ffmpeg / libavcodec / dsputil.h @ 015f9f1a
History | View | Annotate | Download (31.2 KB)
1 |
/*
|
---|---|
2 |
* DSP utils
|
3 |
* Copyright (c) 2000, 2001, 2002 Fabrice Bellard
|
4 |
* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
|
5 |
*
|
6 |
* This file is part of FFmpeg.
|
7 |
*
|
8 |
* FFmpeg is free software; you can redistribute it and/or
|
9 |
* modify it under the terms of the GNU Lesser General Public
|
10 |
* License as published by the Free Software Foundation; either
|
11 |
* version 2.1 of the License, or (at your option) any later version.
|
12 |
*
|
13 |
* FFmpeg is distributed in the hope that it will be useful,
|
14 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
15 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
16 |
* Lesser General Public License for more details.
|
17 |
*
|
18 |
* You should have received a copy of the GNU Lesser General Public
|
19 |
* License along with FFmpeg; if not, write to the Free Software
|
20 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
21 |
*/
|
22 |
|
23 |
/**
|
24 |
* @file
|
25 |
* DSP utils.
|
26 |
* note, many functions in here may use MMX which trashes the FPU state, it is
|
27 |
* absolutely necessary to call emms_c() between dsp & float/double code
|
28 |
*/
|
29 |
|
30 |
#ifndef AVCODEC_DSPUTIL_H
|
31 |
#define AVCODEC_DSPUTIL_H
|
32 |
|
33 |
#include "libavutil/intreadwrite.h" |
34 |
#include "avcodec.h" |
35 |
|
36 |
|
37 |
//#define DEBUG
|
38 |
/* dct code */
|
39 |
typedef short DCTELEM; |
40 |
|
41 |
void fdct_ifast (DCTELEM *data);
|
42 |
void fdct_ifast248 (DCTELEM *data);
|
43 |
void ff_jpeg_fdct_islow (DCTELEM *data);
|
44 |
void ff_fdct248_islow (DCTELEM *data);
|
45 |
|
46 |
void j_rev_dct (DCTELEM *data);
|
47 |
void j_rev_dct4 (DCTELEM *data);
|
48 |
void j_rev_dct2 (DCTELEM *data);
|
49 |
void j_rev_dct1 (DCTELEM *data);
|
50 |
void ff_wmv2_idct_c(DCTELEM *data);
|
51 |
|
52 |
void ff_fdct_mmx(DCTELEM *block);
|
53 |
void ff_fdct_mmx2(DCTELEM *block);
|
54 |
void ff_fdct_sse2(DCTELEM *block);
|
55 |
|
56 |
void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride); |
57 |
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride); |
58 |
void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); |
59 |
void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); |
60 |
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block); |
61 |
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block); |
62 |
void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); |
63 |
void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); |
64 |
void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); |
65 |
void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); |
66 |
|
67 |
void ff_h264_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qmul); |
68 |
void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp); |
69 |
void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc); |
70 |
|
71 |
void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1, |
72 |
const float *win, float add_bias, int len); |
73 |
void ff_float_to_int16_c(int16_t *dst, const float *src, long len); |
74 |
void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels); |
75 |
|
76 |
/* encoding scans */
|
77 |
extern const uint8_t ff_alternate_horizontal_scan[64]; |
78 |
extern const uint8_t ff_alternate_vertical_scan[64]; |
79 |
extern const uint8_t ff_zigzag_direct[64]; |
80 |
extern const uint8_t ff_zigzag248_direct[64]; |
81 |
|
82 |
/* pixel operations */
|
83 |
#define MAX_NEG_CROP 1024 |
84 |
|
85 |
/* temporary */
|
86 |
extern uint32_t ff_squareTbl[512]; |
87 |
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP]; |
88 |
|
89 |
void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride); |
90 |
void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride); |
91 |
void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride); |
92 |
void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride); |
93 |
|
94 |
/* VP3 DSP functions */
|
95 |
void ff_vp3_idct_c(DCTELEM *block/* align 16*/); |
96 |
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
97 |
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
98 |
void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/); |
99 |
|
100 |
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values); |
101 |
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values); |
102 |
|
103 |
/* Bink functions */
|
104 |
void ff_bink_idct_c (DCTELEM *block);
|
105 |
void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block); |
106 |
void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block); |
107 |
|
108 |
/* EA functions */
|
109 |
void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block); |
110 |
|
111 |
/* 1/2^n downscaling functions from imgconvert.c */
|
112 |
#if LIBAVCODEC_VERSION_MAJOR < 53 |
113 |
/**
|
114 |
* @deprecated Use av_image_copy_plane() instead.
