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
|
/*
* 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
#ifdef __arm__
#include <arm_neon.h>
#include "../asmlib.hpp"
template<>
template<typename T>
inline void TransformImpl<6, 1, false, 4, 4, false>::Transform(T *out, const T *in, int ldin, int y0, int ymax, int k0, int kmax) {
uint32_t *outptr = reinterpret_cast<uint32_t *>(out);
const uint32_t *inptr = reinterpret_cast<const uint32_t *>(in);
bool first = true;
uint32_t zerobuff[16] = { 0 }; // 8 for asm loop plus up to 7 for overflow loop
for (int y=y0; y<ymax; y+=6) {
const uint32_t *inptr0 = inptr + y * ldin + k0;
const uint32_t *inptr1 = inptr0 + ldin;
const uint32_t *inptr2 = inptr1 + ldin;
const uint32_t *inptr3 = inptr2 + ldin;
const uint32_t *inptr4 = inptr3 + ldin;
const uint32_t *inptr5 = inptr4 + ldin;
//prefetch_2x(inptr0);
//prefetch_2x(inptr1);
//prefetch_2x(inptr2);
//prefetch_2x(inptr3);
//prefetch_2x(inptr4);
//prefetch_2x(inptr5);
int x=(kmax-k0);
for (;(x>7) || first;x-=8) {
/* Cope with ragged cases by copying from a buffer of zeroes instead */
/* 'first' forces this to always run at least once, needed if the total size is <=7. */
if ((y + 5) >= ymax) {
switch ((y + 5) - ymax) {
case 4:
inptr1 = zerobuff;
// fall through
case 3:
inptr2 = zerobuff;
// fall through
case 2:
inptr3 = zerobuff;
// fall through
case 1:
inptr4 = zerobuff;
// fall through
case 0:
inptr5 = zerobuff;
break;
default:
UNREACHABLE("Impossible.");
}
}
if (first) {
if (x<=7) {
break;
}
first = false;
}
__asm __volatile (
// Load up 8 elements (2 vectors) from each of 8 sources.
"VLD1.32 {d0-d3}, [%[inptr0]]!\n" // q0=A0A1A2A3
"VLD1.32 {d4-d7}, [%[inptr1]]!\n" // q2=B0B1B2B3
"VLD1.32 {d8-d11}, [%[inptr2]]!\n" // q4=C0C1C2C3
"VZIP.32 q0, q4\n" // q0=A0C0A1C1, q4 = A2C2A3C3
"VLD1.32 {d12-d15}, [%[inptr3]]!\n" // q6=D0D1D2D3
"VZIP.32 q2, q6\n" // q2=B0D0B1D1, q6 = B2D2B3D3
"VLD1.32 {d16-d19}, [%[inptr4]]!\n"
"VLD1.32 {d20-d23}, [%[inptr5]]!\n"
"VZIP.32 q8, q10\n" // q8=E0F0E1F1, q10 = E2F2E3F3
ASM_PREFETCH("[%[inptr0], #128]")
"VZIP.32 q0, q2\n" // q0 = A0B0C0D0, q2 = A1B1C1D1
// Store first elements
"VST1.32 {d0-d1}, [%[outptr]]!\n"
"VST1.32 {d16}, [%[outptr]]!\n"
"VZIP.32 q4, q6\n" // q4 = A2B2C2D2, q6 = A3B3C3D3
// Store second elements
"VST1.32 {d4-d5}, [%[outptr]]!\n"
"VZIP.32 q1, q5\n"
ASM_PREFETCH("[%[inptr1], #128]")
"VST1.32 {d17}, [%[outptr]]!\n"
"VZIP.32 q3, q7\n"
// Store third elements
"VZIP.32 q9, q11\n"
"VST1.32 {d8-d9}, [%[outptr]]!\n"
"VZIP.32 q1, q3\n"
ASM_PREFETCH("[%[inptr2], #128]")
"VST1.32 {d20}, [%[outptr]]!\n"
// Store fourth elements
"VZIP.32 q5, q7\n"
"VST1.32 {d12-d13}, [%[outptr]]!\n"
ASM_PREFETCH("[%[inptr3], #128]")
"VST1.32 {d21}, [%[outptr]]!\n"
// Fifth
"VST1.32 {d2-d3}, [%[outptr]]!\n"
ASM_PREFETCH("[%[inptr4], #128]")
"VST1.32 {d18}, [%[outptr]]!\n"
// Sixth
"VST1.32 {d6-d7}, [%[outptr]]!\n"
ASM_PREFETCH("[%[inptr5], #128]")
"VST1.32 {d19}, [%[outptr]]!\n"
// Seventh
"VST1.32 {d10-d11}, [%[outptr]]!\n"
"VST1.32 {d22}, [%[outptr]]!\n"
// Eighth
"VST1.32 {d14-d15}, [%[outptr]]!\n"
"VST1.32 {d23}, [%[outptr]]!\n"
: [inptr0] "+r" (inptr0), [inptr1] "+r" (inptr1), [inptr2] "+r" (inptr2), [inptr3] "+r" (inptr3),
[inptr4] "+r" (inptr4), [inptr5] "+r" (inptr5), [outptr] "+r" (outptr)
:
: "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "memory"
);
}
for (;x>0;x--) {
*outptr++ = *inptr0++;
*outptr++ = *inptr1++;
*outptr++ = *inptr2++;
*outptr++ = *inptr3++;
*outptr++ = *inptr4++;
*outptr++ = *inptr5++;
}
}
}
#endif // __arm__
|
