permute.cu
12.2 KB
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
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#include <cuda_runtime.h>
#include <cuda_fp16.h>
#include "array.hpp"
#include "types.hpp"
#include "grid_stride_range.hpp"
#include "execution.hpp"
#include "kernel_dispatcher.hpp"
#include "../cuda4dnn/csl/stream.hpp"
#include "../cuda4dnn/csl/tensor.hpp"
#include "../cuda4dnn/csl/span.hpp"
#include "../cuda4dnn/kernels/fill_copy.hpp"
#include <opencv2/core.hpp>
#include <cstddef>
#include <vector>
using namespace cv::dnn::cuda4dnn::csl;
using namespace cv::dnn::cuda4dnn::csl::device;
namespace cv { namespace dnn { namespace cuda4dnn { namespace kernels {
namespace raw {
template <class T, std::size_t Rank>
__global__ void permute(
array<index_type, Rank> axis_order,
Span<T> output, array<size_type, Rank> outStrides,
View<T> input, array<size_type, Rank> inStrides)
{
for (auto i : grid_stride_range(input.size())) {
index_type oldPosition = 0;
index_type newPosition = i;
for (int j = 0; j < Rank; j++)
{
auto order = axis_order[j];
oldPosition += (newPosition / outStrides[j]) * inStrides[order];
newPosition %= outStrides[j];
}
output[i] = input[oldPosition];
}
}
template <class T, int TILE_SIZE, int ROWS_PER_THREAD>
__global__ void transpose(Span<T> output, View<T> input, size_type in_width, size_type out_width)
{
__shared__ T tile[TILE_SIZE][TILE_SIZE + 1];
/* blockDim.y = TILE_SIZE / ROWS_PER_THREAD, blockDim.x = TILE_SIZE */
const index_type in_x = blockIdx.x * TILE_SIZE + threadIdx.x;
const index_type in_y_begin = blockIdx.y * TILE_SIZE + threadIdx.y;
/* Every valid input location has a corresponding output location and vice versa.
* Hence, if we do not load values into the shared memory for a given location, we
* also won't read them for storing in the output.
*/
for (int j = 0; j < TILE_SIZE; j += TILE_SIZE / ROWS_PER_THREAD)
{
const auto in_y_current = in_y_begin + j;
if (in_x < in_width && in_y_current < out_width)
tile[threadIdx.y + j][threadIdx.x] = input[in_y_current * in_width + in_x];
}
__syncthreads();
/* We interchange `threadIdx.x` and `threadIdx.y` so that consecutive output indices map to
* consecutive threads. This would allow writes across threds in a warp to be coalesced.
*/
const index_type out_x = blockIdx.y * TILE_SIZE + threadIdx.x;
const index_type out_y_begin = blockIdx.x * TILE_SIZE + threadIdx.y;
for (int j = 0; j < TILE_SIZE; j += TILE_SIZE / ROWS_PER_THREAD)
{
const auto out_y_current = out_y_begin + j;
if (out_x < out_width && out_y_current < in_width)
output[out_y_current * out_width + out_x] = tile[threadIdx.x][threadIdx.y + j];
}
}
}
template <class T>
void transpose(const Stream& stream, Span<T> output, View<T> input, std::size_t in_width, std::size_t out_width)
{
/* Each block processes a TILE_SIZE x TILE_SIZE piece */
constexpr int TILE_SIZE = 32;
/* Each thread processes ROWS_PER_THREAD rows. We do this to decrease the number of threads required
* in a block so that the cost of the block-wide synchronization is minimized.
*/
constexpr int ROWS_PER_THREAD = 4;
dim3 grid_size((in_width + TILE_SIZE - 1) / TILE_SIZE, (out_width + TILE_SIZE - 1) / TILE_SIZE);
dim3 block_size(TILE_SIZE, TILE_SIZE / ROWS_PER_THREAD);
auto policy = execution_policy(grid_size, block_size, stream);
auto kernel = raw::transpose<T, TILE_SIZE, ROWS_PER_THREAD>;
launch_kernel(kernel, policy, output, input, in_width, out_width);
}
template void transpose(const Stream&, Span<__half>, View<__half>, std::size_t, std::size_t);
template void transpose(const Stream&, Span<float>, View<float>, std::size_t, std::size_t);
template <class T, std::size_t Rank> static
void launch_permute_kernel(
const Stream& stream,
const std::vector<std::size_t>& order,
Span<T> output, const std::vector<std::size_t>& outStride,
View<T> input, const std::vector<std::size_t>& inStride)
{
CV_Assert(order.size() == Rank);
CV_Assert(outStride.size() == Rank);
CV_Assert(inStride.size() == Rank);
array<index_type, Rank> order_k;
order_k.assign(std::begin(order), std::end(order));
array<size_type, Rank> outStride_k, inStride_k;
outStride_k.assign(std::begin(outStride), std::end(outStride));
inStride_k.assign(std::begin(inStride), std::end(inStride));
auto kernel = raw::permute<T, Rank>;
auto policy = make_policy(kernel, input.size(), 0, stream);
launch_kernel(kernel, policy, order_k, output, outStride_k, input, inStride_k);
}
GENERATE_KERNEL_DISPATCHER(permute_dispatcher, launch_permute_kernel);
template <class T>
void permute(
const Stream& stream,
TensorSpan<T> output, TensorView<T> input,
std::vector<std::size_t> order)
{
CV_Assert(output.rank() == input.rank());
CV_Assert(input.rank() == order.size());
CV_Assert(input.size() == output.size());
auto rank = output.rank();
auto inShape = input.shape_as_vector();
auto outShape = output.shape_as_vector();
/* singleton axes do not contribute towards address calculation
*
* Reasoning:
* ----------
* Suppose an item's indices in the input tensor is [i1, i2, ...]. The indices in the
* output tensor will be some permutation of the input tensor indices. Let the output
* tensor indices be [o1, o2, ...]. The permutation operation essentially copies items
* from the input tensor to new locations in the output tensor as dictated by the indices.
