Mercurial > libavcodec.hg
annotate libpostproc/postprocess.c @ 2038:02b59a3c62cd libavcodec
faster c lowpass filter
author | michael |
---|---|
date | Thu, 27 May 2004 16:16:09 +0000 |
parents | 98d8283534bb |
children | f25e485a7850 |
rev | line source |
---|---|
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
1 /* |
1067 | 2 Copyright (C) 2001-2003 Michael Niedermayer (michaelni@gmx.at) |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
3 |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
4 AltiVec optimizations (C) 2004 Romain Dolbeau <romain@dolbeau.org> |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
5 |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
6 This program is free software; you can redistribute it and/or modify |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
7 it under the terms of the GNU General Public License as published by |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
8 the Free Software Foundation; either version 2 of the License, or |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
9 (at your option) any later version. |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
10 |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
11 This program is distributed in the hope that it will be useful, |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
12 but WITHOUT ANY WARRANTY; without even the implied warranty of |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
14 GNU General Public License for more details. |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
15 |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
16 You should have received a copy of the GNU General Public License |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
17 along with this program; if not, write to the Free Software |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
18 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
19 */ |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
20 |
1109 | 21 /** |
22 * @file postprocess.c | |
23 * postprocessing. | |
24 */ | |
25 | |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
26 /* |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
27 C MMX MMX2 3DNow AltiVec |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
28 isVertDC Ec Ec Ec |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
29 isVertMinMaxOk Ec Ec Ec |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
30 doVertLowPass E e e Ec |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
31 doVertDefFilter Ec Ec e e Ec |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
32 isHorizDC Ec Ec |
128
e5266b8e79be
much better horizontal filters (transpose & use the vertical ones) :)
michael
parents:
126
diff
changeset
|
33 isHorizMinMaxOk a E |
e5266b8e79be
much better horizontal filters (transpose & use the vertical ones) :)
michael
parents:
126
diff
changeset
|
34 doHorizLowPass E e e |
163 | 35 doHorizDefFilter Ec Ec e e |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
36 deRing E e e* Ecp |
106
389391a6d0bf
rewrote the horizontal lowpass filter to fix a bug which caused a blocky look
michael
parents:
105
diff
changeset
|
37 Vertical RKAlgo1 E a a |
129
be35346e27c1
fixed difference with -vo md5 between doVertDefFilter() C and MMX / MMX2 versions
michael
parents:
128
diff
changeset
|
38 Horizontal RKAlgo1 a a |
156 | 39 Vertical X1# a E E |
40 Horizontal X1# a E E | |
111 | 41 LinIpolDeinterlace e E E* |
42 CubicIpolDeinterlace a e e* | |
43 LinBlendDeinterlace e E E* | |
1029
804cc05a3f61
C implementation of the median deinterlacer (seems to be the only one
rfelker
parents:
957
diff
changeset
|
44 MedianDeinterlace# E Ec Ec |
157
bc12fd7e6153
temp denoiser changes: (a-b)^2 instead of |a-b| and MMX2/3DNOW version
michael
parents:
156
diff
changeset
|
45 TempDeNoiser# E e e |
156 | 46 |
47 * i dont have a 3dnow CPU -> its untested, but noone said it doesnt work so it seems to work | |
48 # more or less selfinvented filters so the exactness isnt too meaningfull | |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
49 E = Exact implementation |
111 | 50 e = allmost exact implementation (slightly different rounding,...) |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
51 a = alternative / approximate impl |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
52 c = checked against the other implementations (-vo md5) |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
53 p = partially optimized, still some work to do |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
54 */ |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
55 |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
56 /* |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
57 TODO: |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
58 reduce the time wasted on the mem transfer |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
59 unroll stuff if instructions depend too much on the prior one |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
60 move YScale thing to the end instead of fixing QP |
96 | 61 write a faster and higher quality deblocking filter :) |
97
e57b1d38d71f
bugfixes: last 3 lines not brightness/contrast corrected
michael
parents:
96
diff
changeset
|
62 make the mainloop more flexible (variable number of blocks at once |
e57b1d38d71f
bugfixes: last 3 lines not brightness/contrast corrected
michael
parents:
96
diff
changeset
|
63 (the if/else stuff per block is slowing things down) |
99 | 64 compare the quality & speed of all filters |
65 split this huge file | |
140 | 66 optimize c versions |
156 | 67 try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
68 ... |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
69 */ |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
70 |
107 | 71 //Changelog: use the CVS log |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
72 |
1067 | 73 #include "config.h" |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
74 #include <inttypes.h> |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
75 #include <stdio.h> |
97
e57b1d38d71f
bugfixes: last 3 lines not brightness/contrast corrected
michael
parents:
96
diff
changeset
|
76 #include <stdlib.h> |
116 | 77 #include <string.h> |
133 | 78 #ifdef HAVE_MALLOC_H |
79 #include <malloc.h> | |
80 #endif | |
96 | 81 //#undef HAVE_MMX2 |
82 //#define HAVE_3DNOW | |
83 //#undef HAVE_MMX | |
169 | 84 //#undef ARCH_X86 |
163 | 85 //#define DEBUG_BRIGHTNESS |
1069 | 86 #ifdef USE_FASTMEMCPY |
1775 | 87 #include "fastmemcpy.h" |
837 | 88 #endif |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
89 #include "postprocess.h" |
829 | 90 #include "postprocess_internal.h" |
1069 | 91 |
92 #include "mangle.h" //FIXME should be supressed | |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
93 |
1071 | 94 #ifndef HAVE_MEMALIGN |
95 #define memalign(a,b) malloc(b) | |
96 #endif | |
97 | |
104 | 98 #define MIN(a,b) ((a) > (b) ? (b) : (a)) |
99 #define MAX(a,b) ((a) < (b) ? (b) : (a)) | |
100 #define ABS(a) ((a) > 0 ? (a) : (-(a))) | |
101 #define SIGN(a) ((a) > 0 ? 1 : -1) | |
102 | |
116 | 103 #define GET_MODE_BUFFER_SIZE 500 |
104 #define OPTIONS_ARRAY_SIZE 10 | |
787 | 105 #define BLOCK_SIZE 8 |
106 #define TEMP_STRIDE 8 | |
107 //#define NUM_BLOCKS_AT_ONCE 16 //not used yet | |
116 | 108 |
1847
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
109 #if defined(__GNUC__) && (__GNUC__ > 3 || __GNUC__ == 3 && __GNUC_MINOR__ > 0) |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
110 # define attribute_used __attribute__((used)) |
2037
