Mercurial > mplayer.hg
annotate libmpcodecs/vf_spp.c @ 14022:803be9a78854
removing strange csp matching code (was copy&pasted from vf_pp where it originated from arpi 2.5 years ago) -> fixes spp+scale+x11 crash
dont disable the filter by default (100l for iive)
author | michael |
---|---|
date | Tue, 23 Nov 2004 21:21:54 +0000 |
parents | 045dbaccaccc |
children | 1fab95e4513c |
rev | line source |
---|---|
11277 | 1 /* |
2 Copyright (C) 2003 Michael Niedermayer <michaelni@gmx.at> | |
3 | |
4 This program is free software; you can redistribute it and/or modify | |
5 it under the terms of the GNU General Public License as published by | |
6 the Free Software Foundation; either version 2 of the License, or | |
7 (at your option) any later version. | |
8 | |
9 This program is distributed in the hope that it will be useful, | |
10 but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 GNU General Public License for more details. | |
13 | |
14 You should have received a copy of the GNU General Public License | |
15 along with this program; if not, write to the Free Software | |
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
17 */ | |
18 | |
19 /* | |
20 * This implementation is based on an algorithm described in | |
21 * "Aria Nosratinia Embedded Post-Processing for | |
22 * Enhancement of Compressed Images (1999)" | |
23 * (http://citeseer.nj.nec.com/nosratinia99embedded.html) | |
24 */ | |
25 | |
11334 | 26 |
11277 | 27 #include <stdio.h> |
28 #include <stdlib.h> | |
29 #include <string.h> | |
30 #include <inttypes.h> | |
31 #include <math.h> | |
32 | |
33 #include "../config.h" | |
11335 | 34 |
35 #ifdef USE_LIBAVCODEC | |
36 | |
11277 | 37 #include "../mp_msg.h" |
38 #include "../cpudetect.h" | |
11568
cf76671b3d77
Fix compilation if lavc is compiled as a shared lib. Patch by Panagiotis Issaris <takis@lumumba.luc.ac.be>
alex
parents:
11477
diff
changeset
|
39 |
cf76671b3d77
Fix compilation if lavc is compiled as a shared lib. Patch by Panagiotis Issaris <takis@lumumba.luc.ac.be>
alex
parents:
11477
diff
changeset
|
40 #ifdef USE_LIBAVCODEC_SO |
cf76671b3d77
Fix compilation if lavc is compiled as a shared lib. Patch by Panagiotis Issaris <takis@lumumba.luc.ac.be>
alex
parents:
11477
diff
changeset
|
41 #include <ffmpeg/avcodec.h> |
cf76671b3d77
Fix compilation if lavc is compiled as a shared lib. Patch by Panagiotis Issaris <takis@lumumba.luc.ac.be>
alex
parents:
11477
diff
changeset
|
42 #include <ffmpeg/dsputil.h> |
cf76671b3d77
Fix compilation if lavc is compiled as a shared lib. Patch by Panagiotis Issaris <takis@lumumba.luc.ac.be>
alex
parents:
11477
diff
changeset
|
43 #else |
11277 | 44 #include "../libavcodec/avcodec.h" |
45 #include "../libavcodec/dsputil.h" | |
11568
cf76671b3d77
Fix compilation if lavc is compiled as a shared lib. Patch by Panagiotis Issaris <takis@lumumba.luc.ac.be>
alex
parents:
11477
diff
changeset
|
46 #endif |
11277 | 47 |
48 #ifdef HAVE_MALLOC_H | |
49 #include <malloc.h> | |
50 #endif | |
51 | |
52 #include "img_format.h" | |
53 #include "mp_image.h" | |
54 #include "vf.h" | |
55 #include "../libvo/fastmemcpy.h" | |
56 | |
57 #define XMIN(a,b) ((a) < (b) ? (a) : (b)) | |
58 | |
59 //===========================================================================// | |
11477
9785bff83777
memcpy pix instead of black screen if no DR and codec provides no qscale table
michael
parents:
11335
diff
changeset
|
60 static const uint8_t __attribute__((aligned(8))) dither[8][8]={ |
