annotate arm/vp56dsp_neon.S @ 12492:58a960d6e34c libavcodec

Rename h264_idct_sse2.asm to h264_idct.asm; move inline IDCT asm from h264dsp_mmx.c to h264_idct.asm (as yasm code). Because the loops are now coded in asm instead of C, this is (depending on the function) up to 50% faster for cases where gcc didn't do a great job at looping. Since h264_idct_add8() is now faster than the manual loop setup in h264.c, in-asm idct calling can now be enabled for chroma as well (see r16207). For MMX, this is 5% faster. For SSE2 (which isn't done for chroma if h264.c does the looping), this makes it up to 50% faster. Speed gain overall is ~0.5-1.0%.
author rbultje
date Tue, 14 Sep 2010 13:36:26 +0000
parents 1c6d78234e67
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
11666
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
1 /*
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
2 * Copyright (c) 2010 Mans Rullgard <mans@mansr.com>
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
3 *
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
4 * This file is part of FFmpeg.
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
5 *
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
6 * FFmpeg is free software; you can redistribute it and/or
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
7 * modify it under the terms of the GNU Lesser General Public
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
8 * License as published by the Free Software Foundation; either
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
9 * version 2.1 of the License, or (at your option) any later version.
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
10 *
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
11 * FFmpeg is distributed in the hope that it will be useful,
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
14 * Lesser General Public License for more details.
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
15 *
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
16 * You should have received a copy of the GNU Lesser General Public
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
17 * License along with FFmpeg; if not, write to the Free Software
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
19 */
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
20
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
21 #include "asm.S"
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
22
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
23 .macro vp6_edge_filter
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
24 vdup.16 q3, r2 @ t
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
25 vmov.i16 q13, #1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
26 vsubl.u8 q0, d20, d18 @ p[ 0] - p[-s]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
27 vsubl.u8 q1, d16, d22 @ p[-2*s] - p[ s]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
28 vsubl.u8 q14, d21, d19
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
29 vsubl.u8 q15, d17, d23
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
30 vadd.i16 q2, q0, q0 @ 2*(p[0]-p[-s])
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
31 vadd.i16 d29, d28, d28
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
32 vadd.i16 q0, q0, q1 @ p[0]-p[-s] + p[-2*s]-p[s]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
33 vadd.i16 d28, d28, d30
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
34 vadd.i16 q0, q0, q2 @ 3*(p[0]-p[-s]) + p[-2*s]-p[s]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
35 vadd.i16 d28, d28, d29
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
36 vrshr.s16 q0, q0, #3 @ v
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
37 vrshr.s16 d28, d28, #3
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
38 vsub.i16 q8, q3, q13 @ t-1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
39 vabs.s16 q1, q0 @ V
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
40 vshr.s16 q2, q0, #15 @ s
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
41 vabs.s16 d30, d28
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
42 vshr.s16 d29, d28, #15
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
43 vsub.i16 q12, q1, q3 @ V-t
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
44 vsub.i16 d31, d30, d6
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
45 vsub.i16 q12, q12, q13 @ V-t-1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
46 vsub.i16 d31, d31, d26
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
47 vcge.u16 q12, q12, q8 @ V-t-1 >= t-1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
48 vcge.u16 d31, d31, d16
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
49 vadd.i16 q13, q3, q3 @ 2*t
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
50 vadd.i16 d16, d6, d6
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
51 vsub.i16 q13, q13, q1 @ 2*t - V
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
52 vsub.i16 d16, d16, d30
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
53 vadd.i16 q13, q13, q2 @ += s
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
54 vadd.i16 d16, d16, d29
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
55 veor q13, q13, q2 @ ^= s
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
56 veor d16, d16, d29
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
57 vbif q0, q13, q12
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
58 vbif d28, d16, d31
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
59 vmovl.u8 q1, d20
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
60 vmovl.u8 q15, d21
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
61 vaddw.u8 q2, q0, d18
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
62 vaddw.u8 q3, q14, d19
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
63 vsub.i16 q1, q1, q0
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
64 vsub.i16 d30, d30, d28
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
65 vqmovun.s16 d18, q2
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
66 vqmovun.s16 d19, q3
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
67 vqmovun.s16 d20, q1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
68 vqmovun.s16 d21, q15
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
69 .endm
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
70
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
71 function ff_vp6_edge_filter_ver_neon, export=1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
72 sub r0, r0, r1, lsl #1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
73 vld1.8 {q8}, [r0], r1 @ p[-2*s]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
74 vld1.8 {q9}, [r0], r1 @ p[-s]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
75 vld1.8 {q10}, [r0], r1 @ p[0]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
76 vld1.8 {q11}, [r0] @ p[s]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
77 vp6_edge_filter
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
78 sub r0, r0, r1, lsl #1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
79 sub r1, r1, #8
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
80 vst1.8 {d18}, [r0]!
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
81 vst1.32 {d19[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
82 vst1.8 {d20}, [r0]!
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
83 vst1.32 {d21[0]}, [r0]
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
84 bx lr
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
85 endfunc
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
86
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
87 function ff_vp6_edge_filter_hor_neon, export=1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
88 sub r3, r0, #1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
89 sub r0, r0, #2
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
90 vld1.32 {d16[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
91 vld1.32 {d18[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
92 vld1.32 {d20[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
93 vld1.32 {d22[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
94 vld1.32 {d16[1]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
95 vld1.32 {d18[1]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
96 vld1.32 {d20[1]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
97 vld1.32 {d22[1]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
98 vld1.32 {d17[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
99 vld1.32 {d19[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
100 vld1.32 {d21[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
101 vld1.32 {d23[0]}, [r0], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
102 vtrn.8 q8, q9
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
103 vtrn.8 q10, q11
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
104 vtrn.16 q8, q10
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
105 vtrn.16 q9, q11
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
106 vp6_edge_filter
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
107 vtrn.8 q9, q10
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
108 vst1.16 {d18[0]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
109 vst1.16 {d20[0]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
110 vst1.16 {d18[1]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
111 vst1.16 {d20[1]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
112 vst1.16 {d18[2]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
113 vst1.16 {d20[2]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
114 vst1.16 {d18[3]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
115 vst1.16 {d20[3]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
116 vst1.16 {d19[0]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
117 vst1.16 {d21[0]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
118 vst1.16 {d19[1]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
119 vst1.16 {d21[1]}, [r3], r1
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
120 bx lr
1c6d78234e67 ARM: NEON optimised VP6 edge filter
mru
parents:
diff changeset
121 endfunc