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