11277 | 61 { 0, 48, 12, 60, 3, 51, 15, 63, }, |
62 { 32, 16, 44, 28, 35, 19, 47, 31, }, | |
63 { 8, 56, 4, 52, 11, 59, 7, 55, }, | |
64 { 40, 24, 36, 20, 43, 27, 39, 23, }, | |
65 { 2, 50, 14, 62, 1, 49, 13, 61, }, | |
66 { 34, 18, 46, 30, 33, 17, 45, 29, }, | |
67 { 10, 58, 6, 54, 9, 57, 5, 53, }, | |
68 { 42, 26, 38, 22, 41, 25, 37, 21, }, | |
69 }; | |
70 | |
11477
9785bff83777
memcpy pix instead of black screen if no DR and codec provides no qscale table
michael
parents:
11335
diff
changeset
|
71 static const uint8_t offset[127][2]= { |
11298 | 72 {0,0}, |
73 {0,0}, {4,4}, | |
74 {0,0}, {2,2}, {6,4}, {4,6}, | |
75 {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7}, | |
76 | |
77 {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3}, | |
78 {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7}, | |
79 | |
80 {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7}, | |
81 {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7}, | |
82 {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7}, | |
83 {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7}, | |
84 | |
11277 | 85 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, |
86 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0}, | |
87 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3}, | |
88 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1}, | |
89 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3}, | |
90 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1}, | |
91 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2}, | |
92 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0}, | |
93 }; | |
94 | |
95 struct vf_priv_s { | |
96 int log2_count; | |
97 int qp; | |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
98 int mode; |
11277 | 99 int mpeg2; |
100 int temp_stride; | |
101 uint8_t *src; | |
102 int16_t *temp; | |
103 AVCodecContext *avctx; | |
104 DSPContext dsp; | |
105 }; | |
106 | |
107 #define SHIFT 22 | |
108 | |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
109 static void hardthresh_c(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){ |
11277 | 110 int i; |
111 int bias= 0; //FIXME | |
112 unsigned int threshold1, threshold2; | |
11296 | 113 |
11301 | 114 threshold1= qp*((1<<4) - bias) - 1; |
11277 | 115 threshold2= (threshold1<<1); |
116 | |
117 memset(dst, 0, 64*sizeof(DCTELEM)); | |
11296 | 118 dst[0]= (src[0] + 4)>>3; |
11277 | 119 |
11296 | 120 for(i=1; i<64; i++){ |
121 int level= src[i]; | |
11277 | 122 if(((unsigned)(level+threshold1))>threshold2){ |
123 const int j= permutation[i]; | |
11296 | 124 dst[j]= (level + 4)>>3; |
11277 | 125 } |
126 } | |
127 } | |
128 | |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
129 static void softthresh_c(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){ |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
130 int i; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
131 int bias= 0; //FIXME |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
132 unsigned int threshold1, threshold2; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
133 |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
134 threshold1= qp*((1<<4) - bias) - 1; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
135 threshold2= (threshold1<<1); |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
136 |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
137 memset(dst, 0, 64*sizeof(DCTELEM)); |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