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
111 # define always_inline __attribute__((always_inline)) inline |
1847
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
112 #else |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
113 # define attribute_used |
2037
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
114 # define always_inline inline |
1847
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
115 #endif |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
116 |
169 | 117 #ifdef ARCH_X86 |
1847
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
118 static uint64_t __attribute__((aligned(8))) attribute_used w05= 0x0005000500050005LL; |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
119 static uint64_t __attribute__((aligned(8))) attribute_used w20= 0x0020002000200020LL; |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
120 static uint64_t __attribute__((aligned(8))) attribute_used b00= 0x0000000000000000LL; |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
121 static uint64_t __attribute__((aligned(8))) attribute_used b01= 0x0101010101010101LL; |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
122 static uint64_t __attribute__((aligned(8))) attribute_used b02= 0x0202020202020202LL; |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
123 static uint64_t __attribute__((aligned(8))) attribute_used b08= 0x0808080808080808LL; |
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
124 static uint64_t __attribute__((aligned(8))) attribute_used b80= 0x8080808080808080LL; |
148
1cfc4d567c0a
minor changes (fixed some warnings, added attribute aligned(8) stuff)
michael
parents:
142
diff
changeset
|
125 #endif |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
126 |
1157 | 127 static uint8_t clip_table[3*256]; |
128 static uint8_t * const clip_tab= clip_table + 256; | |
129 | |
2031
4225c131a2eb
warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents:
2024
diff
changeset
|
130 static const int verbose= 0; |
179 | 131 |
1847
ef661c4dc5a6
attribute_used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))
michael
parents:
1775
diff
changeset
|
132 static const int attribute_used deringThreshold= 20; |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
133 |
787 | 134 |
116 | 135 static struct PPFilter filters[]= |
136 { | |
137 {"hb", "hdeblock", 1, 1, 3, H_DEBLOCK}, | |
138 {"vb", "vdeblock", 1, 2, 4, V_DEBLOCK}, | |
787 | 139 /* {"hr", "rkhdeblock", 1, 1, 3, H_RK1_FILTER}, |
140 {"vr", "rkvdeblock", 1, 2, 4, V_RK1_FILTER},*/ | |
116 | 141 {"h1", "x1hdeblock", 1, 1, 3, H_X1_FILTER}, |
142 {"v1", "x1vdeblock", 1, 2, 4, V_X1_FILTER}, | |
2037
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
143 {"ha", "ahdeblock", 1, 1, 3, H_A_DEBLOCK}, |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
144 {"va", "avdeblock", 1, 2, 4, V_A_DEBLOCK}, |
116 | 145 {"dr", "dering", 1, 5, 6, DERING}, |
146 {"al", "autolevels", 0, 1, 2, LEVEL_FIX}, | |
181 | 147 {"lb", "linblenddeint", 1, 1, 4, LINEAR_BLEND_DEINT_FILTER}, |
148 {"li", "linipoldeint", 1, 1, 4, LINEAR_IPOL_DEINT_FILTER}, | |
149 {"ci", "cubicipoldeint", 1, 1, 4, CUBIC_IPOL_DEINT_FILTER}, | |
150 {"md", "mediandeint", 1, 1, 4, MEDIAN_DEINT_FILTER}, | |
787 | 151 {"fd", "ffmpegdeint", 1, 1, 4, FFMPEG_DEINT_FILTER}, |
1157 | 152 {"l5", "lowpass5", 1, 1, 4, LOWPASS5_DEINT_FILTER}, |
156 | 153 {"tn", "tmpnoise", 1, 7, 8, TEMP_NOISE_FILTER}, |
183 | 154 {"fq", "forcequant", 1, 0, 0, FORCE_QUANT}, |
116 | 155 {NULL, NULL,0,0,0,0} //End Marker |
156 }; | |
157 | |
158 static char *replaceTable[]= | |
159 { | |
2037
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
160 "default", "hdeblock:a,vdeblock:a,dering:a", |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
161 "de", "hdeblock:a,vdeblock:a,dering:a", |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
162 "fast", "x1hdeblock:a,x1vdeblock:a,dering:a", |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
163 "fa", "x1hdeblock:a,x1vdeblock:a,dering:a", |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
164 "ac", "ha:a:128:7,va:a,dering:a", |
116 | 165 NULL //End Marker |
166 }; | |
167 | |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
168 |
787 | 169 #ifdef ARCH_X86 |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
170 static inline void prefetchnta(void *p) |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
171 { |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
172 asm volatile( "prefetchnta (%0)\n\t" |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
173 : : "r" (p) |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
174 ); |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
175 } |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
176 |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
177 static inline void prefetcht0(void *p) |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
178 { |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
179 asm volatile( "prefetcht0 (%0)\n\t" |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
180 : : "r" (p) |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
181 ); |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
182 } |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
183 |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
184 static inline void prefetcht1(void *p) |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
185 { |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
186 asm volatile( "prefetcht1 (%0)\n\t" |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
187 : : "r" (p) |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
188 ); |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
189 } |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
190 |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
191 static inline void prefetcht2(void *p) |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
192 { |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
193 asm volatile( "prefetcht2 (%0)\n\t" |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
194 : : "r" (p) |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
195 ); |
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
196 } |
102 | 197 #endif |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
198 |
169 | 199 // The horizontal Functions exist only in C cuz the MMX code is faster with vertical filters and transposing |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
200 |
165 | 201 /** |
202 * Check if the given 8x8 Block is mostly "flat" | |
203 */ | |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
204 static inline int isHorizDC_C(uint8_t src[], int stride, PPContext *c) |
165 | 205 { |
206 int numEq= 0; | |
207 int y; | |
1196 | 208 const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1; |
791 | 209 const int dcThreshold= dcOffset*2 + 1; |
1196 | 210 |
165 | 211 for(y=0; y<BLOCK_SIZE; y++) |
212 { | |
787 | 213 if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++; |
214 if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++; | |
215 if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++; | |
216 if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++; | |
217 if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++; | |
218 if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++; | |
219 if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++; | |
165 | 220 src+= stride; |
221 } | |
787 | 222 return numEq > c->ppMode.flatnessThreshold; |
223 } | |
224 | |
225 /** | |
226 * Check if the middle 8x8 Block in the given 8x16 block is flat | |
227 */ | |
