1cabdff1aSopenharmony_ci/*
2cabdff1aSopenharmony_ci * This file is part of FFmpeg.
3cabdff1aSopenharmony_ci *
4cabdff1aSopenharmony_ci * FFmpeg is free software; you can redistribute it and/or
5cabdff1aSopenharmony_ci * modify it under the terms of the GNU Lesser General Public
6cabdff1aSopenharmony_ci * License as published by the Free Software Foundation; either
7cabdff1aSopenharmony_ci * version 2.1 of the License, or (at your option) any later version.
8cabdff1aSopenharmony_ci *
9cabdff1aSopenharmony_ci * FFmpeg is distributed in the hope that it will be useful,
10cabdff1aSopenharmony_ci * but WITHOUT ANY WARRANTY; without even the implied warranty of
11cabdff1aSopenharmony_ci * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12cabdff1aSopenharmony_ci * Lesser General Public License for more details.
13cabdff1aSopenharmony_ci *
14cabdff1aSopenharmony_ci * You should have received a copy of the GNU Lesser General Public
15cabdff1aSopenharmony_ci * License along with FFmpeg; if not, write to the Free Software
16cabdff1aSopenharmony_ci * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17cabdff1aSopenharmony_ci */
18cabdff1aSopenharmony_ci
19cabdff1aSopenharmony_ci#include "libavutil/aarch64/asm.S"
20cabdff1aSopenharmony_ci
21cabdff1aSopenharmony_ci           // 0.85..^1    0.85..^2    0.85..^3    0.85..^4
22cabdff1aSopenharmony_ciconst tab_st, align=4
23cabdff1aSopenharmony_ci        .word 0x3f599a00, 0x3f38f671, 0x3f1d382a, 0x3f05a32f
24cabdff1aSopenharmony_ciendconst
25cabdff1aSopenharmony_ciconst tab_x0, align=4
26cabdff1aSopenharmony_ci        .word 0x0,        0x3f599a00, 0x3f38f671, 0x3f1d382a
27cabdff1aSopenharmony_ciendconst
28cabdff1aSopenharmony_ciconst tab_x1, align=4
29cabdff1aSopenharmony_ci        .word 0x0,        0x0,        0x3f599a00, 0x3f38f671
30cabdff1aSopenharmony_ciendconst
31cabdff1aSopenharmony_ciconst tab_x2, align=4
32cabdff1aSopenharmony_ci        .word 0x0,        0x0,        0x0,        0x3f599a00
33cabdff1aSopenharmony_ciendconst
34cabdff1aSopenharmony_ci
35cabdff1aSopenharmony_cifunction ff_opus_deemphasis_neon, export=1
36cabdff1aSopenharmony_ci        movrel  x4, tab_st
37cabdff1aSopenharmony_ci        ld1    {v4.4s}, [x4]
38cabdff1aSopenharmony_ci        movrel  x4, tab_x0
39cabdff1aSopenharmony_ci        ld1    {v5.4s}, [x4]
40cabdff1aSopenharmony_ci        movrel  x4, tab_x1
41cabdff1aSopenharmony_ci        ld1    {v6.4s}, [x4]
42cabdff1aSopenharmony_ci        movrel  x4, tab_x2
43cabdff1aSopenharmony_ci        ld1    {v7.4s}, [x4]
44cabdff1aSopenharmony_ci
45cabdff1aSopenharmony_ci        fmul v0.4s, v4.4s, v0.s[0]
46cabdff1aSopenharmony_ci
47cabdff1aSopenharmony_ci1:      ld1  {v1.4s, v2.4s}, [x1], #32
48cabdff1aSopenharmony_ci
49cabdff1aSopenharmony_ci        fmla v0.4s, v5.4s, v1.s[0]
50cabdff1aSopenharmony_ci        fmul v3.4s, v7.4s, v2.s[2]