138 dst[0]= (src[0] + 4)>>3; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
139 |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
140 for(i=1; i<64; i++){ |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
141 int level= src[i]; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
142 if(((unsigned)(level+threshold1))>threshold2){ |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
143 const int j= permutation[i]; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
144 if(level>0) |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
145 dst[j]= (level - threshold1 + 4)>>3; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
146 else |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
147 dst[j]= (level + threshold1 + 4)>>3; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
148 } |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
149 } |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
150 } |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
151 |
11301 | 152 #ifdef HAVE_MMX |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
153 static void hardthresh_mmx(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){ |
11301 | 154 int bias= 0; //FIXME |
155 unsigned int threshold1; | |
156 | |
157 threshold1= qp*((1<<4) - bias) - 1; | |
158 | |
159 asm volatile( | |
160 #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \ | |
161 "movq " #src0 ", %%mm0 \n\t"\ | |
162 "movq " #src1 ", %%mm1 \n\t"\ | |
163 "movq " #src2 ", %%mm2 \n\t"\ | |
164 "movq " #src3 ", %%mm3 \n\t"\ | |
165 "psubw %%mm4, %%mm0 \n\t"\ | |
166 "psubw %%mm4, %%mm1 \n\t"\ | |
167 "psubw %%mm4, %%mm2 \n\t"\ | |
168 "psubw %%mm4, %%mm3 \n\t"\ | |
169 "paddusw %%mm5, %%mm0 \n\t"\ | |
170 "paddusw %%mm5, %%mm1 \n\t"\ | |
171 "paddusw %%mm5, %%mm2 \n\t"\ | |
172 "paddusw %%mm5, %%mm3 \n\t"\ | |
173 "paddw %%mm6, %%mm0 \n\t"\ | |
174 "paddw %%mm6, %%mm1 \n\t"\ | |
175 "paddw %%mm6, %%mm2 \n\t"\ | |
176 "paddw %%mm6, %%mm3 \n\t"\ | |
177 "psubusw %%mm6, %%mm0 \n\t"\ | |
178 "psubusw %%mm6, %%mm1 \n\t"\ | |
179 "psubusw %%mm6, %%mm2 \n\t"\ | |
180 "psubusw %%mm6, %%mm3 \n\t"\ | |
181 "psraw $3, %%mm0 \n\t"\ | |
182 "psraw $3, %%mm1 \n\t"\ | |
183 "psraw $3, %%mm2 \n\t"\ | |
184 "psraw $3, %%mm3 \n\t"\ | |
185 \ | |
186 "movq %%mm0, %%mm7 \n\t"\ | |
187 "punpcklwd %%mm2, %%mm0 \n\t" /*A*/\ | |
188 "punpckhwd %%mm2, %%mm7 \n\t" /*C*/\ | |
189 "movq %%mm1, %%mm2 \n\t"\ | |
190 "punpcklwd %%mm3, %%mm1 \n\t" /*B*/\ | |
191 "punpckhwd %%mm3, %%mm2 \n\t" /*D*/\ | |
192 "movq %%mm0, %%mm3 \n\t"\ | |
193 "punpcklwd %%mm1, %%mm0 \n\t" /*A*/\ | |
194 "punpckhwd %%mm7, %%mm3 \n\t" /*C*/\ | |
13393 | 195 "punpcklwd %%mm2, %%mm7 \n\t" /*B*/\ |
11301 | 196 "punpckhwd %%mm2, %%mm1 \n\t" /*D*/\ |
197 \ | |
198 "movq %%mm0, " #dst0 " \n\t"\ | |
199 "movq %%mm7, " #dst1 " \n\t"\ | |
200 "movq %%mm3, " #dst2 " \n\t"\ | |
201 "movq %%mm1, " #dst3 " \n\t" | |
202 | |
203 "movd %2, %%mm4 \n\t" | |
204 "movd %3, %%mm5 \n\t" | |
205 "movd %4, %%mm6 \n\t" | |
206 "packssdw %%mm4, %%mm4 \n\t" | |
207 "packssdw %%mm5, %%mm5 \n\t" | |
208 "packssdw %%mm6, %%mm6 \n\t" | |
209 "packssdw %%mm4, %%mm4 \n\t" | |
210 "packssdw %%mm5, %%mm5 \n\t" | |
211 "packssdw %%mm6, %%mm6 \n\t" | |
212 REQUANT_CORE( (%1), 8(%1), 16(%1), 24(%1), (%0), 8(%0), 64(%0), 72(%0)) | |
213 REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0)) | |
214 REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0)) | |
215 REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0)) | |