228 static inline int isVertDC_C(uint8_t src[], int stride, PPContext *c){ | |
229 int numEq= 0; | |
230 int y; | |
1196 | 231 const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1; |
791 | 232 const int dcThreshold= dcOffset*2 + 1; |
1196 | 233 |
787 | 234 src+= stride*4; // src points to begin of the 8x8 Block |
235 for(y=0; y<BLOCK_SIZE-1; y++) | |
236 { | |
237 if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++; | |
238 if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++; | |
239 if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++; | |
240 if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++; | |
241 if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++; | |
242 if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++; | |
243 if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++; | |
244 if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++; | |
245 src+= stride; | |
246 } | |
247 return numEq > c->ppMode.flatnessThreshold; | |
165 | 248 } |
249 | |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
250 static inline int isHorizMinMaxOk_C(uint8_t src[], int stride, int QP) |
165 | 251 { |
1327
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
252 int i; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
253 #if 1 |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
254 for(i=0; i<2; i++){ |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
255 if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
256 src += stride; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
257 if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
258 src += stride; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
259 if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
260 src += stride; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
261 if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
262 src += stride; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
263 } |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
264 #else |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
265 for(i=0; i<8; i++){ |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
266 if((unsigned)(src[0] - src[7] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
267 src += stride; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
268 } |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
269 #endif |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
270 return 1; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
271 } |
165 | 272 |
1327
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
273 static inline int isVertMinMaxOk_C(uint8_t src[], int stride, int QP) |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
274 { |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
275 #if 1 |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
276 #if 1 |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
277 int x; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
278 src+= stride*4; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
279 for(x=0; x<BLOCK_SIZE; x+=4) |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
280 { |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
281 if((unsigned)(src[ x + 0*stride] - src[ x + 5*stride] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
282 if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
283 if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
284 if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
285 } |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
286 #else |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
287 int x; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
288 src+= stride*3; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
289 for(x=0; x<BLOCK_SIZE; x++) |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
290 { |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
291 if((unsigned)(src[x + stride] - src[x + (stride<<3)] + 2*QP) > 4*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
292 } |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
293 #endif |
165 | 294 return 1; |
1327
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
295 #else |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
296 int x; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
297 src+= stride*4; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
298 for(x=0; x<BLOCK_SIZE; x++) |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
299 { |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
300 int min=255; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
301 int max=0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
302 int y; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
303 for(y=0; y<8; y++){ |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
304 int v= src[x + y*stride]; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
305 if(v>max) max=v; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
306 if(v<min) min=v; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
307 } |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
308 if(max-min > 2*QP) return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
309 } |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
310 return 1; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
311 #endif |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
312 } |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
313 |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
314 static inline int horizClassify_C(uint8_t src[], int stride, PPContext *c){ |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
315 if( isHorizDC_C(src, stride, c) ){ |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
316 if( isHorizMinMaxOk_C(src, stride, c->QP) ) |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
317 return 1; |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
318 else |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
319 return 0; |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
320 }else{ |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
321 return 2; |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
322 } |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
323 } |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
324 |
1327
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
325 static inline int vertClassify_C(uint8_t src[], int stride, PPContext *c){ |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
326 if( isVertDC_C(src, stride, c) ){ |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
327 if( isVertMinMaxOk_C(src, stride, c->QP) ) |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
328 return 1; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
329 else |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
330 return 0; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
331 }else{ |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
332 return 2; |
854571532c89
blinking blocks around thin vertical lines and dots bugfix
michaelni
parents:
1282
diff
changeset
|
333 } |
165 | 334 } |
335 | |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
336 static inline void doHorizDefFilter_C(uint8_t dst[], int stride, PPContext *c) |
165 | 337 { |
338 int y; | |
339 for(y=0; y<BLOCK_SIZE; y++) | |
340 { | |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
341 const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]); |
165 | 342 |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
343 if(ABS(middleEnergy) < 8*c->QP) |
165 | 344 { |
345 const int q=(dst[3] - dst[4])/2; | |
