1 /*
2  * Copyright (c) 2021 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 
sve_transpose_interleave_4VL(uint16_t * out,const uint16_t * in,size_t width,size_t in_stride,size_t height)32 void sve_transpose_interleave_4VL(uint16_t *out, const uint16_t *in, size_t width, size_t in_stride, size_t height)
33 {
34     size_t out_stride = 4 * height * get_vector_length<uint8_t>();
35 
36     __asm__ __volatile__(
37       "ptrue p3.b\n"
38       "cmp %x[height], #0x4\n"
39       "blt 4f\n"
40       "1:"  // Main row loop: Head
41       "mov x26, %x[in]\n"
42       "mov x25, %x[out]\n"
43       "add x24, x26, %x[in_stride]\n"
44       "add x23, x24, %x[in_stride]\n"
45       "add x22, x23, %x[in_stride]\n"
46       "add %x[in], x22, %x[in_stride]\n"
47       "sub %x[height], %x[height], #0x4\n"
48       "mov x21, %x[width]\n"
49       "2:"  // Main row loop: Column loop
50       "mov x20, x21\n"
51       "mov x19, x25\n"
52       "whilelt p0.h, XZR, x20\n"
53       "ld1h { z31.h }, p0/Z, [x26]\n"
54       "ld1h { z30.h }, p0/Z, [x24]\n"
55       "dech x20\n"
56       "ld1h { z29.h }, p0/Z, [x23]\n"
57       "whilelt p2.h, XZR, x20\n"
58       "ld1h { z28.h }, p0/Z, [x22]\n"
59       "dech x20\n"
60       "ld1h { z27.h }, p2/Z, [x26, #1, MUL VL]\n"
61       "whilelt p1.h, XZR, x20\n"
62       "ld1h { z26.h }, p2/Z, [x24, #1, MUL VL]\n"
63       "dech x20\n"
64       "ld1h { z25.h }, p1/Z, [x26, #2, MUL VL]\n"
65       "whilelt p0.h, XZR, x20\n"
66       "ld1h { z24.h }, p1/Z, [x24, #2, MUL VL]\n"
67       "add x25, x25, %x[out_stride]\n"
68       "ld1h { z23.h }, p0/Z, [x26, #3, MUL VL]\n"
69       "addvl x26, x26, #4\n"
70       "ld1h { z22.h }, p0/Z, [x24, #3, MUL VL]\n"
71       "addvl x24, x24, #4\n"
72       "ld1h { z21.h }, p2/Z, [x23, #1, MUL VL]\n"
73       "dech x21, ALL, MUL #4\n"
74       "ld1h { z20.h }, p1/Z, [x23, #2, MUL VL]\n"
75       "cmp x21, #0x0\n"
76       "ld1h { z19.h }, p0/Z, [x23, #3, MUL VL]\n"
77       "addvl x23, x23, #4\n"
78       "ld1h { z18.h }, p2/Z, [x22, #1, MUL VL]\n"
79       "ld1h { z17.h }, p1/Z, [x22, #2, MUL VL]\n"
80       "ld1h { z16.h }, p0/Z, [x22, #3, MUL VL]\n"
81       "addvl x22, x22, #4\n"
82       "st1h { z31.h }, p3, [x19]\n"
83       "st1h { z27.h }, p3, [x19, #1, MUL VL]\n"
84       "st1h { z25.h }, p3, [x19, #2, MUL VL]\n"
85       "st1h { z23.h }, p3, [x19, #3, MUL VL]\n"
86       "st1h { z30.h }, p3, [x19, #4, MUL VL]\n"
87       "st1h { z26.h }, p3, [x19, #5, MUL VL]\n"
88       "st1h { z24.h }, p3, [x19, #6, MUL VL]\n"
89       "st1h { z22.h }, p3, [x19, #7, MUL VL]\n"
90       "addvl x19, x19, #16\n"
91       "st1h { z29.h }, p3, [x19, #-8, MUL VL]\n"
92       "st1h { z21.h }, p3, [x19, #-7, MUL VL]\n"
93       "st1h { z20.h }, p3, [x19, #-6, MUL VL]\n"
94       "st1h { z19.h }, p3, [x19, #-5, MUL VL]\n"
95       "st1h { z28.h }, p3, [x19, #-4, MUL VL]\n"
96       "st1h { z18.h }, p3, [x19, #-3, MUL VL]\n"
97       "st1h { z17.h }, p3, [x19, #-2, MUL VL]\n"
98       "st1h { z16.h }, p3, [x19, #-1, MUL VL]\n"
99       "bgt 2b\n"
100       "3:"  // Main row loop: Column loop skip
101       "addvl %x[out], %x[out], #16\n"
102       "cmp %x[height], #0x4\n"
103       "bge 1b\n"