216 : : "r" (src), "r" (dst), "g" (threshold1+1), "g" (threshold1+5), "g" (threshold1-4) //FIXME maybe more accurate then needed? | |
217 ); | |
218 dst[0]= (src[0] + 4)>>3; | |
219 } | |
11994 | 220 |
221 static void softthresh_mmx(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){ | |
222 int bias= 0; //FIXME | |
223 unsigned int threshold1; | |
224 | |
225 threshold1= qp*((1<<4) - bias) - 1; | |
226 | |
227 asm volatile( | |
228 #undef REQUANT_CORE | |
229 #define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3) \ | |
230 "movq " #src0 ", %%mm0 \n\t"\ | |
231 "movq " #src1 ", %%mm1 \n\t"\ | |
232 "pxor %%mm6, %%mm6 \n\t"\ | |
233 "pxor %%mm7, %%mm7 \n\t"\ | |
234 "pcmpgtw %%mm0, %%mm6 \n\t"\ | |
235 "pcmpgtw %%mm1, %%mm7 \n\t"\ | |
236 "pxor %%mm6, %%mm0 \n\t"\ | |
237 "pxor %%mm7, %%mm1 \n\t"\ | |
238 "psubusw %%mm4, %%mm0 \n\t"\ | |
239 "psubusw %%mm4, %%mm1 \n\t"\ | |
240 "pxor %%mm6, %%mm0 \n\t"\ | |
241 "pxor %%mm7, %%mm1 \n\t"\ | |
242 "movq " #src2 ", %%mm2 \n\t"\ | |
243 "movq " #src3 ", %%mm3 \n\t"\ | |
244 "pxor %%mm6, %%mm6 \n\t"\ | |
245 "pxor %%mm7, %%mm7 \n\t"\ | |
246 "pcmpgtw %%mm2, %%mm6 \n\t"\ | |
247 "pcmpgtw %%mm3, %%mm7 \n\t"\ | |
248 "pxor %%mm6, %%mm2 \n\t"\ | |
249 "pxor %%mm7, %%mm3 \n\t"\ | |
250 "psubusw %%mm4, %%mm2 \n\t"\ | |
251 "psubusw %%mm4, %%mm3 \n\t"\ | |
252 "pxor %%mm6, %%mm2 \n\t"\ | |
253 "pxor %%mm7, %%mm3 \n\t"\ | |
254 \ | |
255 "paddsw %%mm5, %%mm0 \n\t"\ | |
256 "paddsw %%mm5, %%mm1 \n\t"\ | |
257 "paddsw %%mm5, %%mm2 \n\t"\ | |
258 "paddsw %%mm5, %%mm3 \n\t"\ | |
259 "psraw $3, %%mm0 \n\t"\ | |
260 "psraw $3, %%mm1 \n\t"\ | |
261 "psraw $3, %%mm2 \n\t"\ | |
262 "psraw $3, %%mm3 \n\t"\ | |
263 \ | |
264 "movq %%mm0, %%mm7 \n\t"\ | |
265 "punpcklwd %%mm2, %%mm0 \n\t" /*A*/\ | |
266 "punpckhwd %%mm2, %%mm7 \n\t" /*C*/\ | |
267 "movq %%mm1, %%mm2 \n\t"\ | |
268 "punpcklwd %%mm3, %%mm1 \n\t" /*B*/\ | |
269 "punpckhwd %%mm3, %%mm2 \n\t" /*D*/\ | |
270 "movq %%mm0, %%mm3 \n\t"\ | |
271 "punpcklwd %%mm1, %%mm0 \n\t" /*A*/\ | |
272 "punpckhwd %%mm7, %%mm3 \n\t" /*C*/\ | |
13393 | 273 "punpcklwd %%mm2, %%mm7 \n\t" /*B*/\ |
11994 | 274 "punpckhwd %%mm2, %%mm1 \n\t" /*D*/\ |
275 \ | |
276 "movq %%mm0, " #dst0 " \n\t"\ | |
277 "movq %%mm7, " #dst1 " \n\t"\ | |
278 "movq %%mm3, " #dst2 " \n\t"\ | |
279 "movq %%mm1, " #dst3 " \n\t" | |
280 | |
281 "movd %2, %%mm4 \n\t" | |
282 "movd %3, %%mm5 \n\t" | |
283 "packssdw %%mm4, %%mm4 \n\t" | |
284 "packssdw %%mm5, %%mm5 \n\t" | |
285 "packssdw %%mm4, %%mm4 \n\t" | |
286 "packssdw %%mm5, %%mm5 \n\t" | |
287 REQUANT_CORE( (%1), 8(%1), 16(%1), 24(%1), (%0), 8(%0), 64(%0), 72(%0)) | |
288 REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0)) | |
289 REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0)) | |
290 REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0)) | |
291 : : "r" (src), "r" (dst), "g" (threshold1), "rm" (4) //FIXME maybe more accurate then needed? | |
292 ); | |
293 | |
294 dst[0]= (src[0] + 4)>>3; | |
295 } | |
11301 | 296 #endif |
297 | |
11277 | 298 static inline void add_block(int16_t *dst, int stride, DCTELEM block[64]){ |
11280 | 299 int y; |
11277 | 300 |
301 for(y=0; y<8; y++){ | |
11280 | 302 *(uint32_t*)&dst[0 + y*stride]+= *(uint32_t*)&block[0 + y*8]; |
303 *(uint32_t*)&dst[2 + y*stride]+= *(uint32_t*)&block[2 + y*8]; | |
304 *(uint32_t*)&dst[4 + y*stride]+= *(uint32_t*)&block[4 + y*8]; | |
305 *(uint32_t*)&dst[6 + y*stride]+= *(uint32_t*)&block[6 + y*8]; | |
11277 | 306 } |
307 } | |
308 | |
