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
|
/*
* Copyright (c) 2017-2018 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
/* As some of the merges need these headers, but are all included in the
* arm_gemm namespace, put these headers here. */
#include <arm_neon.h>
#include "asmlib.hpp"
#include "utils.hpp"
namespace arm_gemm {
template<unsigned int twidth, unsigned int height, bool sve=false, typename Tin, typename Tout>
inline void MergeResults(Tout * out, const Tin * in, int ldc, int y0, int ymax, int x0, int xmax, const Tout alpha, const Tout beta) {
// For SVE cases, multiply the width up by the vector length.
// Use the *input* type to determine this, since this will be what the kernel operated on.
const int width = twidth * (sve ? get_vector_length<Tin>() : 1);
const int full_y_blocks = (ymax - y0) / height;
const int y_remainder = (ymax - y0) % height;
const int y_blocks = full_y_blocks + (y_remainder ? 1 : 0);
const int full_x_blocks = (xmax - x0) / width;
const int x_remainder = (xmax - x0) % width;
const int x_blocks = full_x_blocks + (x_remainder ? 1 : 0);
for (int y_block = 0; y_block < y_blocks; y_block++) {
int ybase = y0 + (y_block * height);
int fill_rows = (y_block < full_y_blocks) ? height : y_remainder;
for (int x_block = 0; x_block < x_blocks; x_block++) {
int xbase = x0 + (x_block * width);
int fill_cols = (x_block < full_x_blocks) ? width : x_remainder;
for (int row=0; row < fill_rows; row++) {
for (int col=0; col < fill_cols; col++) {
Tout &p = out[(ybase + row) * ldc + xbase + col];
// Special case for beta==0 - don't read the input;
// (0 * x == 0) is not always true for FP types.
if (beta == static_cast<Tout>(0)) {
p = (alpha * in[row * width + col]);
} else {
p = (p * beta) + (alpha * in[row * width + col]);
}
}
}
in += (width * height);
}
}
}
#include "merges/list.hpp"
} // namespace arm_gemm
|
