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
|
/*
* 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.
*/
#include "Simd/SimdMemory.h"
#include "Simd/SimdStore.h"
namespace Simd
{
#ifdef SIMD_AVX_ENABLE
namespace Avx
{
static void transpose4x4(const double* a, size_t rows, size_t cols, double * b, size_t i, size_t j)
{
__m256d a0 = _mm256_loadu_pd(&a[i*cols + j]);
__m256d a1 = _mm256_loadu_pd(&a[(i+1)*cols + j]);
__m256d a2 = _mm256_loadu_pd(&a[(i+2)*cols + j]);
__m256d a3 = _mm256_loadu_pd(&a[(i+3)*cols + j]);
__m256d T0 = _mm256_shuffle_pd(a0, a1, 15);
__m256d T1 = _mm256_shuffle_pd(a0, a1, 0);
__m256d T2 = _mm256_shuffle_pd(a2, a3, 15);
__m256d T3 = _mm256_shuffle_pd(a2, a3, 0);
a1 = _mm256_permute2f128_pd(T0, T2, 32);
a3 = _mm256_permute2f128_pd(T0, T2, 49);
a0 = _mm256_permute2f128_pd(T1, T3, 32);
a2 = _mm256_permute2f128_pd(T1, T3, 49);
_mm256_storeu_pd(&b[j*rows + i], a0);
_mm256_storeu_pd(&b[(j+1)*rows + i], a1);
_mm256_storeu_pd(&b[(j+2)*rows + i], a2);
_mm256_storeu_pd(&b[(j+3)*rows + i], a3);
}
void transpose16x16(const double * a, size_t rows, size_t cols, double * b, size_t i, size_t j)
{
transpose4x4(a, rows, cols, b, i, j);
transpose4x4(a, rows, cols, b, i, j + 4);
transpose4x4(a, rows, cols, b, i, j + 8);
transpose4x4(a, rows, cols, b, i, j + 12);
transpose4x4(a, rows, cols, b, i + 4, j);
transpose4x4(a, rows, cols, b, i + 4, j + 4);
transpose4x4(a, rows, cols, b, i + 4, j + 8);
transpose4x4(a, rows, cols, b, i + 4, j + 12);
transpose4x4(a, rows, cols, b, i + 8, j);
transpose4x4(a, rows, cols, b, i + 8, j + 4);
transpose4x4(a, rows, cols, b, i + 8, j + 8);
transpose4x4(a, rows, cols, b, i + 8, j + 12);
transpose4x4(a, rows, cols, b, i + 12, j);
transpose4x4(a, rows, cols, b, i + 12, j + 4);
transpose4x4(a, rows, cols, b, i + 12, j + 8);
transpose4x4(a, rows, cols, b, i + 12, j + 12);
}
void SimdMatTranspose(const double * mat, size_t rows, size_t cols, double * dst)
{
// Matrix transpose using tiling
const int nrows = static_cast<int>(rows);
const int ncols = static_cast<int>(cols);
const int tileSize = 16;
for (int i = 0; i < nrows;) {
for (; i <= nrows - tileSize; i += tileSize) {
int j = 0;
for (; j <= ncols - tileSize; j += tileSize) {
transpose16x16(mat, rows, cols, dst, i, j);
}
for (int k = i; k < i + tileSize; k++) {
for (int l = j; l < ncols; l++) {
dst[l*rows + k] = mat[k*cols + l];
}
}
}
for (; i < nrows; i++) {
for (int j = 0; j < ncols; j++) {
dst[j*rows + i] = mat[i*cols + j];
}
}
}
_mm256_zeroupper();
}
}
#endif// SIMD_AVX_ENABLE
}
|