11305 | 309 static void store_slice_c(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){ |
11299 | 310 int y, x; |
311 | |
312 #define STORE(pos) \ | |
313 temp= ((src[x + y*src_stride + pos]<<log2_scale) + d[pos])>>6;\ | |
314 if(temp & 0x100) temp= ~(temp>>31);\ | |
315 dst[x + y*dst_stride + pos]= temp; | |
316 | |
11305 | 317 for(y=0; y<height; y++){ |
12157 | 318 const uint8_t *d= dither[y]; |
11299 | 319 for(x=0; x<width; x+=8){ |
320 int temp; | |
321 STORE(0); | |
322 STORE(1); | |
323 STORE(2); | |
324 STORE(3); | |
325 STORE(4); | |
326 STORE(5); | |
327 STORE(6); | |
328 STORE(7); | |
329 } | |
330 } | |
331 } | |
332 | |
333 #ifdef HAVE_MMX | |
11305 | 334 static void store_slice_mmx(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale){ |
11299 | 335 int y; |
336 | |
11305 | 337 for(y=0; y<height; y++){ |
11299 | 338 uint8_t *dst1= dst; |
339 int16_t *src1= src; | |
340 asm volatile( | |
341 "movq (%3), %%mm3 \n\t" | |
342 "movq (%3), %%mm4 \n\t" | |
343 "movd %4, %%mm2 \n\t" | |
344 "pxor %%mm0, %%mm0 \n\t" | |
345 "punpcklbw %%mm0, %%mm3 \n\t" | |
346 "punpckhbw %%mm0, %%mm4 \n\t" | |
347 "psraw %%mm2, %%mm3 \n\t" | |
348 "psraw %%mm2, %%mm4 \n\t" | |
349 "movd %5, %%mm2 \n\t" | |
350 "1: \n\t" | |
351 "movq (%0), %%mm0 \n\t" | |
352 "movq 8(%0), %%mm1 \n\t" | |
353 "paddw %%mm3, %%mm0 \n\t" | |
354 "paddw %%mm4, %%mm1 \n\t" | |
355 "psraw %%mm2, %%mm0 \n\t" | |
356 "psraw %%mm2, %%mm1 \n\t" | |
357 "packuswb %%mm1, %%mm0 \n\t" | |
358 "movq %%mm0, (%1) \n\t" | |
13720
821f464b4d90
adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64
aurel
parents:
13393
diff
changeset
|
359 "add $16, %0 \n\t" |
821f464b4d90
adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64
aurel
parents:
13393
diff
changeset
|
360 "add $8, %1 \n\t" |
821f464b4d90
adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64
aurel
parents:
13393
diff
changeset
|
361 "cmp %2, %1 \n\t" |
11299 | 362 " jb 1b \n\t" |
363 : "+r" (src1), "+r"(dst1) | |
364 : "r"(dst + width), "r"(dither[y]), "g"(log2_scale), "g"(6-log2_scale) | |
365 ); | |
366 src += src_stride; | |
367 dst += dst_stride; | |
368 } | |
369 // if(width != mmxw) | |
370 // store_slice_c(dst + mmxw, src + mmxw, dst_stride, src_stride, width - mmxw, log2_scale); | |
371 } | |
372 #endif | |
373 | |
11305 | 374 static void (*store_slice)(uint8_t *dst, int16_t *src, int dst_stride, int src_stride, int width, int height, int log2_scale)= store_slice_c; |
11299 | 375 |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
376 static void (*requantize)(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation)= hardthresh_c; |
11301 | 377 |
11277 | 378 static void filter(struct vf_priv_s *p, uint8_t *dst, uint8_t *src, int dst_stride, int src_stride, int width, int height, uint8_t *qp_store, int qp_stride, int is_luma){ |
379 int x, y, i; | |
380 const int count= 1<<p->log2_count; | |
11307 | 381 const int stride= is_luma ? p->temp_stride : ((width+16+15)&(~15)); |
11277 | 382 uint64_t block_align[32]; |
383 DCTELEM *block = (DCTELEM *)block_align; | |
384 DCTELEM *block2= (DCTELEM *)(block_align+16); | |
11301 | 385 |
13729 | 386 if (!src || !dst) return; // HACK avoid crash for Y8 colourspace |
11277 | 387 for(y=0; y<height; y++){ |
11299 | 388 int index= 8 + 8*stride + y*stride; |
389 memcpy(p->src + index, src + y*src_stride, width); | |
11277 | 390 for(x=0; x<8; x++){ |
391 p->src[index - x - 1]= p->src[index + x ]; | |
392 p->src[index + width + x ]= p->src[index + width - x - 1]; | |
