24.04
sme_interleave4VL_block4_u8_u8_summing.hpp
Go to the documentation of this file.
1
/*
2
* Copyright (c) 2022-2023 Arm Limited.
3
*
4
* SPDX-License-Identifier: MIT
5
*
6
* Permission is hereby granted, free of charge, to any person obtaining a copy
7
* of this software and associated documentation files (the "Software"), to
8
* deal in the Software without restriction, including without limitation the
9
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10
* sell copies of the Software, and to permit persons to whom the Software is
11
* furnished to do so, subject to the following conditions:
12
*
13
* The above copyright notice and this permission notice shall be included in all
14
* copies or substantial portions of the Software.
15
*
16
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22
* SOFTWARE.
23
*/
24
25
#if defined(ARM_COMPUTE_ENABLE_SME)
26
27
template
<>
28
void
interleave_block<4, 4, VLType::SME, true>(
29
uint8_t * &out,
const
uint8_t *
const
*in,
30
size_t
width,
size_t
height,
size_t
row_offset,
bool
first
31
)
32
{
33
__asm__ __volatile__(
34
".inst 0xd503477f // SMSTART ZA\n"
35
"cntw x16\n"
36
"cntw x15\n"
37
"mov z24.b, #0x1\n"
38
"cntw x14, ALL, MUL #2\n"
39
"cntw x13, ALL, MUL #3\n"
40
"mov z23.s, #0x0\n"
41
"mov z22.s, #0x0\n"
42
"cmp %x[height], x16\n"
43
"csel x16, %x[height], x16, LT\n"
44
"mov z21.s, #0x0\n"
45
"mov z20.s, #0x0\n"
46
"whilelt p12.b, XZR, %x[height]\n"
47
"whilelt p10.b, x15, %x[height]\n"
48
"whilelt p9.b, x14, %x[height]\n"
49
"whilelt p8.b, x13, %x[height]\n"
50
"zip1 p12.b, p12.b, p9.b\n"
51
"zip1 p10.b, p10.b, p8.b\n"
52
"ptrue p2.b\n"
53
"cntb x11\n"
54
"ptrue p11.s\n"
55
"sub x16, x16, #0x1\n"
56
"zip1 p10.b, p12.b, p10.b\n"
57
"mov x10, %x[row_offset]\n"
58
"mov x9, %x[out]\n"
59
"cbnz %x[first], 1f\n"
60
"addvl x9, x9, #-4\n"
61
"ld1w { z23.s }, p2/Z, [x9]\n"
62
"ld1w { z22.s }, p2/Z, [x9, #1, MUL VL]\n"
63
"ld1w { z21.s }, p2/Z, [x9, #2, MUL VL]\n"
64
"ld1w { z20.s }, p2/Z, [x9, #3, MUL VL]\n"
65
"1:"
// Initialise row sums: End
66
"mov x28, #0x0\n"
67
"whilelt p9.b, x28, %x[width]\n"
68
"whilelt p8.b, x28, %x[width]\n"
69
"2:"
// Width loop
70
"add x27, %x[in], XZR, LSL #3\n"
71
"add x26, %x[in], x15, LSL #3\n"
72
"ldr x25, [x27], #0x8\n"
73
"add x24, %x[in], x14, LSL #3\n"
74
"add x23, %x[in], x13, LSL #3\n"
75
"ldr x20, [x26], #0x8\n"
76
"mov x12, #0x0\n"
77
"ldr x22, [x24], #0x8\n"
78
"ldr x21, [x23], #0x8\n"
79
"cbz x16, 4f\n"
80
"3:"
// Loads: Loop
81
".inst 0x25246140 // psel p0.b, p8.b/Z, p10.b[w12]\n"
82
".inst 0xe00a0320 // ld1b { za0h.b[x12] }, p0/Z, [x25, x10]\n"
83
".inst 0x252c6140 // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
84
"ldr x25, [x27], #0x8\n"
85
".inst 0xe00a0281 // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x10]\n"
86
".inst 0x25346141 // psel p1.b, p8.b/Z, p10.b[w12, #2]\n"
87
".inst 0x253c6140 // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
88
"ldr x20, [x26], #0x8\n"
89
".inst 0xe00a06c2 // ld1b { za0h.b[x12, #2] }, p1/Z, [x22, x10]\n"
90
"ldr x22, [x24], #0x8\n"
91
".inst 0xe00a02a3 // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x10]\n"
