|
12367
|
1 ;*****************************************************************************
|
|
|
2 ;* SSE2-optimized weighted prediction code
|
|
|
3 ;*****************************************************************************
|
|
|
4 ;* Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
|
|
|
5 ;* Copyright (C) 2010 Eli Friedman <eli.friedman@gmail.com>
|
|
|
6 ;*
|
|
|
7 ;* This file is part of FFmpeg.
|
|
|
8 ;*
|
|
|
9 ;* FFmpeg is free software; you can redistribute it and/or
|
|
|
10 ;* modify it under the terms of the GNU Lesser General Public
|
|
|
11 ;* License as published by the Free Software Foundation; either
|
|
|
12 ;* version 2.1 of the License, or (at your option) any later version.
|
|
|
13 ;*
|
|
|
14 ;* FFmpeg is distributed in the hope that it will be useful,
|
|
|
15 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
16 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
17 ;* Lesser General Public License for more details.
|
|
|
18 ;*
|
|
|
19 ;* You should have received a copy of the GNU Lesser General Public
|
|
|
20 ;* License along with FFmpeg; if not, write to the Free Software
|
|
|
21 ;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
22 ;******************************************************************************
|
|
|
23
|
|
|
24 %include "x86inc.asm"
|
|
|
25
|
|
|
26 SECTION .text
|
|
|
27 INIT_XMM
|
|
|
28
|
|
|
29 ;-----------------------------------------------------------------------------
|
|
|
30 ; biweight pred:
|
|
|
31 ;
|
|
|
32 ; void h264_biweight_16x16_sse2(uint8_t *dst, uint8_t *src, int stride,
|
|
|
33 ; int log2_denom, int weightd, int weights,
|
|
|
34 ; int offset);
|
|
|
35 ;-----------------------------------------------------------------------------
|
|
|
36
|
|
|
37 %macro BIWEIGHT_SSE2_SETUP 0
|
|
|
38 add r6, 1
|
|
|
39 or r6, 1
|
|
|
40 add r3, 1
|
|
|
41 movd m3, r4
|
|
|
42 movd m4, r5
|
|
|
43 movd m5, r6
|
|
|
44 movd m6, r3
|
|
|
45 pslld m5, m6
|
|
|
46 psrld m5, 1
|
|
|
47 pshuflw m3, m3, 0
|
|
|
48 pshuflw m4, m4, 0
|
|
|
49 pshuflw m5, m5, 0
|
|
|
50 punpcklqdq m3, m3
|
|
|
51 punpcklqdq m4, m4
|
|
|
52 punpcklqdq m5, m5
|
|
|
53 pxor m7, m7
|
|
|
54 %endmacro
|
|
|
55
|
|
|
56 %macro BIWEIGHT_SSE2_STEPA 3
|
|
|
57 movh m%1, [r0+%3]
|
|
|
58 movh m%2, [r1+%3]
|
|
|
59 punpcklbw m%1, m7
|
|
|
60 punpcklbw m%2, m7
|
|
|
61 pmullw m%1, m3
|
|
|
62 pmullw m%2, m4
|
|
|
63 paddsw m%1, m%2
|
|
|
64 %endmacro
|
|
|
65
|
|
|
66 %macro BIWEIGHT_SSE2_STEPB 0
|
