f4334277
 Hu Chunming
提交3rdparty
|
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
|
/*************************************************************************************
* Copyright (c) 2015, Advanced Micro Devices, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification,
* are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation and/or
* other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
* INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA,
* OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
**************************************************************************************/
__kernel void LRNComputeOutput(const int nthreads, __global T* in, __global T* scale, const T negative_beta, __global T* out) {
int index = get_global_id(0);
int tmp = get_global_size(0);
for(index; index < nthreads; index += tmp)
out[index] = in[index] * pow(scale[index], negative_beta);
}
__kernel void LRNFillScale(const int nthreads, __global T* in, const int num, const int channels, const int height, const int width, const int size, const T alpha_over_size, const T k, __global T* scale) {
int index = get_global_id(0);
int tmp = get_global_size(0);
for(index; index < nthreads; index += tmp) {
// find out the local offset
const int w = index % width;
const int h = (index / width) % height;
const int n = index / width / height;
const int offset = (n * channels * height + h) * width + w;
const int step = height * width;
in = in + offset;
scale = scale + offset;
int head = 0;
const int pre_pad = (size - 1) / 2;
const int post_pad = size - pre_pad - 1;
T accum_scale = 0;
// fill the scale at [n, :, h, w]
// accumulate values
while (head < post_pad && head < channels) {
accum_scale += in[head * step] * in[head * step];
++head;
}
// both add and subtract
while (head < channels) {
accum_scale += in[head * step] * in[head * step];
if (head - size >= 0) {
accum_scale -= in[(head - size) * step]
* in[(head - size) * step];
}
scale[(head - post_pad) * step] = k + accum_scale * alpha_over_size;
++head;
}
// subtract only
while (head < channels + post_pad) {
if (head - size >= 0) {
accum_scale -= in[(head - size) * step]
* in[(head - size) * step];
}
scale[(head - post_pad) * step] = k + accum_scale * alpha_over_size;
++head;
}
}
}
|