393 } | |
394 } | |
395 for(y=0; y<8; y++){ | |
396 memcpy(p->src + ( 7-y)*stride, p->src + ( y+8)*stride, stride); | |
11281 | 397 memcpy(p->src + (height+8+y)*stride, p->src + (height-y+7)*stride, stride); |
11277 | 398 } |
399 //FIXME (try edge emu) | |
400 | |
401 for(y=0; y<height+8; y+=8){ | |
11299 | 402 memset(p->temp + (8+y)*stride, 0, 8*stride*sizeof(int16_t)); |
11277 | 403 for(x=0; x<width+8; x+=8){ |
404 const int qps= 3 + is_luma; | |
405 int qp; | |
406 | |
407 if(p->qp) | |
408 qp= p->qp; | |
409 else{ | |
410 qp= qp_store[ (XMIN(x, width-1)>>qps) + (XMIN(y, height-1)>>qps) * qp_stride]; | |
411 if(p->mpeg2) qp>>=1; | |
412 } | |
413 for(i=0; i<count; i++){ | |
11298 | 414 const int x1= x + offset[i+count-1][0]; |
415 const int y1= y + offset[i+count-1][1]; | |
11277 | 416 const int index= x1 + y1*stride; |
417 p->dsp.get_pixels(block, p->src + index, stride); | |
418 p->dsp.fdct(block); | |
419 requantize(block2, block, qp, p->dsp.idct_permutation); | |
420 p->dsp.idct(block2); | |
421 add_block(p->temp + index, stride, block2); | |
422 } | |
423 } | |
11299 | 424 if(y) |
11305 | 425 store_slice(dst + (y-8)*dst_stride, p->temp + 8 + y*stride, dst_stride, stride, width, XMIN(8, height+8-y), 6-p->log2_count); |
11277 | 426 } |
11295 | 427 #if 0 |
428 for(y=0; y<height; y++){ | |
429 for(x=0; x<width; x++){ | |
430 if((((x>>6) ^ (y>>6)) & 1) == 0) | |
431 dst[x + y*dst_stride]= p->src[8 + 8*stride + x + y*stride]; | |
432 if((x&63) == 0 || (y&63)==0) | |
433 dst[x + y*dst_stride] += 128; | |
434 } | |
435 } | |
436 #endif | |
11277 | 437 //FIXME reorder for better caching |
438 } | |
439 | |
440 static int config(struct vf_instance_s* vf, | |
441 int width, int height, int d_width, int d_height, | |
442 unsigned int flags, unsigned int outfmt){ | |
11305 | 443 int h= (height+16+15)&(~15); |
11277 | 444 |
445 vf->priv->temp_stride= (width+16+15)&(~15); | |
11305 | 446 vf->priv->temp= malloc(vf->priv->temp_stride*h*sizeof(int16_t)); |
447 vf->priv->src = malloc(vf->priv->temp_stride*h*sizeof(uint8_t)); | |
11277 | 448 |
449 return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt); | |
450 } | |
451 | |
452 static void get_image(struct vf_instance_s* vf, mp_image_t *mpi){ | |
453 if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change | |
454 // ok, we can do pp in-place (or pp disabled): | |
455 vf->dmpi=vf_get_image(vf->next,mpi->imgfmt, | |
456 mpi->type, mpi->flags, mpi->w, mpi->h); | |
457 mpi->planes[0]=vf->dmpi->planes[0]; | |
458 mpi->stride[0]=vf->dmpi->stride[0]; | |
459 mpi->width=vf->dmpi->width; | |
460 if(mpi->flags&MP_IMGFLAG_PLANAR){ | |
461 mpi->planes[1]=vf->dmpi->planes[1]; | |
462 mpi->planes[2]=vf->dmpi->planes[2]; | |
463 mpi->stride[1]=vf->dmpi->stride[1]; | |
464 mpi->stride[2]=vf->dmpi->stride[2]; | |
465 } | |
466 mpi->flags|=MP_IMGFLAG_DIRECT; | |
467 } | |
468 | |
469 static int put_image(struct vf_instance_s* vf, mp_image_t *mpi){ | |
470 mp_image_t *dmpi; | |
471 | |
472 if(!(mpi->flags&MP_IMGFLAG_DIRECT)){ | |
473 // no DR, so get a new image! hope we'll get DR buffer: | |
14022
803be9a78854
removing strange csp matching code (was copy&pasted from vf_pp where it originated from arpi 2.5 years ago) -> fixes spp+scale+x11 crash
michael
parents:
13729
diff
changeset
|
474 dmpi=vf_get_image(vf->next,mpi->imgfmt, |
12631
6cbccc0c7d7b
Fix memory corruption, noticable at reallocate image
iive
parents:
12163
diff
changeset
|
475 MP_IMGTYPE_TEMP, |
6cbccc0c7d7b