346 const int leftEnergy= 5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]); | |
347 const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]); | |
348 | |
349 int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) ); | |
350 d= MAX(d, 0); | |
351 | |
352 d= (5*d + 32) >> 6; | |
353 d*= SIGN(-middleEnergy); | |
354 | |
355 if(q>0) | |
356 { | |
357 d= d<0 ? 0 : d; | |
358 d= d>q ? q : d; | |
359 } | |
360 else | |
361 { | |
362 d= d>0 ? 0 : d; | |
363 d= d<q ? q : d; | |
364 } | |
365 | |
366 dst[3]-= d; | |
367 dst[4]+= d; | |
368 } | |
369 dst+= stride; | |
370 } | |
371 } | |
372 | |
373 /** | |
374 * Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block) | |
375 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version) | |
376 */ | |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
377 static inline void doHorizLowPass_C(uint8_t dst[], int stride, PPContext *c) |
165 | 378 { |
379 int y; | |
380 for(y=0; y<BLOCK_SIZE; y++) | |
381 { | |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
382 const int first= ABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0]; |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
383 const int last= ABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7]; |
165 | 384 |
2038 | 385 int sums[10]; |
386 sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4; | |
387 sums[1] = sums[0] - first + dst[3]; | |
388 sums[2] = sums[1] - first + dst[4]; | |
389 sums[3] = sums[2] - first + dst[5]; | |
390 sums[4] = sums[3] - first + dst[6]; | |
391 sums[5] = sums[4] - dst[0] + dst[7]; | |
392 sums[6] = sums[5] - dst[1] + last; | |
393 sums[7] = sums[6] - dst[2] + last; | |
394 sums[8] = sums[7] - dst[3] + last; | |
395 sums[9] = sums[8] - dst[4] + last; | |
165 | 396 |
2038 | 397 dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4; |
398 dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4; | |
399 dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4; | |
400 dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4; | |
401 dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4; | |
402 dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4; | |
403 dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4; | |
404 dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4; | |
165 | 405 |
406 dst+= stride; | |
407 } | |
408 } | |
409 | |
169 | 410 /** |
411 * Experimental Filter 1 (Horizontal) | |
412 * will not damage linear gradients | |
413 * Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter | |
414 * can only smooth blocks at the expected locations (it cant smooth them if they did move) | |
415 * MMX2 version does correct clipping C version doesnt | |
416 * not identical with the vertical one | |
417 */ | |
418 static inline void horizX1Filter(uint8_t *src, int stride, int QP) | |
95
8bce253b537c
new postprocess code by Michael Niedermayer (michaelni@gmx.at)
arpi
parents:
diff
changeset
|
419 { |
169 | 420 int y; |
421 static uint64_t *lut= NULL; | |
422 if(lut==NULL) | |
423 { | |
424 int i; | |
425 lut= (uint64_t*)memalign(8, 256*8); | |
426 for(i=0; i<256; i++) | |
427 { | |
428 int v= i < 128 ? 2*i : 2*(i-256); | |
429 /* | |
430 //Simulate 112242211 9-Tap filter | |
431 uint64_t a= (v/16) & 0xFF; | |
432 uint64_t b= (v/8) & 0xFF; | |
433 uint64_t c= (v/4) & 0xFF; | |
434 uint64_t d= (3*v/8) & 0xFF; | |
435 */ | |
436 //Simulate piecewise linear interpolation | |
437 uint64_t a= (v/16) & 0xFF; | |
438 uint64_t b= (v*3/16) & 0xFF; | |
439 uint64_t c= (v*5/16) & 0xFF; | |
440 uint64_t d= (7*v/16) & 0xFF; | |
441 uint64_t A= (0x100 - a)&0xFF; | |
442 uint64_t B= (0x100 - b)&0xFF; | |
443 uint64_t C= (0x100 - c)&0xFF; | |
444 uint64_t D= (0x100 - c)&0xFF; | |
130 | 445 |
169 | 446 lut[i] = (a<<56) | (b<<48) | (c<<40) | (d<<32) | |
447 (D<<24) | (C<<16) | (B<<8) | (A); | |
448 //lut[i] = (v<<32) | (v<<24); | |
134 | 449 } |
450 } | |
451 | |
169 | 452 for(y=0; y<BLOCK_SIZE; y++) |
134 | 453 { |
169 | 454 int a= src[1] - src[2]; |
455 int b= src[3] - src[4]; | |
456 int c= src[5] - src[6]; | |
134 | 457 |
169 | 458 int d= MAX(ABS(b) - (ABS(a) + ABS(c))/2, 0); |
167 | 459 |
169 | 460 if(d < QP) |
167 | 461 { |
169 | 462 int v = d * SIGN(-b); |
106
389391a6d0bf
rewrote the horizontal lowpass filter to fix a bug which caused a blocky look
michael
parents:
105
diff
changeset
|
463 |
169 | 464 src[1] +=v/8; |
465 src[2] +=v/4; | |
466 src[3] +=3*v/8; | |
467 src[4] -=3*v/8; | |
468 src[5] -=v/4; | |
469 src[6] -=v/8; | |
106
389391a6d0bf
rewrote the horizontal lowpass filter to fix a bug which caused a blocky look
michael
parents:
105
diff
changeset
|
470 |
169 | 471 } |
472 src+=stride; | |
106
389391a6d0bf
rewrote the horizontal lowpass filter to fix a bug which caused a blocky look
michael
parents:
105
diff
changeset
|
473 } |
389391a6d0bf
rewrote the horizontal lowpass filter to fix a bug which caused a blocky look
michael
parents:
105
diff
changeset
|
474 } |
389391a6d0bf
rewrote the horizontal lowpass filter to fix a bug which caused a blocky look
michael
parents:
105
diff
changeset
|
475 |
2037
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
476 /** |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
477 * accurate deblock filter |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
478 */ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
479 static always_inline void do_a_deblock(uint8_t *src, int step, int stride, PPContext *c){ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
480 int y; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
481 const int QP= c->QP; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
482 const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
483 const int dcThreshold= dcOffset*2 + 1; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
484 //START_TIMER |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
485 src+= step*4; // src points to begin of the 8x8 Block |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
486 for(y=0; y<8; y++){ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
487 int numEq= 0; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
488 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
489 if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
490 if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
491 if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
492 if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
493 if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
494 if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
495 if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
496 if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
497 if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
498 if(numEq > c->ppMode.flatnessThreshold){ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
499 int min, max, x; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
500 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
501 if(src[0] > src[step]){ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
502 max= src[0]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
503 min= src[step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
504 }else{ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
505 max= src[step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
506 min= src[0]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
507 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
508 for(x=2; x<8; x+=2){ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
509 if(src[x*step] > src[(x+1)*step]){ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