|
115 |
*/
|
116 |
attribute_deprecated |
117 |
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); |
118 |
#endif
|
119 |
|
120 |
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); |
121 |
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); |
122 |
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); |
123 |
|
124 |
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, |
125 |
int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
126 |
|
127 |
/* minimum alignment rules ;)
|
128 |
If you notice errors in the align stuff, need more alignment for some ASM code
|
129 |
for some CPU or need to use a function with less aligned data then send a mail
|
130 |
to the ffmpeg-devel mailing list, ...
|
131 |
|
132 |
!warning These alignments might not match reality, (missing attribute((align))
|
133 |
stuff somewhere possible).
|
134 |
I (Michael) did not check them, these are just the alignments which I think
|
135 |
could be reached easily ...
|
136 |
|
137 |
!future video codecs might need functions with less strict alignment
|
138 |
*/
|
139 |
|
140 |
/*
|
141 |
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
|
142 |
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
|
143 |
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
|
144 |
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
|
145 |
void clear_blocks_c(DCTELEM *blocks);
|
146 |
*/
|
147 |
|
148 |
/* add and put pixel (decoding) */
|
149 |
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
|
150 |
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
|
151 |
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h); |
152 |
typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h); |
153 |
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
154 |
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); |
155 |
|
156 |
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h); |
157 |
|
158 |
#define DEF_OLD_QPEL(name)\
|
159 |
void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
160 |
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
161 |
void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); |
162 |
|
163 |
DEF_OLD_QPEL(qpel16_mc11_old_c) |
164 |
DEF_OLD_QPEL(qpel16_mc31_old_c) |
165 |
DEF_OLD_QPEL(qpel16_mc12_old_c) |
166 |
DEF_OLD_QPEL(qpel16_mc32_old_c) |
167 |
DEF_OLD_QPEL(qpel16_mc13_old_c) |
168 |
DEF_OLD_QPEL(qpel16_mc33_old_c) |
169 |
DEF_OLD_QPEL(qpel8_mc11_old_c) |
170 |
DEF_OLD_QPEL(qpel8_mc31_old_c) |
171 |
DEF_OLD_QPEL(qpel8_mc12_old_c) |
172 |
DEF_OLD_QPEL(qpel8_mc32_old_c) |
173 |
DEF_OLD_QPEL(qpel8_mc13_old_c) |
174 |
DEF_OLD_QPEL(qpel8_mc33_old_c) |
175 |
|
176 |
#define CALL_2X_PIXELS(a, b, n)\
|
177 |
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ |
178 |
b(block , pixels , line_size, h);\ |
179 |
b(block+n, pixels+n, line_size, h);\ |
180 |
} |
181 |
|
182 |
/* motion estimation */
|
183 |
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
|
184 |
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
|
185 |
typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/; |
186 |
|
187 |
/**
|
188 |
* Scantable.
|
189 |
*/
|
190 |
typedef struct ScanTable{ |
191 |
const uint8_t *scantable;
|
192 |
uint8_t permutated[64];
|
193 |
uint8_t raster_end[64];
|
194 |
#if ARCH_PPC
|
195 |
/** Used by dct_quantize_altivec to find last-non-zero */
|
196 |
DECLARE_ALIGNED(16, uint8_t, inverse)[64]; |
197 |
#endif
|
198 |
} ScanTable; |
199 |
|
200 |
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); |
201 |
|
202 |
void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize, |
203 |
int block_w, int block_h, |
204 |
int src_x, int src_y, int w, int h); |
205 |
|
206 |
/**
|
207 |
* DSPContext.
|
208 |
*/
|
209 |
typedef struct DSPContext { |
210 |
/* pixel ops : interface with DCT */
|
211 |
void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
212 |
void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); |
213 |
void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
214 |
void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
215 |
void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
216 |
void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); |
217 |
void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); |
218 |
void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); |
219 |
int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/); |
220 |
/**
|
221 |
* translational global motion compensation.
|
222 |
*/
|
223 |
void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
224 |
/**
|
225 |
* global motion compensation.