Fix memory corruption, noticable at reallocate image
iive
parents:
12163
diff
changeset
|
476 MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE, |
6cbccc0c7d7b
Fix memory corruption, noticable at reallocate image
iive
parents:
12163
diff
changeset
|
477 mpi->w,mpi->h); |
6cbccc0c7d7b
Fix memory corruption, noticable at reallocate image
iive
parents:
12163
diff
changeset
|
478 vf_clone_mpi_attributes(dmpi, mpi); |
6cbccc0c7d7b
Fix memory corruption, noticable at reallocate image
iive
parents:
12163
diff
changeset
|
479 }else{ |
13149 | 480 dmpi=vf->dmpi; |
12631
6cbccc0c7d7b
Fix memory corruption, noticable at reallocate image
iive
parents:
12163
diff
changeset
|
481 } |
11277 | 482 |
483 vf->priv->mpeg2= mpi->qscale_type; | |
11307 | 484 if(vf->priv->log2_count || !(mpi->flags&MP_IMGFLAG_DIRECT)){ |
11308
0c8d12a58a29
skip filter if codec doesnt provide the QP array and user didnt force a QP (fixes diegos segfault)
michael
parents:
11307
diff
changeset
|
485 if(mpi->qscale || vf->priv->qp){ |
11307 | 486 filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, mpi->qscale, mpi->qstride, 1); |
487 filter(vf->priv, dmpi->planes[1], mpi->planes[1], dmpi->stride[1], mpi->stride[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, mpi->qscale, mpi->qstride, 0); | |
488 filter(vf->priv, dmpi->planes[2], mpi->planes[2], dmpi->stride[2], mpi->stride[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, mpi->qscale, mpi->qstride, 0); | |
11477
9785bff83777
memcpy pix instead of black screen if no DR and codec provides no qscale table
michael
parents:
11335
diff
changeset
|
489 }else{ |
9785bff83777
memcpy pix instead of black screen if no DR and codec provides no qscale table
michael
parents:
11335
diff
changeset
|
490 memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h, dmpi->stride[0], mpi->stride[0]); |
9785bff83777
memcpy pix instead of black screen if no DR and codec provides no qscale table
michael
parents:
11335
diff
changeset
|
491 memcpy_pic(dmpi->planes[1], mpi->planes[1], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, dmpi->stride[1], mpi->stride[1]); |
9785bff83777
memcpy pix instead of black screen if no DR and codec provides no qscale table
michael
parents:
11335
diff
changeset
|
492 memcpy_pic(dmpi->planes[2], mpi->planes[2], mpi->w>>mpi->chroma_x_shift, mpi->h>>mpi->chroma_y_shift, dmpi->stride[2], mpi->stride[2]); |
11308
0c8d12a58a29
skip filter if codec doesnt provide the QP array and user didnt force a QP (fixes diegos segfault)
michael
parents:
11307
diff
changeset
|
493 } |
11307 | 494 } |
11277 | 495 |
496 #ifdef HAVE_MMX | |
497 if(gCpuCaps.hasMMX) asm volatile ("emms\n\t"); | |
498 #endif | |
499 #ifdef HAVE_MMX2 | |
500 if(gCpuCaps.hasMMX2) asm volatile ("sfence\n\t"); | |
501 #endif | |
502 | |
503 return vf_next_put_image(vf,dmpi); | |
504 } | |
505 | |
506 static void uninit(struct vf_instance_s* vf){ | |
507 if(!vf->priv) return; | |
508 | |
509 if(vf->priv->temp) free(vf->priv->temp); | |
510 vf->priv->temp= NULL; | |
511 if(vf->priv->src) free(vf->priv->src); | |
512 vf->priv->src= NULL; | |
513 if(vf->priv->avctx) free(vf->priv->avctx); | |
514 vf->priv->avctx= NULL; | |
515 | |
516 free(vf->priv); | |
517 vf->priv=NULL; | |
518 } | |
519 | |
520 //===========================================================================// | |
521 static int query_format(struct vf_instance_s* vf, unsigned int fmt){ | |
11307 | 522 switch(fmt){ |
523 case IMGFMT_YVU9: | |
524 case IMGFMT_IF09: | |
