annotate arm/dcadsp_neon.S @ 12187:fb3fcaf3c1b6 libavcodec

Fix memleak when using mp*float decoder. Patch by flybird2k at gmail
author cehoyos
date Sun, 18 Jul 2010 19:53:24 +0000
parents 8d3539d6ba3d
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
11619
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
1 /*
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
2 * Copyright (c) 2010 Mans Rullgard <mans@mansr.com>
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
3 *
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
4 * This file is part of FFmpeg.
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
5 *
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
6 * FFmpeg is free software; you can redistribute it and/or
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
7 * modify it under the terms of the GNU Lesser General Public
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
8 * License as published by the Free Software Foundation; either
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
9 * version 2.1 of the License, or (at your option) any later version.
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
10 *
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
11 * FFmpeg is distributed in the hope that it will be useful,
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
14 * Lesser General Public License for more details.
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
15 *
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
16 * You should have received a copy of the GNU Lesser General Public
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
17 * License along with FFmpeg; if not, write to the Free Software
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
19 */
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
20
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
21 #include "asm.S"
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
22
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
23 function ff_dca_lfe_fir_neon, export=1
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
24 push {r4-r6,lr}
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
25
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
26 add r4, r0, r3, lsl #2 @ out2
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
27 add r5, r2, #256*4-16 @ cf1
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
28 sub r1, r1, #12
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
29 cmp r3, #32
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
30 moveq r6, #256/32
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
31 movne r6, #256/64
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
32 NOVFP vldr d0, [sp, #16] @ scale, bias
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
33 mov lr, #-16
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
34 1:
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
35 vmov.f32 q2, #0.0 @ v0
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
36 vmov.f32 q3, #0.0 @ v1
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
37 mov r12, r6
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
38 2:
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
39 vld1.32 {q8}, [r2,:128]! @ cf0
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
40 vld1.32 {q9}, [r5,:128], lr @ cf1
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
41 vld1.32 {q1}, [r1], lr @ in
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
42 subs r12, r12, #4
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
43 vrev64.32 q10, q8
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
44 vmla.f32 q3, q1, q9
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
45 vmla.f32 d4, d2, d21
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
46 vmla.f32 d5, d3, d20
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
47 bne 2b
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
48
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
49 add r1, r1, r6, lsl #2
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
50 subs r3, r3, #1
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
51 vadd.f32 d4, d4, d5
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
52 vadd.f32 d6, d6, d7
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
53 vpadd.f32 d4, d4, d6
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
54 vdup.32 d5, d0[1]
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
55 vmla.f32 d5, d4, d0[0]
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
56 vst1.32 {d5[0]}, [r0,:32]!
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
57 vst1.32 {d5[1]}, [r4,:32]!
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
58 bne 1b
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
59
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
60 pop {r4-r6,pc}
8d3539d6ba3d DCA: ARM/NEON optimised lfe_fir
mru
parents:
diff changeset
61 endfunc