510 if(src[x *step] > max) max= src[ x *step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
511 if(src[(x+1)*step] < min) min= src[(x+1)*step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
512 }else{ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
513 if(src[(x+1)*step] > max) max= src[(x+1)*step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
514 if(src[ x *step] < min) min= src[ x *step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
515 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
516 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
517 if(max-min < 2*QP){ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
518 const int first= ABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
519 const int last= ABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
520 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
521 int sums[10]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
522 sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
523 sums[1] = sums[0] - first + src[3*step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
524 sums[2] = sums[1] - first + src[4*step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
525 sums[3] = sums[2] - first + src[5*step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
526 sums[4] = sums[3] - first + src[6*step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
527 sums[5] = sums[4] - src[0*step] + src[7*step]; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
528 sums[6] = sums[5] - src[1*step] + last; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
529 sums[7] = sums[6] - src[2*step] + last; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
530 sums[8] = sums[7] - src[3*step] + last; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
531 sums[9] = sums[8] - src[4*step] + last; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
532 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
533 src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
534 src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
535 src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
536 src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
537 src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
538 src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
539 src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
540 src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
541 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
542 }else{ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
543 const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]); |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
544 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
545 if(ABS(middleEnergy) < 8*QP) |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
546 { |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
547 const int q=(src[3*step] - src[4*step])/2; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
548 const int leftEnergy= 5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]); |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
549 const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]); |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
550 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
551 int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) ); |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
552 d= MAX(d, 0); |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
553 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
554 d= (5*d + 32) >> 6; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
555 d*= SIGN(-middleEnergy); |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
556 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
557 if(q>0) |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
558 { |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
559 d= d<0 ? 0 : d; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
560 d= d>q ? q : d; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
561 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
562 else |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
563 { |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
564 d= d>0 ? 0 : d; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
565 d= d<q ? q : d; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
566 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
567 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
568 src[3*step]-= d; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
569 src[4*step]+= d; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
570 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
571 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
572 |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
573 src += stride; |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
574 } |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
575 /*if(step==16){ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
576 STOP_TIMER("step16") |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
577 }else{ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
578 STOP_TIMER("stepX") |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
579 }*/ |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
580 } |
106
389391a6d0bf
rewrote the horizontal lowpass filter to fix a bug which caused a blocky look
michael
parents:
105
diff
changeset
|
581 |
171 | 582 //Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one |
169 | 583 //Plain C versions |
171 | 584 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) |
585 #define COMPILE_C | |
586 #endif | |
587 | |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
588 #ifdef ARCH_POWERPC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
589 #ifdef HAVE_ALTIVEC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
590 #define COMPILE_ALTIVEC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
591 #ifndef CONFIG_DARWIN |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
592 #warning "################################################################################" |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
593 #warning "WARNING: No gcc available as of today (2004-05-25) seems to be able to compile properly some of the code under non-Darwin PPC OSes. Some functions result in wrong results, while others simply won't compile (gcc explodes after allocating 1GiB+)." |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
594 #warning "################################################################################" |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
595 #endif //CONFIG_DARWIN |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
596 #endif //HAVE_ALTIVEC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
597 #endif //ARCH_POWERPC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
598 |
787 | 599 #ifdef ARCH_X86 |
171 | 600 |
601 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
602 #define COMPILE_MMX | |
603 #endif | |
604 | |
605 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT) | |
606 #define COMPILE_MMX2 | |
607 #endif | |
608 | |
609 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
610 #define COMPILE_3DNOW | |
611 #endif | |
787 | 612 #endif //ARCH_X86 |
171 | 613 |
614 #undef HAVE_MMX | |
615 #undef HAVE_MMX2 | |
616 #undef HAVE_3DNOW | |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
617 #undef HAVE_ALTIVEC |
171 | 618 #undef ARCH_X86 |
619 | |
620 #ifdef COMPILE_C | |
169 | 621 #undef HAVE_MMX |