11277 | 525 case IMGFMT_YV12: |
526 case IMGFMT_I420: | |
527 case IMGFMT_IYUV: | |
11307 | 528 case IMGFMT_CLPL: |
529 case IMGFMT_Y800: | |
530 case IMGFMT_Y8: | |
531 case IMGFMT_NV12: | |
532 case IMGFMT_444P: | |
533 case IMGFMT_422P: | |
534 case IMGFMT_411P: | |
535 return vf_next_query_format(vf,fmt); | |
536 } | |
537 return 0; | |
11277 | 538 } |
539 | |
540 static unsigned int fmt_list[]={ | |
11307 | 541 IMGFMT_YVU9, |
542 IMGFMT_IF09, | |
543 IMGFMT_YV12, | |
544 IMGFMT_I420, | |
545 IMGFMT_IYUV, | |
546 IMGFMT_CLPL, | |
547 IMGFMT_Y800, | |
548 IMGFMT_Y8, | |
549 IMGFMT_NV12, | |
550 IMGFMT_444P, | |
551 IMGFMT_422P, | |
552 IMGFMT_411P, | |
553 0 | |
11277 | 554 }; |
555 | |
11307 | 556 static int control(struct vf_instance_s* vf, int request, void* data){ |
557 switch(request){ | |
558 case VFCTRL_QUERY_MAX_PP_LEVEL: | |
559 return 6; | |
560 case VFCTRL_SET_PP_LEVEL: | |
561 vf->priv->log2_count= *((unsigned int*)data); | |
562 return CONTROL_TRUE; | |
563 } | |
564 return vf_next_control(vf,request,data); | |
565 } | |
566 | |
11277 | 567 static int open(vf_instance_t *vf, char* args){ |
12157 | 568 |
14022
803be9a78854
removing strange csp matching code (was copy&pasted from vf_pp where it originated from arpi 2.5 years ago) -> fixes spp+scale+x11 crash
michael
parents:
13729
diff
changeset
|
569 int log2c=-1; |
12157 | 570 |
11277 | 571 vf->config=config; |
572 vf->put_image=put_image; | |
573 vf->get_image=get_image; | |
574 vf->query_format=query_format; | |
575 vf->uninit=uninit; | |
11307 | 576 vf->control= control; |
11277 | 577 vf->priv=malloc(sizeof(struct vf_priv_s)); |
578 memset(vf->priv, 0, sizeof(struct vf_priv_s)); | |
11280 | 579 |
580 avcodec_init(); | |
581 | |
11277 | 582 vf->priv->avctx= avcodec_alloc_context(); |
583 dsputil_init(&vf->priv->dsp, vf->priv->avctx); | |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
584 |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
585 vf->priv->log2_count= 3; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
586 |
12157 | 587 if (args) sscanf(args, "%d:%d:%d", &log2c, &vf->priv->qp, &vf->priv->mode); |
588 | |
589 if( log2c >=0 && log2c <=6 ) | |
590 vf->priv->log2_count = log2c; | |
591 | |
592 if(vf->priv->qp < 0) | |
593 vf->priv->qp = 0; | |
594 | |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
595 switch(vf->priv->mode){ |
12157 | 596 default: |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
597 case 0: requantize= hardthresh_c; break; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
598 case 1: requantize= softthresh_c; break; |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
599 } |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
600 |
11304 | 601 #ifdef HAVE_MMX |
11299 | 602 if(gCpuCaps.hasMMX){ |
603 store_slice= store_slice_mmx; | |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
604 switch(vf->priv->mode){ |
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
605 case 0: requantize= hardthresh_mmx; break; |
11994 | 606 case 1: requantize= softthresh_mmx; break; |
11993
4e2d99dbef78
spp soft thresholding patch by (James Crowson <jbcrowso at ncsu dot edu>)
michael
parents:
11568
diff
changeset
|
607 } |
11299 | 608 } |
11304 | 609 #endif |
11299 | 610 |
11277 | 611 return 1; |
612 } | |
613 | |
614 vf_info_t vf_info_spp = { | |
615 "simple postprocess", | |
616 "spp", | |
617 "Michael Niedermayer", | |
618 "", | |
619 open, | |
620 NULL | |
621 }; | |
622 | |
11335 | 623 #endif //USE_LIBAVCODEC |