*
* If the size of the nth axis (say i2) of the input is one the input and output indicies for
* all the elements will be of the form be [i1, 0, ...] and [..., 0, ...] respectively.
* The index does not contribute to the element's address calculation and hence would give
* identical result if it weren't there.
*/
for (int i = 0; i < rank; i++)
{
/* index `i` corresponds to the axis index in the output; order[i] has the corresponding axis index in the input */
while (i < rank && outShape[i] == 1)
{
int in_i = order[i];
CV_Assert(inShape[in_i] == 1);
/* delete axis `i` */
inShape.erase(std::begin(inShape) + in_i);
outShape.erase(std::begin(outShape) + i);
/* deletion of an axis reduces an axis in the input tensor which would cause the indices
* of the axes that come after the deleted axis to reduce by one
*/
order.erase(order.begin() + i);
for (auto& axis : order)
if (axis > in_i)
axis--;
rank--;
/* optimizations should not break the invariants */
CV_Assert(rank == order.size());
CV_Assert(inShape.size() == order.size());
CV_Assert(outShape.size() == order.size());
CV_Assert(input.size() == output.size());
}
}
/* contiguous axes whose relative ordering stays same before and after permutation can be merged into one axis
* example: in permute order 0 2 3 1, axes 2 and 3 can be grouped into a single axis
*
* Reasoning:
* ----------
* Suppose an item's indices in the input tensor is [i0, i1, i2, i3, ...]. Let the permutation order be [0, 3, 1, 2, ...].
* Note that i1 and i2 are adjacent axes in the same order in input as well as output. The indices in the output tensor
* will be [i0, i3, i1, i2, ...].
*
* Each axis in the contiguous axes sequence will add an offset of iN * strideN. In the above example,
* the two axes add a total offset of `i1 * (size2 * stride2) + i2 * stride2` which is `(i1 * size2 + i2) * stride2`,
* in both input and output. Note stride2 can be different in the input and output. We can merge the two axes into one axis
* with a size of `size1 * size2`. The new offset added will be `i12 * stride12` as the kernel iterates through `i12`. Note
* that `i12` is actually `(i1 * size2 + i2)` and `stride12` is `stride2`.
*/
for (int i = 0; i < rank; i++) {
/* the indices used in the loops such as `i` and `j` are axis indices in the output tensor */
/* the corresponding input axis indices are `order[i]` and `order[j]`*/
/* loop invariant: `i` is the first axis in the contiguous unpermuted axis sequence */
int j = i + 1; /* `j` is the axis which we will attempt to merge */
while (j < rank && (order[i] + 1) == order[j]) {
/* axis `i` and axis `j` do not change relative order */
auto in_i = order[i], in_j = order[j];
auto new_size = inShape[in_i] * inShape[in_j];
inShape[in_i] = new_size;
outShape[i] = new_size;
/* delete axis `j` */
inShape.erase(std::begin(inShape) + in_j);
outShape.erase(std::begin(outShape) + j);
/* deletion of an axis reduces an axis in the input tensor which would cause the indices
* of the axes that come after the deleted axis to reduce by one
*/
order.erase(order.begin() + j);
for (auto& axis : order)
if (axis > order[i])
axis--;
rank--;
/* optimizations should not break the invariants */
CV_Assert(rank == order.size());
CV_Assert(inShape.size() == order.size());
CV_Assert(outShape.size() == order.size());
CV_Assert(input.size() == output.size());
}
}
std::vector<std::size_t> inStride(rank), outStride(rank);
inStride.back() = 1;
outStride.back() = 1;
/* garbage, ..., garbage, 1 */
std::copy(std::begin(inShape) + 1, std::end(inShape), std::begin(inStride));
std::copy(std::begin(outShape) + 1, std::end(outShape), std::begin(outStride));
/* dim[0], dim[1], ..., dim[-1], 1 */
std::partial_sum(inStride.rbegin(), inStride.rend(), inStride.rbegin(), std::multiplies<std::size_t>());
std::partial_sum(outStride.rbegin(), outStride.rend(), outStride.rbegin(), std::multiplies<std::size_t>());
/* stride[0], stride[1], ..., stride[-2], 1 */
const bool is_in_order = [&order] {
for (int i = 0; i < order.size(); i++)
if (order[i] != i)
return false;
return true;
}();
if (is_in_order)
{
kernels::copy<T>(stream, output, input);
}
else if(rank == 2)
{
/* use the more efficient transpose kernel */
transpose<T>(stream, output, input, inShape[1], outShape[1]);
}
else
{
CV_Assert(3 <= rank && rank <= CSL_MAX_TENSOR_RANK);
permute_dispatcher<T, 3, CSL_MAX_TENSOR_RANK>(rank, stream, order, output, outStride, input, inStride);
}
}
#if !defined(__CUDA_ARCH__) || (__CUDA_ARCH__ >= 530)
template void permute(const Stream&, TensorSpan<__half>, TensorView<__half>, std::vector<std::size_t>);
#endif
template void permute(const Stream&, TensorSpan<float>, TensorView<float>, std::vector<std::size_t>);
}}}} /* namespace cv::dnn::cuda4dnn::kernels */