622 #undef HAVE_MMX2 | |
623 #undef HAVE_3DNOW | |
624 #undef ARCH_X86 | |
625 #define RENAME(a) a ## _C | |
626 #include "postprocess_template.c" | |
171 | 627 #endif |
106
389391a6d0bf
rewrote the horizontal lowpass filter to fix a bug which caused a blocky look
michael
parents:
105
diff
changeset
|
628 |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
629 #ifdef ARCH_POWERPC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
630 #ifdef COMPILE_ALTIVEC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
631 #undef RENAME |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
632 #define HAVE_ALTIVEC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
633 #define RENAME(a) a ## _altivec |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
634 #include "postprocess_altivec_template.c" |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
635 #include "postprocess_template.c" |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
636 #endif |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
637 #endif //ARCH_POWERPC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
638 |
169 | 639 //MMX versions |
171 | 640 #ifdef COMPILE_MMX |
169 | 641 #undef RENAME |
642 #define HAVE_MMX | |
643 #undef HAVE_MMX2 | |
644 #undef HAVE_3DNOW | |
645 #define ARCH_X86 | |
646 #define RENAME(a) a ## _MMX | |
647 #include "postprocess_template.c" | |
171 | 648 #endif |
128
e5266b8e79be
much better horizontal filters (transpose & use the vertical ones) :)
michael
parents:
126
diff
changeset
|
649 |
169 | 650 //MMX2 versions |
171 | 651 #ifdef COMPILE_MMX2 |
169 | 652 #undef RENAME |
653 #define HAVE_MMX | |
654 #define HAVE_MMX2 | |
655 #undef HAVE_3DNOW | |
656 #define ARCH_X86 | |
657 #define RENAME(a) a ## _MMX2 | |
658 #include "postprocess_template.c" | |
171 | 659 #endif |
128
e5266b8e79be
much better horizontal filters (transpose & use the vertical ones) :)
michael
parents:
126
diff
changeset
|
660 |
169 | 661 //3DNOW versions |
171 | 662 #ifdef COMPILE_3DNOW |
169 | 663 #undef RENAME |
664 #define HAVE_MMX | |
665 #undef HAVE_MMX2 | |
666 #define HAVE_3DNOW | |
667 #define ARCH_X86 | |
668 #define RENAME(a) a ## _3DNow | |
669 #include "postprocess_template.c" | |
171 | 670 #endif |
128
e5266b8e79be
much better horizontal filters (transpose & use the vertical ones) :)
michael
parents:
126
diff
changeset
|
671 |
169 | 672 // minor note: the HAVE_xyz is messed up after that line so dont use it |
128
e5266b8e79be
much better horizontal filters (transpose & use the vertical ones) :)
michael
parents:
126
diff
changeset
|
673 |
169 | 674 static inline void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height, |
829 | 675 QP_STORE_T QPs[], int QPStride, int isColor, pp_mode_t *vm, pp_context_t *vc) |
156 | 676 { |
787 | 677 PPContext *c= (PPContext *)vc; |
829 | 678 PPMode *ppMode= (PPMode *)vm; |
787 | 679 c->ppMode= *ppMode; //FIXME |
680 | |
169 | 681 // useing ifs here as they are faster than function pointers allthough the |
682 // difference wouldnt be messureable here but its much better because | |
683 // someone might exchange the cpu whithout restarting mplayer ;) | |
171 | 684 #ifdef RUNTIME_CPUDETECT |
787 | 685 #ifdef ARCH_X86 |
169 | 686 // ordered per speed fasterst first |
805 | 687 if(c->cpuCaps & PP_CPU_CAPS_MMX2) |
787 | 688 postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
805 | 689 else if(c->cpuCaps & PP_CPU_CAPS_3DNOW) |
787 | 690 postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
805 | 691 else if(c->cpuCaps & PP_CPU_CAPS_MMX) |
787 | 692 postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
169 | 693 else |
787 | 694 postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
157
bc12fd7e6153
temp denoiser changes: (a-b)^2 instead of |a-b| and MMX2/3DNOW version
michael
parents:
156
diff
changeset
|
695 #else |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
696 #ifdef ARCH_POWERPC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
697 #ifdef HAVE_ALTIVEC |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
698 else if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC) |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
699 postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
700 else |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
701 #endif |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
702 #endif |
787 | 703 postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
157
bc12fd7e6153
temp denoiser changes: (a-b)^2 instead of |a-b| and MMX2/3DNOW version
michael
parents:
156
diff
changeset
|
704 #endif |
171 | 705 #else //RUNTIME_CPUDETECT |
706 #ifdef HAVE_MMX2 | |
787 | 707 postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
171 | 708 #elif defined (HAVE_3DNOW) |
787 | 709 postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
171 | 710 #elif defined (HAVE_MMX) |
787 | 711 postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
2036
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
712 #elif defined (HAVE_ALTIVEC) |
6a6c678517b3
altivec optimizations and horizontal filter fix by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents:
2031
diff
changeset
|
713 postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
171 | 714 #else |
787 | 715 postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c); |
171 | 716 #endif |
717 #endif //!RUNTIME_CPUDETECT | |
156 | 718 } |
719 | |
169 | 720 //static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height, |
721 // QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode); | |
96 | 722 |
116 | 723 /* -pp Command line Help |
724 */ | |
804 | 725 char *pp_help= |
833 | 726 "<filterName>[:<option>[:<option>...]][[,|/][-]<filterName>[:<option>...]]...\n" |
184 | 727 "long form example:\n" |
833 | 728 "vdeblock:autoq/hdeblock:autoq/linblenddeint default,-vdeblock\n" |
184 | 729 "short form example:\n" |
833 | 730 "vb:a/hb:a/lb de,-vb\n" |
184 | 731 "more examples:\n" |
806 | 732 "tn:64:128:256\n" |
184 | 733 "Filters Options\n" |
734 "short long name short long option Description\n" | |
1467
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
735 "* * a autoq CPU power dependent enabler\n" |
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
736 " c chrom chrominance filtering enabled\n" |
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
737 " y nochrom chrominance filtering disabled\n" |
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
738 "hb hdeblock (2 threshold) horizontal deblocking filter\n" |
1197
d9cbc8ef5a33
better? default thresholds, if this is worse for any files, then tell us ASAP
michaelni
parents:
1196
diff
changeset
|
739 " 1. difference factor: default=32, higher -> more deblocking\n" |
d9cbc8ef5a33
better? default thresholds, if this is worse for any files, then tell us ASAP
michaelni
parents:
1196
diff
changeset
|
740 " 2. flatness threshold: default=39, lower -> more deblocking\n" |
184 | 741 " the h & v deblocking filters share these\n" |
1467
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
742 " so you can't set different thresholds for h / v\n" |
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
743 "vb vdeblock (2 threshold) vertical deblocking filter\n" |
2037
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
744 "ha hadeblock (2 threshold) horizontal deblocking filter\n" |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
745 "va vadeblock (2 threshold) vertical deblocking filter\n" |
1467
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
746 "h1 x1hdeblock experimental h deblock filter 1\n" |
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