92
"add x12, x12, #0x4\n"
93
"cmp x12, x16, LSL #2\n"
94
"ldr x21, [x23], #0x8\n"
95
"blt 3b\n"
96
"4:"
// Loads: Tail
97
".inst 0x25246140 // psel p0.b, p8.b/Z, p10.b[w12]\n"
98
".inst 0xe00a0320 // ld1b { za0h.b[x12] }, p0/Z, [x25, x10]\n"
99
".inst 0x252c6140 // psel p0.b, p8.b/Z, p10.b[w12, #1]\n"
100
".inst 0xe00a0281 // ld1b { za0h.b[x12, #1] }, p0/Z, [x20, x10]\n"
101
".inst 0x25346140 // psel p0.b, p8.b/Z, p10.b[w12, #2]\n"
102
"sub x20, %x[width], x28\n"
103
".inst 0xe00a02c2 // ld1b { za0h.b[x12, #2] }, p0/Z, [x22, x10]\n"
104
"cmp x20, x11\n"
105
"csel x20, x20, x11, LT\n"
106
".inst 0x253c6140 // psel p0.b, p8.b/Z, p10.b[w12, #3]\n"
107
"add x20, x20, #0x3\n"
108
".inst 0xe00a02a3 // ld1b { za0h.b[x12, #3] }, p0/Z, [x21, x10]\n"
109
"mov x12, #0x0\n"
110
"lsr x20, x20, #0x2\n"
111
"5:"
// Stores: Loop
112
".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
113
".inst 0xe0bf8120 // st1w { za0v.s[x12] }, p0/Z, [x9, XZR, LSL #2]\n"
114
".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
115
".inst 0xc0828810 // mova z16.s, p2/M, za0v.s[x12]\n"
116
".inst 0xe0af8124 // st1w { za1v.s[x12] }, p0/Z, [x9, x15, LSL #2]\n"
117
".inst 0x25306d21 // psel p1.s, p11.s/Z, p9.s[w12]\n"
118
".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n"
119
".inst 0xc0828891 // mova z17.s, p2/M, za1v.s[x12]\n"
120
".inst 0xe0ae8528 // st1w { za2v.s[x12] }, p1/Z, [x9, x14, LSL #2]\n"
121
".inst 0xc0828913 // mova z19.s, p2/M, za2v.s[x12]\n"
122
"udot z23.s, z16.b, z24.b\n"
123
".inst 0xe0ad812c // st1w { za3v.s[x12] }, p0/Z, [x9, x13, LSL #2]\n"
124
".inst 0xc0828992 // mova z18.s, p2/M, za3v.s[x12]\n"
125
"add x12, x12, #0x1\n"
126
"cmp x12, x20\n"
127
"udot z22.s, z17.b, z24.b\n"
128
"udot z21.s, z19.b, z24.b\n"
129
"addvl x9, x9, #4\n"
130
"udot z20.s, z18.b, z24.b\n"
131
"blt 5b\n"
132
"incb x28\n"
133
"whilelt p9.b, x28, %x[width]\n"
134
"whilelt p8.b, x28, %x[width]\n"
135
"incb x10\n"
136
"b.any 2b\n"
137
"st1w { z23.s }, p2, [x9]\n"
138
"st1w { z22.s }, p2, [x9, #1, MUL VL]\n"
139
"st1w { z21.s }, p2, [x9, #2, MUL VL]\n"
140
"st1w { z20.s }, p2, [x9, #3, MUL VL]\n"
141
"addvl x9, x9, #4\n"
142
"mov %x[out], x9\n"
143
".inst 0xd503467f // SMSTOP\n"
144
: [out]
"+&r"
(out)
145
: [first]
"r"
(first), [height]
"r"
(height), [in]
"r"
(in), [row_offset]
"r"
(row_offset), [width]
"r"
(width)
146
:
"cc"
,
"memory"
,
"p0"
,
"p1"
,
"p2"
,
"p3"
,
"p4"
,
"p5"
,
"p6"
,
"p7"
,
"p8"
,
"p9"
,
"p10"
,
"p11"
,
"p12"
,
"p13"
,
"p14"
,
"p15"
,
"x9"
,
"x10"
,
"x11"
,
"x12"
,
"x13"
,
"x14"
,
"x15"
,
"x16"
,
"x20"
,
"x21"
,
"x22"
,
"x23"
,
"x24"
,
"x25"
,
"x26"
,
"x27"
,
"x28"
,
"z0"
,
"z1"
,
"z2"
,
"z3"
,
"z4"
,
"z5"
,
"z6"
,
"z7"
,
"z8"
,
"z9"
,
"z10"
,
"z11"
,
"z12"
,
"z13"
,
"z14"
,
"z15"
,
"z16"
,
"z17"
,
"z18"
,
"z19"
,
"z20"
,
"z21"
,
"z22"
,
"z23"
,
"z24"
,
"z25"
,
"z26"
,
"z27"
,
"z28"
,
"z29"
,
"z30"
,
"z31"
147
);
148
}
149
150
#endif // defined(ARM_COMPUTE_ENABLE_SME)
src
core
NEON
kernels
arm_gemm
indirect-interleaves
sme_interleave4VL_block4_u8_u8_summing.hpp
Generated on Mon Apr 29 2024 10:53:55 for Compute Library by
1.8.17