1cabdff1aSopenharmony_ci/*
2cabdff1aSopenharmony_ci * ARM NEON optimised DSP functions
3cabdff1aSopenharmony_ci * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
4cabdff1aSopenharmony_ci *
5cabdff1aSopenharmony_ci * This file is part of FFmpeg.
6cabdff1aSopenharmony_ci *
7cabdff1aSopenharmony_ci * FFmpeg is free software; you can redistribute it and/or
8cabdff1aSopenharmony_ci * modify it under the terms of the GNU Lesser General Public
9cabdff1aSopenharmony_ci * License as published by the Free Software Foundation; either
10cabdff1aSopenharmony_ci * version 2.1 of the License, or (at your option) any later version.
11cabdff1aSopenharmony_ci *
12cabdff1aSopenharmony_ci * FFmpeg is distributed in the hope that it will be useful,
13cabdff1aSopenharmony_ci * but WITHOUT ANY WARRANTY; without even the implied warranty of
14cabdff1aSopenharmony_ci * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15cabdff1aSopenharmony_ci * Lesser General Public License for more details.
16cabdff1aSopenharmony_ci *
17cabdff1aSopenharmony_ci * You should have received a copy of the GNU Lesser General Public
18cabdff1aSopenharmony_ci * License along with FFmpeg; if not, write to the Free Software
19cabdff1aSopenharmony_ci * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20cabdff1aSopenharmony_ci */
21cabdff1aSopenharmony_ci
22cabdff1aSopenharmony_ci#include "libavutil/arm/asm.S"
23cabdff1aSopenharmony_ci
24cabdff1aSopenharmony_cifunction ff_vorbis_inverse_coupling_neon, export=1
25cabdff1aSopenharmony_ci        vmov.i32        q10, #1<<31
26cabdff1aSopenharmony_ci        subs            r2,  r2,  #4
27cabdff1aSopenharmony_ci        mov             r3,  r0
28cabdff1aSopenharmony_ci        mov             r12, r1
29cabdff1aSopenharmony_ci        beq             3f
30cabdff1aSopenharmony_ci
31cabdff1aSopenharmony_ci        vld1.32         {d24-d25},[r1,:128]!
32cabdff1aSopenharmony_ci        vld1.32         {d22-d23},[r0,:128]!
33cabdff1aSopenharmony_ci        vcle.s32        q8,  q12, #0
34cabdff1aSopenharmony_ci        vand            q9,  q11, q10
35cabdff1aSopenharmony_ci        veor            q12, q12, q9
36cabdff1aSopenharmony_ci        vand            q2,  q12, q8
37cabdff1aSopenharmony_ci        vbic            q3,  q12, q8
38cabdff1aSopenharmony_ci        vadd.f32        q12, q11, q2
39cabdff1aSopenharmony_ci        vsub.f32        q11, q11, q3
40cabdff1aSopenharmony_ci1:      vld1.32         {d2-d3},  [r1,:128]!
41cabdff1aSopenharmony_ci        vld1.32         {d0-d1},  [r0,:128]!
42cabdff1aSopenharmony_ci        vcle.s32        q8,  q1,  #0
43cabdff1aSopenharmony_ci        vand            q9,  q0,  q10
44cabdff1aSopenharmony_ci        veor            q1,  q1,  q9
45cabdff1aSopenharmony_ci        vst1.32         {d24-d25},[r3, :128]!
46cabdff1aSopenharmony_ci        vst1.32         {d22-d23},[r12,:128]!
47cabdff1aSopenharmony_ci        vand            q2,  q1,  q8
48cabdff1aSopenharmony_ci        vbic            q3,  q1,  q8
49cabdff1aSopenharmony_ci        vadd.f32        q1,  q0,  q2
50cabdff1aSopenharmony_ci        vsub.f32        q0,  q0,  q3
51cabdff1aSopenharmony_ci        subs            r2,  r2,  #8
52cabdff1aSopenharmony_ci        ble             2f
53cabdff1aSopenharmony_ci        vld1.32         {d24-d25},[r1,:128]!
54cabdff1aSopenharmony_ci        vld1.32         {d22-d23},[r0,:128]!
55cabdff1aSopenharmony_ci        vcle.s32        q8,  q12, #0
56cabdff1aSopenharmony_ci        vand            q9,  q11, q10
57cabdff1aSopenharmony_ci        veor            q12, q12, q9
58cabdff1aSopenharmony_ci        vst1.32         {d2-d3},  [r3, :128]!
59cabdff1aSopenharmony_ci        vst1.32         {d0-d1},  [r12,:128]!
60cabdff1aSopenharmony_ci        vand            q2,  q12, q8
61cabdff1aSopenharmony_ci        vbic            q3,  q12, q8
62cabdff1aSopenharmony_ci        vadd.f32        q12, q11, q2
63cabdff1aSopenharmony_ci        vsub.f32        q11, q11, q3
64cabdff1aSopenharmony_ci        b               1b
65cabdff1aSopenharmony_ci
66cabdff1aSopenharmony_ci2:      vst1.32         {d2-d3},  [r3, :128]!
67cabdff1aSopenharmony_ci        vst1.32         {d0-d1},  [r12,:128]!
68cabdff1aSopenharmony_ci        it              lt
69cabdff1aSopenharmony_ci        bxlt            lr
70cabdff1aSopenharmony_ci
71cabdff1aSopenharmony_ci3:      vld1.32         {d2-d3},  [r1,:128]
72cabdff1aSopenharmony_ci        vld1.32         {d0-d1},  [r0,:128]
73cabdff1aSopenharmony_ci        vcle.s32        q8,  q1,  #0
74cabdff1aSopenharmony_ci        vand            q9,  q0,  q10
75cabdff1aSopenharmony_ci        veor            q1,  q1,  q9
76cabdff1aSopenharmony_ci        vand            q2,  q1,  q8
77cabdff1aSopenharmony_ci        vbic            q3,  q1,  q8
78cabdff1aSopenharmony_ci        vadd.f32        q1,  q0,  q2
79cabdff1aSopenharmony_ci        vsub.f32        q0,  q0,  q3
80cabdff1aSopenharmony_ci        vst1.32         {d2-d3},  [r0,:128]!
81cabdff1aSopenharmony_ci        vst1.32         {d0-d1},  [r1,:128]!
82cabdff1aSopenharmony_ci        bx              lr
83cabdff1aSopenharmony_ciendfunc
84