104       "cbz %x[height], 8f\n"
105       "4:"  // Main loop skip
106 
107       "5:"  // Tail row loop: Head
108       "mov x26, %x[in]\n"
109       "mov x25, %x[out]\n"
110       "add %x[in], x26, %x[in_stride]\n"
111       "sub %x[height], %x[height], #0x1\n"
112       "mov x20, %x[width]\n"
113       "6:"  // Tail row loop: Column loop
114       "mov x19, x20\n"
115       "dech x20, ALL, MUL #4\n"
116       "whilelt p0.h, XZR, x19\n"
117       "ld1h { z19.h }, p0/Z, [x26]\n"
118       "dech x19\n"
119       "whilelt p0.h, XZR, x19\n"
120       "ld1h { z18.h }, p0/Z, [x26, #1, MUL VL]\n"
121       "dech x19\n"
122       "whilelt p0.h, XZR, x19\n"
123       "ld1h { z17.h }, p0/Z, [x26, #2, MUL VL]\n"
124       "dech x19\n"
125       "whilelt p0.h, XZR, x19\n"
126       "ld1h { z16.h }, p0/Z, [x26, #3, MUL VL]\n"
127       "st1h { z19.h }, p3, [x25]\n"
128       "addvl x26, x26, #4\n"
129       "st1h { z18.h }, p3, [x25, #1, MUL VL]\n"
130       "cmp x20, #0x0\n"
131       "st1h { z17.h }, p3, [x25, #2, MUL VL]\n"
132       "st1h { z16.h }, p3, [x25, #3, MUL VL]\n"
133       "add x25, x25, %x[out_stride]\n"
134       "bgt 6b\n"
135       "7:"  // Tail row loop: Column loop skip
136       "addvl %x[out], %x[out], #4\n"
137       "cmp %x[height], #0x1\n"
138       "bge 5b\n"
139       "8:"  // Done
140 
141       : [height] "+&r" (height), [in] "+&r" (in), [out] "+&r" (out)
142       : [in_stride] "r" (in_stride), [out_stride] "r" (out_stride), [width] "r" (width)
143       : "cc", "memory", "p0", "p1", "p2", "p3", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
144     );
145 }
146 
147 } // anonymous namespace
148 
149 template<>
Transform(float * out,const float * in,int stride,int x0,int xmax,int k0,int kmax)150 void Transform<4, 1, true, VLType::SVE>(
151     float *out, const float *in, int stride, int x0, int xmax, int k0, int kmax)
152 {
153     sve_transpose_interleave_4VL(
154         reinterpret_cast<uint16_t *>(out),
155         reinterpret_cast<const uint16_t *>(in + k0 * stride + x0),
156         (xmax-x0) * sizeof(float) / 2,
157         stride * sizeof(float),
158         (kmax-k0)
159     );
160 }
161 
162 template<>
Transform(__fp16 * out,const __fp16 * in,int stride,int x0,int xmax,int k0,int kmax)163 void Transform<4, 1, true, VLType::SVE>(
164     __fp16 *out, const __fp16 *in, int stride, int x0, int xmax, int k0, int kmax)
165 {
166     sve_transpose_interleave_4VL(
167         reinterpret_cast<uint16_t *>(out),
168         reinterpret_cast<const uint16_t *>(in + k0 * stride + x0),
169         (xmax-x0) * sizeof(__fp16) / 2,
170         stride * sizeof(__fp16),
171         (kmax-k0)
172     );
173 }
174 
175 template<>
Transform(double * out,const double * in,int stride,int x0,int xmax,int k0,int kmax)176 void Transform<4, 1, true, VLType::SVE>(
177     double *out, const double *in, int stride, int x0, int xmax, int k0, int kmax)
178 {
179     sve_transpose_interleave_4VL(
180         reinterpret_cast<uint16_t *>(out),
181         reinterpret_cast<const uint16_t *>(in + k0 * stride + x0),
182         (xmax-x0) * sizeof(double) / 2,
183         stride * sizeof(double),
184         (kmax-k0)
185     );
186 }
187 
188 #endif
189