1 /*
2 * Copyright (c) 2022 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
14 * all 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
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 * IN THE SOFTWARE.
23 */
24
25 #pragma once
26
27 #ifdef __ARM_FEATURE_SVE
28
29
30 namespace {
31
sme_transpose_interleave_4VL_1x4(uint8_t * out,const uint8_t * in,size_t width,size_t in_stride,size_t height)32 void sme_transpose_interleave_4VL_1x4(uint8_t *out, const uint8_t *in, size_t width, size_t in_stride, size_t height)
33 {
34 uint8_t *pad_row = reinterpret_cast<uint8_t *>(alloca(width * sizeof(uint8_t)));
35
36 if (height % 4) {
37 memset(pad_row, 0, width * sizeof(uint8_t));
38 }
39
40 size_t out_stride = 4 * roundup<size_t>(height, 4) * sme::get_vector_length<uint32_t>();
41
42 __asm__ __volatile__(
43 ".inst 0xd503477f // SMSTART ZA\n"
44 "ptrue p1.b\n"
45 "1:" // Main row loop: Head
46 "mov x24, %x[in]\n"
47 "add x23, x24, %x[in_stride]\n"
48 "add x22, x23, %x[in_stride]\n"
49 "add x21, x22, %x[in_stride]\n"
50 "cmp %x[height], #0x3\n"
51 "add %x[in], x21, %x[in_stride]\n"
52 "csel x21, x21, %x[pad_row], GT\n"
53 "csel x22, x22, %x[pad_row], GE\n"
54 "cmp %x[height], #0x1\n"
55 "mov x20, %x[out]\n"
56 "csel x23, x23, %x[pad_row], GT\n"
57 "sub %x[height], %x[height], #0x4\n"
58 "mov x19, %x[width]\n"
59 "2:" // Main row loop: Column loop
60 "whilelt p0.b, XZR, x19\n"
61 "ld1b { z17.b }, p0/Z, [x24]\n"
62 "decw x19, ALL, MUL #4\n"
63 "ld1b { z19.b }, p0/Z, [x23]\n"
64 "cmp x19, #0x0\n"
65 "addvl x24, x24, #1\n"
66 "ld1b { z16.b }, p0/Z, [x22]\n"
67 "zip1 z18.b, z17.b, z16.b\n"
68 "zip2 z20.b, z17.b, z16.b\n"
69 "addvl x23, x23, #1\n"
70 "ld1b { z16.b }, p0/Z, [x21]\n"
71 "zip1 z17.b, z19.b, z16.b\n"
72 "zip2 z19.b, z19.b, z16.b\n"
73 "addvl x22, x22, #1\n"
74 "addvl x21, x21, #1\n"
75 "zip1 z16.b, z18.b, z17.b\n"
76 "zip2 z18.b, z18.b, z17.b\n"
77 "st1b { z16.b }, p1, [x20]\n"
78 "zip1 z17.b, z20.b, z19.b\n"
79 "zip2 z16.b, z20.b, z19.b\n"
80 "st1b { z18.b }, p1, [x20, #1, MUL VL]\n"
81 "st1b { z17.b }, p1, [x20, #2, MUL VL]\n"
82 "st1b { z16.b }, p1, [x20, #3, MUL VL]\n"
83 "add x20, x20, %x[out_stride]\n"
84 "bgt 2b\n"
85 "3:" // Main row loop: Column loop skip
86 "cmp %x[height], #0x1\n"
87 "addvl %x[out], %x[out], #4\n"
88 "bge 1b\n"
89 ".inst 0xd503467f // SMSTOP\n"
90 : [height] "+&r" (height), [in] "+&r" (in), [out] "+&r" (out)
91 : [in_stride] "r" (in_stride), [out_stride] "r" (out_stride), [pad_row] "r" (pad_row), [width] "r" (width)
92 : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x19", "x20", "x21", "x22", "x23", "x24", "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"
93 );
94 }
95
96 } // anonymous namespace
97
98 template<>
Transform(uint8_t * out,const uint8_t * in,int stride,int x0,int xmax,int k0,int kmax)99 void Transform<4, 4, true, VLType::SME>(
100 uint8_t *out, const uint8_t *in, int stride, int x0, int xmax, int k0, int kmax)
101 {
102 sme_transpose_interleave_4VL_1x4(
103 reinterpret_cast<uint8_t *>(out),
104 reinterpret_cast<const uint8_t *>(in + k0 * stride + x0),
105 (xmax-x0) * sizeof(uint8_t) / 1,
106 stride * sizeof(uint8_t),
107 (kmax-k0)
108 );
109 }
110
111 template<>
Transform(int8_t * out,const int8_t * in,int stride,int x0,int xmax,int k0,int kmax)112 void Transform<4, 4, true, VLType::SME>(
113 int8_t *out, const int8_t *in, int stride, int x0, int xmax, int k0, int kmax)
114 {
115 sme_transpose_interleave_4VL_1x4(
116 reinterpret_cast<uint8_t *>(out),
117 reinterpret_cast<const uint8_t *>(in + k0 * stride + x0),
118 (xmax-x0) * sizeof(int8_t) / 1,
119 stride * sizeof(int8_t),
120 (kmax-k0)
121 );
122 }
123
124 #endif
125