51cabdff1aSopenharmony_ci
52cabdff1aSopenharmony_ci        fmla v0.4s, v6.4s, v1.s[1]
53cabdff1aSopenharmony_ci        fmla v3.4s, v6.4s, v2.s[1]
54cabdff1aSopenharmony_ci
55cabdff1aSopenharmony_ci        fmla v0.4s, v7.4s, v1.s[2]
56cabdff1aSopenharmony_ci        fmla v3.4s, v5.4s, v2.s[0]
57cabdff1aSopenharmony_ci
58cabdff1aSopenharmony_ci        fadd v1.4s, v1.4s, v0.4s
59cabdff1aSopenharmony_ci        fadd v2.4s, v2.4s, v3.4s
60cabdff1aSopenharmony_ci
61cabdff1aSopenharmony_ci        fmla v2.4s, v4.4s, v1.s[3]
62cabdff1aSopenharmony_ci
63cabdff1aSopenharmony_ci        st1  {v1.4s, v2.4s}, [x0], #32
64cabdff1aSopenharmony_ci        fmul v0.4s, v4.4s, v2.s[3]
65cabdff1aSopenharmony_ci
66cabdff1aSopenharmony_ci        subs w2, w2, #8
67cabdff1aSopenharmony_ci        b.gt 1b
68cabdff1aSopenharmony_ci
69cabdff1aSopenharmony_ci        mov s0, v2.s[3]
70cabdff1aSopenharmony_ci
71cabdff1aSopenharmony_ci        ret
72cabdff1aSopenharmony_ciendfunc
73cabdff1aSopenharmony_ci
74cabdff1aSopenharmony_cifunction ff_opus_postfilter_neon, export=1
75cabdff1aSopenharmony_ci        ld1 {v0.4s}, [x2]
76cabdff1aSopenharmony_ci        dup v1.4s, v0.s[1]
77cabdff1aSopenharmony_ci        dup v2.4s, v0.s[2]
78cabdff1aSopenharmony_ci        dup v0.4s, v0.s[0]
79cabdff1aSopenharmony_ci
80cabdff1aSopenharmony_ci        add w1, w1, #2
81cabdff1aSopenharmony_ci        sub x1, x0, x1, lsl #2
82cabdff1aSopenharmony_ci
83cabdff1aSopenharmony_ci        ld1 {v3.4s}, [x1]
84cabdff1aSopenharmony_ci        fmul v3.4s, v3.4s, v2.4s
85cabdff1aSopenharmony_ci
86cabdff1aSopenharmony_ci1:      add x1, x1, #4
87cabdff1aSopenharmony_ci        ld1 {v4.4s}, [x1]
88cabdff1aSopenharmony_ci        add x1, x1, #4
89cabdff1aSopenharmony_ci        ld1 {v5.4s}, [x1]
90cabdff1aSopenharmony_ci        add x1, x1, #4
91cabdff1aSopenharmony_ci        ld1 {v6.4s}, [x1]
92cabdff1aSopenharmony_ci        add x1, x1, #4
93cabdff1aSopenharmony_ci        ld1 {v7.4s}, [x1]
94cabdff1aSopenharmony_ci
95cabdff1aSopenharmony_ci        fmla v3.4s, v7.4s, v2.4s
96cabdff1aSopenharmony_ci        fadd v6.4s, v6.4s, v4.4s
97cabdff1aSopenharmony_ci
98cabdff1aSopenharmony_ci        ld1 {v4.4s}, [x0]
99cabdff1aSopenharmony_ci        fmla v4.4s, v5.4s, v0.4s
100cabdff1aSopenharmony_ci
101cabdff1aSopenharmony_ci        fmul v6.4s, v6.4s, v1.4s
102cabdff1aSopenharmony_ci        fadd v6.4s, v6.4s, v3.4s
103cabdff1aSopenharmony_ci
104cabdff1aSopenharmony_ci        fadd v4.4s, v4.4s, v6.4s
105cabdff1aSopenharmony_ci        fmul v3.4s, v7.4s, v2.4s
106cabdff1aSopenharmony_ci
107cabdff1aSopenharmony_ci        st1  {v4.4s}, [x0], #16
108cabdff1aSopenharmony_ci
109cabdff1aSopenharmony_ci        subs w3, w3, #4
110cabdff1aSopenharmony_ci        b.gt 1b
111cabdff1aSopenharmony_ci
112cabdff1aSopenharmony_ci        ret
113cabdff1aSopenharmony_ciendfunc
114