where to start iterating from (In Window coordinates)
* @param[in] end Where to stop iterating (In Window coordinates).
* @param[in] lambda_function Lambda function to call for every iteration between start and end. (It will be called last for end - 1)
*/
WindowIterator(const Window &w, const Coordinates &start, const Coordinates &end, L &&lambda_function)
: _lambda_function(std::move(lambda_function)),
_position(convert_window_coord_to_position(w, start)),
_end(convert_window_coord_to_position(w, end)),
_w(w)
{
}
/** Iterate over the lowest 3 dimensions of the window.
*
* @param[in] on_new_row_size Callback to be called before lambda_function every time the width of the row processed changes.
*/
template <typename M>
void iterate_3D(M &&on_new_row_size)
{
while (_end.z() != _position.z())
{
iterate_2D_internal(on_new_row_size, _w.x().end() - _w.x().step(), _w.y().end() - _w.y().step());
_position[2] += _w.z().step();
_position[1] = _w.y().start();
_position[0] = _w.x().start();
}
// Left over:
iterate_2D(on_new_row_size);
}
/** Iterate over the lowest 2 dimensions of the window.
*
* @param[in] on_new_row_size Callback to be called before lambda_function every time the width of the row processed changes.
*/
template <typename M>
void iterate_2D(M &&on_new_row_size)
{
iterate_2D_internal(on_new_row_size, _end.x(), _end.y());
}
/** Change the step used for the iteration.
*
* @note Does not affect the start and end points.
*
* @param[in] dim Dimension to change
* @param[in] step New step to use for the given dimension.
*/
inline void set_step(size_t dim, int step)
{
_w.set_dimension_step(dim, step);
}
/** Returns the coordinates in absolute coordinates of the end position
*
* @return End position coordinates.
*/
const Coordinates &end_position() const
{
return _end;
}
private:
template <typename M>
void iterate_2D_internal(M &&on_new_row_size, int end_x, int end_y)
{
//Is there more than one row to process ?
if (end_y == _position.y())
{
// Both start and end belong to the same row:
iterate_over_dim0(end_x + _w.x().step(), on_new_row_size);
}
else
{
// Do we start from the beginning of the row ?
if (_w.x().start() != _position.x())
{
//Start in the middle of a row: process left-over X
iterate_over_dim0(_w.x().end(), on_new_row_size);
_position[1] += _w.y().step();
}
//Middle rows
bool no_leftover = end_x + _w.x().step() == _w.x().end();
if (no_leftover)
{
//Switch to full row size:
on_new_row_size(_w[0].start(), _w.x().end());
// Shouldn't be possible to reach that point and not have at least one entire row to process
ARM_COMPUTE_ERROR_ON(_w.y().end() == _position.y());
// No leftover: all the rows lefts to process are full width:
iterate_over_dim1(end_y + _w.y().step());
}
else
{
// Are there full rows to process ?
if (_position[1] != end_y)
{
//Switch to full row size:
on_new_row_size(_w[0].start(), _w.x().end());
iterate_over_dim1(end_y);
}
//Leftover end x
_position[0] = _w.x().start();
iterate_over_dim0(end_x + _w.x().step(), on_new_row_size);
}
}
}
/** Process full rows below 'end'
*
* @param[in] end Y position to stop at.
*/
void iterate_over_dim1(int end)
{
for (; _position[1] != end; _position[1] += _w[1].step())
{
_position[0] = _w[0].start();
iterate_over_dim0(_w[0].end());
}
}
/** Process elements of a given row up to 'end'
*
* @param[in] end X position to stop at.
* @param[in] on_new_row_size Callback to call before starting iterating
*/
template <typename M>
void iterate_over_dim0(int end, M &&on_new_row_size)
{
on_new_row_size(_position.x(), end);
iterate_over_dim0(end);
}
/** Process elements of a given row up to 'end'
*
* @param[in] end X position to stop at.
*/
void iterate_over_dim0(int end)
{
// Both start and end belong to the same row:
ARM_COMPUTE_ERROR_ON(_position[0] > end);
for (; _position.x() < end; _position[0] += _w[0].step())
{
_lambda_function(_position);
}
}
L _lambda_function; /**< Function to call for each iteration */
Coordinates _position; /**< Absolute coordinates of the current position */
Coordinates _end; /**< Absolute coordinates of the point after the last iteration */
Window _w; /**< Window to iterate over */
};
/** Create a WindowIterator object
*
* @param[in] w Window to use for the iteration
* @param[in] start Where to start iterating from (In Window coordinates)
* @param[in] end Where to stop iterating (In Window coordinates).
* @param[in] lambda_function Lambda function to call for every iteration between start and end. (It will be called last for end - 1)
*
* @return A WindowIterator object.
*/
template <typename L>
WindowIterator<L>
create_window_iterator(const Window &w, const Coordinates &start, const Coordinates &end, L &&lambda_function)
{
return WindowIterator<L>(w, start, end, std::move(lambda_function));
}
} // namespace arm_compute
#endif /*ARM_COMPUTE_WINDOW_ITERATOR_H*/
|