|
|
67 paddsw m0, m5
|
|
|
68 paddsw m1, m5
|
|
|
69 psraw m0, m6
|
|
|
70 psraw m1, m6
|
|
|
71 packuswb m0, m1
|
|
|
72 %endmacro
|
|
|
73
|
|
|
74 cglobal h264_biweight_16x16_sse2, 7, 7, 8
|
|
|
75 BIWEIGHT_SSE2_SETUP
|
|
|
76 mov r3, 16
|
|
|
77
|
|
|
78 .nextrow
|
|
|
79 BIWEIGHT_SSE2_STEPA 0, 1, 0
|
|
|
80 BIWEIGHT_SSE2_STEPA 1, 2, 8
|
|
|
81 BIWEIGHT_SSE2_STEPB
|
|
|
82 mova [r0], m0
|
|
|
83 add r0, r2
|
|
|
84 add r1, r2
|
|
|
85 dec r3
|
|
|
86 jnz .nextrow
|
|
|
87 REP_RET
|
|
|
88
|
|
|
89 cglobal h264_biweight_8x8_sse2, 7, 7, 8
|
|
|
90 BIWEIGHT_SSE2_SETUP
|
|
|
91 mov r3, 4
|
|
|
92 lea r4, [r2*2]
|
|
|
93
|
|
|
94 .nextrow
|
|
|
95 BIWEIGHT_SSE2_STEPA 0, 1, 0
|
|
|
96 BIWEIGHT_SSE2_STEPA 1, 2, r2
|
|
|
97 BIWEIGHT_SSE2_STEPB
|
|
|
98 movh [r0], m0
|
|
|
99 movhps [r0+r2], m0
|
|
|
100 add r0, r4
|
|
|
101 add r1, r4
|
|
|
102 dec r3
|
|
|
103 jnz .nextrow
|
|
|
104 REP_RET
|
|
|
105
|
|
|
106 %macro BIWEIGHT_SSSE3_SETUP 0
|
|
|
107 add r6, 1
|
|
|
108 or r6, 1
|
|
|
109 add r3, 1
|
|
|
110 movd m4, r4
|
|
|
111 movd m0, r5
|
|
|
112 movd m5, r6
|
|
|
113 movd m6, r3
|
|
|
114 pslld m5, m6
|
|
|
115 psrld m5, 1
|
|
|
116 punpcklbw m4, m0
|
|
|
117 pshuflw m4, m4, 0
|
|
|
118 pshuflw m5, m5, 0
|
|
|
119 punpcklqdq m4, m4
|
|
|
120 punpcklqdq m5, m5
|
|
|
121 %endmacro
|
|
|
122
|
|
|
123 %macro BIWEIGHT_SSSE3_OP 0
|
|
|
124 pmaddubsw m0, m4
|
|
|
125 pmaddubsw m2, m4
|
|
|
126 paddsw m0, m5
|
|
|
127 paddsw m2, m5
|
|
|
128 psraw m0, m6
|
|
|
129 psraw m2, m6
|
|
|
130 packuswb m0, m2
|
|
|
131 %endmacro
|
|
|
132
|
|
|
133 cglobal h264_biweight_16x16_ssse3, 7, 7, 8
|
|
|
134 BIWEIGHT_SSSE3_SETUP
|
|
|
135 mov r3, 16
|
|
|
136
|
|
|
137 .nextrow
|
|
|
138 movh m0, [r0]
|
|
|
139 movh m2, [r0+8]
|
|
|
140 movh m3, [r1+8]
|
|
|
141 punpcklbw m0, [r1]
|
|
|
142 punpcklbw m2, m3
|
|
|
143 BIWEIGHT_SSSE3_OP
|
|
|
144 mova [r0], m0
|
|
|
145 add r0, r2
|
|
|
146 add r1, r2
|
|
|
147 dec r3
|
|
|
148 jnz .nextrow
|
|
|
149 REP_RET
|
|
|
150
|
|
|
151 cglobal h264_biweight_8x8_ssse3, 7, 7, 8
|
|
|
152 BIWEIGHT_SSSE3_SETUP
|
|
|
153 mov r3, 4
|
|
|
154 lea r4, [r2*2]
|
|
|
155
|
|
|
156 .nextrow
|
|
|
157 movh m0, [r0]
|
|
|
158 movh m1, [r1]
|
|
|
159 movh m2, [r0+r2]
|
|
|
160 movh m3, [r1+r2]
|
|
|
161 punpcklbw m0, m1
|
|
|
162 punpcklbw m2, m3
|
|
|
163 BIWEIGHT_SSSE3_OP
|
|
|
164 movh [r0], m0
|
|
|
165 movhps [r0+r2], m0
|
|
|
166 add r0, r4
|
|
|
167 add r1, r4
|
|
|
168 dec r3
|
|
|
169 jnz .nextrow
|
|
|
170 REP_RET
|