|
226 |
*/
|
227 |
void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
228 |
int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
229 |
void (*clear_block)(DCTELEM *block/*align 16*/); |
230 |
void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
231 |
int (*pix_sum)(uint8_t * pix, int line_size); |
232 |
int (*pix_norm1)(uint8_t * pix, int line_size); |
233 |
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
|
234 |
|
235 |
me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */ |
236 |
me_cmp_func sse[6];
|
237 |
me_cmp_func hadamard8_diff[6];
|
238 |
me_cmp_func dct_sad[6];
|
239 |
me_cmp_func quant_psnr[6];
|
240 |
me_cmp_func bit[6];
|
241 |
me_cmp_func rd[6];
|
242 |
me_cmp_func vsad[6];
|
243 |
me_cmp_func vsse[6];
|
244 |
me_cmp_func nsse[6];
|
245 |
me_cmp_func w53[6];
|
246 |
me_cmp_func w97[6];
|
247 |
me_cmp_func dct_max[6];
|
248 |
me_cmp_func dct264_sad[6];
|
249 |
|
250 |
me_cmp_func me_pre_cmp[6];
|
251 |
me_cmp_func me_cmp[6];
|
252 |
me_cmp_func me_sub_cmp[6];
|
253 |
me_cmp_func mb_cmp[6];
|
254 |
me_cmp_func ildct_cmp[6]; //only width 16 used |
255 |
me_cmp_func frame_skip_cmp[6]; //only width 8 used |
256 |
|
257 |
int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, |
258 |
int size);
|
259 |
|
260 |
/**
|
261 |
* Halfpel motion compensation with rounding (a+b+1)>>1.
|
262 |
* this is an array[4][4] of motion compensation functions for 4
|
263 |
* horizontal blocksizes (8,16) and the 4 halfpel positions<br>
|
264 |
* *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
|
265 |
* @param block destination where the result is stored
|
266 |
* @param pixels source
|
267 |
* @param line_size number of bytes in a horizontal line of block
|
268 |
* @param h height
|
269 |
*/
|
270 |
op_pixels_func put_pixels_tab[4][4]; |
271 |
|
272 |
/**
|
273 |
* Halfpel motion compensation with rounding (a+b+1)>>1.
|
274 |
* This is an array[4][4] of motion compensation functions for 4
|
275 |
* horizontal blocksizes (8,16) and the 4 halfpel positions<br>
|
276 |
* *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
|
277 |
* @param block destination into which the result is averaged (a+b+1)>>1
|
278 |
* @param pixels source
|
279 |
* @param line_size number of bytes in a horizontal line of block
|
280 |
* @param h height
|
281 |
*/
|
282 |
op_pixels_func avg_pixels_tab[4][4]; |
283 |
|
284 |
/**
|
285 |
* Halfpel motion compensation with no rounding (a+b)>>1.
|
286 |
* this is an array[2][4] of motion compensation functions for 2
|
287 |
* horizontal blocksizes (8,16) and the 4 halfpel positions<br>
|
288 |
* *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
|
289 |
* @param block destination where the result is stored
|
290 |
* @param pixels source
|
291 |
* @param line_size number of bytes in a horizontal line of block
|
292 |
* @param h height
|
293 |
*/
|
294 |
op_pixels_func put_no_rnd_pixels_tab[4][4]; |
295 |
|
296 |
/**
|
297 |
* Halfpel motion compensation with no rounding (a+b)>>1.
|
298 |
* this is an array[2][4] of motion compensation functions for 2
|
299 |
* horizontal blocksizes (8,16) and the 4 halfpel positions<br>
|
300 |
* *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
|
301 |
* @param block destination into which the result is averaged (a+b)>>1
|
302 |
* @param pixels source
|
303 |
* @param line_size number of bytes in a horizontal line of block
|
304 |
* @param h height
|
305 |
*/
|
306 |
op_pixels_func avg_no_rnd_pixels_tab[4][4]; |
307 |
|
308 |
void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h); |
309 |
|
310 |
/**
|
311 |
* Thirdpel motion compensation with rounding (a+b+1)>>1.
|
312 |
* this is an array[12] of motion compensation functions for the 9 thirdpe
|
313 |
* positions<br>
|
314 |
* *pixels_tab[ xthirdpel + 4*ythirdpel ]
|
315 |
* @param block destination where the result is stored
|
316 |
* @param pixels source
|
317 |
* @param line_size number of bytes in a horizontal line of block
|
318 |
* @param h height
|
319 |
*/
|
320 |
tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
321 |
tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? |
322 |
|
323 |
qpel_mc_func put_qpel_pixels_tab[2][16]; |
324 |
qpel_mc_func avg_qpel_pixels_tab[2][16]; |
325 |
qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; |
326 |
qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
327 |
qpel_mc_func put_mspel_pixels_tab[8];
|
328 |
|
329 |
/**
|
330 |
* h264 Chroma MC
|
331 |
*/
|
332 |
h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
|
333 |
h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
|
334 |
/* This is really one func used in VC-1 decoding */
|
335 |
h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
|
336 |
h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
|
337 |
|
338 |
qpel_mc_func put_h264_qpel_pixels_tab[4][16]; |
339 |
qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; |
340 |
|
341 |
qpel_mc_func put_2tap_qpel_pixels_tab[4][16]; |
342 |
qpel_mc_func avg_2tap_qpel_pixels_tab[4][16]; |
343 |
|
344 |
me_cmp_func pix_abs[2][4]; |
345 |
|
346 |
/* huffyuv specific */
|
347 |
void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); |
348 |