747 "v1 x1vdeblock experimental v deblock filter 1\n" |
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
748 "dr dering deringing filter\n" |
184 | 749 "al autolevels automatic brightness / contrast\n" |
750 " f fullyrange stretch luminance to (0..255)\n" | |
751 "lb linblenddeint linear blend deinterlacer\n" | |
752 "li linipoldeint linear interpolating deinterlace\n" | |
753 "ci cubicipoldeint cubic interpolating deinterlacer\n" | |
754 "md mediandeint median deinterlacer\n" | |
787 | 755 "fd ffmpegdeint ffmpeg deinterlacer\n" |
2037
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
756 "de default hb:a,vb:a,dr:a\n" |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
757 "fa fast h1:a,v1:a,dr:a\n" |
1467
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
758 "tn tmpnoise (3 threshold) temporal noise reducer\n" |
184 | 759 " 1. <= 2. <= 3. larger -> stronger filtering\n" |
1467
a320fe172086
Ministry of English Composition fixes (courtesy of Diego Biurrun
tmmm
parents:
1327
diff
changeset
|
760 "fq forceQuant <quantizer> force quantizer\n" |
184 | 761 ; |
116 | 762 |
829 | 763 pp_mode_t *pp_get_mode_by_name_and_quality(char *name, int quality) |
116 | 764 { |
765 char temp[GET_MODE_BUFFER_SIZE]; | |
766 char *p= temp; | |
787 | 767 char *filterDelimiters= ",/"; |
116 | 768 char *optionDelimiters= ":"; |
829 | 769 struct PPMode *ppMode; |
116 | 770 char *filterToken; |
771 | |
829 | 772 ppMode= memalign(8, sizeof(PPMode)); |
773 | |
774 ppMode->lumMode= 0; | |
775 ppMode->chromMode= 0; | |
776 ppMode->maxTmpNoise[0]= 700; | |
777 ppMode->maxTmpNoise[1]= 1500; | |
778 ppMode->maxTmpNoise[2]= 3000; | |
779 ppMode->maxAllowedY= 234; | |
780 ppMode->minAllowedY= 16; | |
1197
d9cbc8ef5a33
better? default thresholds, if this is worse for any files, then tell us ASAP
michaelni
parents:
1196
diff
changeset
|
781 ppMode->baseDcDiff= 256/8; |
d9cbc8ef5a33
better? default thresholds, if this is worse for any files, then tell us ASAP
michaelni
parents:
1196
diff
changeset
|
782 ppMode->flatnessThreshold= 56-16-1; |
829 | 783 ppMode->maxClippedThreshold= 0.01; |
784 ppMode->error=0; | |
793 | 785 |
116 | 786 strncpy(temp, name, GET_MODE_BUFFER_SIZE); |
787 | |
202 | 788 if(verbose>1) printf("pp: %s\n", name); |
156 | 789 |
116 | 790 for(;;){ |
791 char *filterName; | |
830 | 792 int q= 1000000; //PP_QUALITY_MAX; |
116 | 793 int chrom=-1; |
794 char *option; | |
795 char *options[OPTIONS_ARRAY_SIZE]; | |
796 int i; | |
797 int filterNameOk=0; | |
798 int numOfUnknownOptions=0; | |
799 int enable=1; //does the user want us to enabled or disabled the filter | |
800 | |
801 filterToken= strtok(p, filterDelimiters); | |
802 if(filterToken == NULL) break; | |
156 | 803 p+= strlen(filterToken) + 1; // p points to next filterToken |
116 | 804 filterName= strtok(filterToken, optionDelimiters); |
202 | 805 if(verbose>1) printf("pp: %s::%s\n", filterToken, filterName); |
116 | 806 |
807 if(*filterName == '-') | |
808 { | |
809 enable=0; | |
810 filterName++; | |
811 } | |
156 | 812 |
116 | 813 for(;;){ //for all options |
814 option= strtok(NULL, optionDelimiters); | |
815 if(option == NULL) break; | |
816 | |
202 | 817 if(verbose>1) printf("pp: option: %s\n", option); |
116 | 818 if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality; |
819 else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0; | |
820 else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1; | |
821 else | |
822 { | |
823 options[numOfUnknownOptions] = option; | |
824 numOfUnknownOptions++; | |
825 } | |
826 if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break; | |
827 } | |
156 | 828 options[numOfUnknownOptions] = NULL; |
116 | 829 |
830 /* replace stuff from the replace Table */ | |
831 for(i=0; replaceTable[2*i]!=NULL; i++) | |
832 { | |
833 if(!strcmp(replaceTable[2*i], filterName)) | |
834 { | |
835 int newlen= strlen(replaceTable[2*i + 1]); | |
836 int plen; | |
837 int spaceLeft; | |
838 | |
839 if(p==NULL) p= temp, *p=0; //last filter | |
840 else p--, *p=','; //not last filter | |
841 | |
842 plen= strlen(p); | |
419
b71190bacce8
applied 64bit patch from Ulrich Hecht <uli at suse dot de>
alex
parents:
212
diff
changeset
|
843 spaceLeft= p - temp + plen; |
116 | 844 if(spaceLeft + newlen >= GET_MODE_BUFFER_SIZE) |
845 { | |
829 | 846 ppMode->error++; |
116 | 847 break; |
848 } | |
849 memmove(p + newlen, p, plen+1); | |
850 memcpy(p, replaceTable[2*i + 1], newlen); | |
851 filterNameOk=1; | |
852 } | |
853 } | |
854 | |
855 for(i=0; filters[i].shortName!=NULL; i++) | |
856 { | |
156 | 857 // printf("Compareing %s, %s, %s\n", filters[i].shortName,filters[i].longName, filterName); |
116 | 858 if( !strcmp(filters[i].longName, filterName) |
859 || !strcmp(filters[i].shortName, filterName)) | |
860 { | |
829 | 861 ppMode->lumMode &= ~filters[i].mask; |
862 ppMode->chromMode &= ~filters[i].mask; | |
116 | 863 |
864 filterNameOk=1; | |
865 if(!enable) break; // user wants to disable it | |
866 | |
867 if(q >= filters[i].minLumQuality) | |
829 | 868 ppMode->lumMode|= filters[i].mask; |
116 | 869 if(chrom==1 || (chrom==-1 && filters[i].chromDefault)) |
870 if(q >= filters[i].minChromQuality) | |
829 | 871 ppMode->chromMode|= filters[i].mask; |
116 | 872 |
873 if(filters[i].mask == LEVEL_FIX) | |
874 { | |
875 int o; | |
829 | 876 ppMode->minAllowedY= 16; |
877 ppMode->maxAllowedY= 234; | |
116 | 878 for(o=0; options[o]!=NULL; o++) |
182
3ccd74a91074
minor brightness/contrast bugfix / moved some global vars into ppMode
michael
parents:
181
diff
changeset
|
879 { |
116 | 880 if( !strcmp(options[o],"fullyrange") |
881 ||!strcmp(options[o],"f")) | |
882 { | |
829 | 883 ppMode->minAllowedY= 0; |
884 ppMode->maxAllowedY= 255; | |
116 | 885 numOfUnknownOptions--; |
886 } | |
182
3ccd74a91074
minor brightness/contrast bugfix / moved some global vars into ppMode
michael
parents:
181
diff
changeset
|
887 } |
116 | 888 } |
156 | 889 else if(filters[i].mask == TEMP_NOISE_FILTER) |
890 { | |
891 int o; | |
892 int numOfNoises=0; | |
893 | |
894 for(o=0; options[o]!=NULL; o++) | |
895 { | |
896 char *tail; | |
829 | 897 ppMode->maxTmpNoise[numOfNoises]= |
156 | 898 strtol(options[o], &tail, 0); |
899 if(tail!=options[o]) | |
900 { | |
901 numOfNoises++; | |
902 numOfUnknownOptions--; | |
903 if(numOfNoises >= 3) break; | |
904 } | |
905 } | |
906 } | |
2037
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
907 else if(filters[i].mask == V_DEBLOCK || filters[i].mask == H_DEBLOCK |
98d8283534bb
accurate/slow (per line instead of per block) deblock filter spport which is identical to what is recommanded in the mpeg4 spec
michael
parents:
2036
diff
changeset
|
908 || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK) |
181 | 909 { |
910 int o; | |
911 | |
912 for(o=0; options[o]!=NULL && o<2; o++) | |
913 { | |
914 char *tail; | |
915 int val= strtol(options[o], &tail, 0); | |
916 if(tail==options[o]) break; | |
917 | |
918 numOfUnknownOptions--; | |
829 | 919 if(o==0) ppMode->baseDcDiff= val; |
920 else ppMode->flatnessThreshold= val; | |
181 | 921 } |
922 } | |
183 | 923 else if(filters[i].mask == FORCE_QUANT) |
924 { | |
925 int o; | |
829 | 926 ppMode->forcedQuant= 15; |
183 | 927 |
928 for(o=0; options[o]!=NULL && o<1; o++) | |
929 { | |
930 char *tail; | |
931 int val= strtol(options[o], &tail, 0); | |
932 if(tail==options[o]) break; | |
933 | |
934 numOfUnknownOptions--; | |
829 | 935 ppMode->forcedQuant= val; |
183 | 936 } |
937 } | |
116 | 938 } |
939 } | |
829 | 940 if(!filterNameOk) ppMode->error++; |
941 ppMode->error += numOfUnknownOptions; | |
116 | 942 } |
943 | |
829 | 944 if(verbose>1) printf("pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode); |
945 if(ppMode->error) | |
946 { | |
947 fprintf(stderr, "%d errors in postprocess string \"%s\"\n", ppMode->error, name); | |
948 free(ppMode); | |
949 return NULL; | |
950 } | |
116 | 951 return ppMode; |
952 } | |
953 | |
829 | 954 void pp_free_mode(pp_mode_t *mode){ |
955 if(mode) free(mode); | |
956 } | |
957 | |
937 | 958 static void reallocAlign(void **p, int alignment, int size){ |
959 if(*p) free(*p); | |
960 *p= memalign(alignment, size); | |
961 memset(*p, 0, size); | |
962 } | |
963 | |
1196 | 964 static void reallocBuffers(PPContext *c, int width, int height, int stride, int qpStride){ |
791 | 965 int mbWidth = (width+15)>>4; |
966 int mbHeight= (height+15)>>4; | |
937 | 967 int i; |
968 | |
969 c->stride= stride; | |
1196 | 970 c->qpStride= qpStride; |
787 | 971 |
937 | 972 reallocAlign((void **)&c->tempDst, 8, stride*24); |
973 reallocAlign((void **)&c->tempSrc, 8, stride*24); | |
974 reallocAlign((void **)&c->tempBlocks, 8, 2*16*8); | |
975 reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t)); | |
787 | 976 for(i=0; i<256; i++) |
977 c->yHistogram[i]= width*height/64*15/256; | |
978 | |
979 for(i=0; i<3; i++) | |
185 | 980 { |
787 | 981 //Note:the +17*1024 is just there so i dont have to worry about r/w over te end |
937 | 982 reallocAlign((void **)&c->tempBlured[i], 8, stride*mbHeight*16 + 17*1024); |
983 reallocAlign((void **)&c->tempBluredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size | |
185 | 984 } |
937 | 985 |
1157 | 986 reallocAlign((void **)&c->deintTemp, 8, 2*width+32); |
1196 | 987 reallocAlign((void **)&c->nonBQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T)); |
988 reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T)); | |
937 | 989 reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T)); |
990 } | |
179 | 991 |
1282 | 992 static void global_init(void){ |
1157 | 993 int i; |
994 memset(clip_table, 0, 256); | |
995 for(i=256; i<512; i++) | |
996 clip_table[i]= i; | |
997 memset(clip_table+512, 0, 256); | |
998 } | |
999 | |
937 | 1000 pp_context_t *pp_get_context(int width, int height, int cpuCaps){ |
1001 PPContext *c= memalign(32, sizeof(PPContext)); | |
1002 int stride= (width+15)&(~15); //assumed / will realloc if needed | |
1196 | 1003 int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed |
937 | 1004 |
1157 | 1005 global_init(); |
1006 | |
937 | 1007 memset(c, 0, sizeof(PPContext)); |
1008 c->cpuCaps= cpuCaps; | |
957 | 1009 if(cpuCaps&PP_FORMAT){ |
1010 c->hChromaSubSample= cpuCaps&0x3; | |
1011 c->vChromaSubSample= (cpuCaps>>4)&0x3; | |
1012 }else{ | |
1013 c->hChromaSubSample= 1; | |
1014 c->vChromaSubSample= 1; | |
1015 } | |
937 | 1016 |
1196 | 1017 reallocBuffers(c, width, height, stride, qpStride); |
937 | 1018 |
787 | 1019 c->frameNum=-1; |
179 | 1020 |
787 | 1021 return c; |
179 | 1022 } |
1023 | |
792 | 1024 void pp_free_context(void *vc){ |
787 | 1025 PPContext *c = (PPContext*)vc; |
1026 int i; | |
1027 | |
1028 for(i=0; i<3; i++) free(c->tempBlured[i]); | |
1029 for(i=0; i<3; i++) free(c->tempBluredPast[i]); | |
1030 | |
1031 free(c->tempBlocks); | |
1032 free(c->yHistogram); | |
1033 free(c->tempDst); | |
1034 free(c->tempSrc); | |
1035 free(c->deintTemp); | |
1196 | 1036 free(c->stdQPTable); |
791 | 1037 free(c->nonBQPTable); |
937 | 1038 free(c->forcedQPTable); |
1039 | |
1040 memset(c, 0, sizeof(PPContext)); | |
1041 | |
787 | 1042 free(c); |
1043 } | |
1044 | |
792 | 1045 void pp_postprocess(uint8_t * src[3], int srcStride[3], |
787 | 1046 uint8_t * dst[3], int dstStride[3], |
791 | 1047 int width, int height, |
787 | 1048 QP_STORE_T *QP_store, int QPStride, |
829 | 1049 pp_mode_t *vm, void *vc, int pict_type) |
116 | 1050 { |
791 | 1051 int mbWidth = (width+15)>>4; |
1052 int mbHeight= (height+15)>>4; | |
829 | 1053 PPMode *mode = (PPMode*)vm; |
791 | 1054 PPContext *c = (PPContext*)vc; |
937 | 1055 int minStride= MAX(srcStride[0], dstStride[0]); |
1196 | 1056 |
1057 if(c->stride < minStride || c->qpStride < QPStride) | |
1058 reallocBuffers(c, width, height, | |
1059 MAX(minStride, c->stride), | |
1060 MAX(c->qpStride, QPStride)); | |
787 | 1061 |
183 | 1062 if(QP_store==NULL || (mode->lumMode & FORCE_QUANT)) |
152 | 1063 { |
183 | 1064 int i; |
937 | 1065 QP_store= c->forcedQPTable; |
787 | 1066 QPStride= 0; |
183 | 1067 if(mode->lumMode & FORCE_QUANT) |
937 | 1068 for(i=0; i<mbWidth; i++) QP_store[i]= mode->forcedQuant; |
183 | 1069 else |
937 | 1070 for(i=0; i<mbWidth; i++) QP_store[i]= 1; |
152 | 1071 } |
1196 | 1072 //printf("pict_type:%d\n", pict_type); |
1073 | |
1074 if(pict_type & PP_PICT_TYPE_QP2){ | |
1075 int i; | |
1076 const int count= mbHeight * QPStride; | |
1077 for(i=0; i<(count>>2); i++){ | |
1078 ((uint32_t*)c->stdQPTable)[i] = (((uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F; | |
1079 } | |
1080 for(i<<=2; i<count; i++){ | |
1081 c->stdQPTable[i] = QP_store[i]>>1; | |
1082 } | |
1083 QP_store= c->stdQPTable; | |
1084 } | |
1085 | |
791 | 1086 if(0){ |
1087 int x,y; | |
1088 for(y=0; y<mbHeight; y++){ | |
1089 for(x=0; x<mbWidth; x++){ | |
1090 printf("%2d ", QP_store[x + y*QPStride]); | |
1091 } | |
1092 printf("\n"); | |
1093 } | |
1094 printf("\n"); | |
1095 } | |
798 | 1096 |
1196 | 1097 if((pict_type&7)!=3) |
791 | 1098 { |
1196 | 1099 int i; |
1100 const int count= mbHeight * QPStride; | |
1101 for(i=0; i<(count>>2); i++){ | |
1724 | 1102 ((uint32_t*)c->nonBQPTable)[i] = ((uint32_t*)QP_store)[i] & 0x3F3F3F3F; |
1196 | 1103 } |
1104 for(i<<=2; i<count; i++){ | |
1724 | 1105 c->nonBQPTable[i] = QP_store[i] & 0x3F; |
791 | 1106 } |
1107 } | |
152 | 1108 |
793 | 1109 if(verbose>2) |
202 | 1110 { |
1111 printf("using npp filters 0x%X/0x%X\n", mode->lumMode, mode->chromMode); | |
1112 } | |
1113 | |
787 | 1114 postProcess(src[0], srcStride[0], dst[0], dstStride[0], |
1076 | 1115 width, height, QP_store, QPStride, 0, mode, c); |
116 | 1116 |
957 | 1117 width = (width )>>c->hChromaSubSample; |
1118 height = (height)>>c->vChromaSubSample; | |
116 | 1119 |
168
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1120 if(mode->chromMode) |
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1121 { |
787 | 1122 postProcess(src[1], srcStride[1], dst[1], dstStride[1], |
1076 | 1123 width, height, QP_store, QPStride, 1, mode, c); |
787 | 1124 postProcess(src[2], srcStride[2], dst[2], dstStride[2], |
1076 | 1125 width, height, QP_store, QPStride, 2, mode, c); |
168
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1126 } |
787 | 1127 else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2]) |
168
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1128 { |
791 | 1129 memcpy(dst[1], src[1], srcStride[1]*height); |
1130 memcpy(dst[2], src[2], srcStride[2]*height); | |
168
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1131 } |
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1132 else |
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1133 { |
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1134 int y; |
791 | 1135 for(y=0; y<height; y++) |
168
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1136 { |
791 | 1137 memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width); |
1138 memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width); | |
168
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1139 } |
712c7a115164
use fastmemcpy for chrominance if no chrominance filtering is done
michael
parents:
167
diff
changeset
|
1140 } |
116 | 1141 } |
1142 |