void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w); |
349 |
void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
350 |
/**
|
351 |
* subtract huffyuv's variant of median prediction
|
352 |
* note, this might read from src1[-1], src2[-1]
|
353 |
*/
|
354 |
void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top); |
355 |
void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top); |
356 |
int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left); |
357 |
void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha); |
358 |
/* this might write to dst[w] */
|
359 |
void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); |
360 |
void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); |
361 |
|
362 |
void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); |
363 |
void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); |
364 |
|
365 |
void (*h261_loop_filter)(uint8_t *src, int stride); |
366 |
|
367 |
void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale); |
368 |
void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale); |
369 |
|
370 |
void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/); |
371 |
void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values); |
372 |
void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values); |
373 |
|
374 |
/* assume len is a multiple of 4, and arrays are 16-byte aligned */
|
375 |
void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); |
376 |
void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len); |
377 |
/* assume len is a multiple of 8, and arrays are 16-byte aligned */
|
378 |
void (*vector_fmul)(float *dst, const float *src0, const float *src1, int len); |
379 |
void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len); |
380 |
/* assume len is a multiple of 8, and src arrays are 16-byte aligned */
|
381 |
void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len); |
382 |
/* assume len is a multiple of 4, and arrays are 16-byte aligned */
|
383 |
void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len); |
384 |
/* assume len is a multiple of 8, and arrays are 16-byte aligned */
|
385 |
void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len); |
386 |
void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */); |
387 |
/**
|
388 |
* Multiply a vector of floats by a scalar float. Source and
|
389 |
* destination vectors must overlap exactly or not at all.
|
390 |
* @param dst result vector, 16-byte aligned
|
391 |
* @param src input vector, 16-byte aligned
|
392 |
* @param mul scalar value
|
393 |
* @param len length of vector, multiple of 4
|
394 |
*/
|
395 |
void (*vector_fmul_scalar)(float *dst, const float *src, float mul, |
396 |
int len);
|
397 |
/**
|
398 |
* Multiply a vector of floats by concatenated short vectors of
|
399 |
* floats and by a scalar float. Source and destination vectors
|
400 |
* must overlap exactly or not at all.
|
401 |
* [0]: short vectors of length 2, 8-byte aligned
|
402 |
* [1]: short vectors of length 4, 16-byte aligned
|
403 |
* @param dst output vector, 16-byte aligned
|
404 |
* @param src input vector, 16-byte aligned
|
405 |
* @param sv array of pointers to short vectors
|
406 |
* @param mul scalar value
|
407 |
* @param len number of elements in src and dst, multiple of 4
|
408 |
*/
|
409 |
void (*vector_fmul_sv_scalar[2])(float *dst, const float *src, |
410 |
const float **sv, float mul, int len); |
411 |
/**
|
412 |
* Multiply short vectors of floats by a scalar float, store
|
413 |
* concatenated result.
|
414 |
* [0]: short vectors of length 2, 8-byte aligned
|
415 |
* [1]: short vectors of length 4, 16-byte aligned
|
416 |
* @param dst output vector, 16-byte aligned
|
417 |
* @param sv array of pointers to short vectors
|
418 |
* @param mul scalar value
|
419 |
* @param len number of output elements, multiple of 4
|
420 |
*/
|
421 |
void (*sv_fmul_scalar[2])(float *dst, const float **sv, |
422 |
float mul, int len); |
423 |
/**
|
424 |
* Calculate the scalar product of two vectors of floats.
|
425 |
* @param v1 first vector, 16-byte aligned
|
426 |
* @param v2 second vector, 16-byte aligned
|
427 |
* @param len length of vectors, multiple of 4
|
428 |
*/
|
429 |
float (*scalarproduct_float)(const float *v1, const float *v2, int len); |
430 |
/**
|
431 |
* Calculate the sum and difference of two vectors of floats.
|
432 |
* @param v1 first input vector, sum output, 16-byte aligned
|
433 |
* @param v2 second input vector, difference output, 16-byte aligned
|
434 |
* @param len length of vectors, multiple of 4
|
435 |
*/
|
436 |
void (*butterflies_float)(float *restrict v1, float *restrict v2, int len); |
437 |
|
438 |
/* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
|
439 |
* simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
|
440 |
void (*float_to_int16)(int16_t *dst, const float *src, long len); |
441 |
void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels); |
442 |
|
443 |
/* (I)DCT */
|
444 |
void (*fdct)(DCTELEM *block/* align 16*/); |
445 |
void (*fdct248)(DCTELEM *block/* align 16*/); |
446 |
|
447 |
/* IDCT really*/
|
448 |
void (*idct)(DCTELEM *block/* align 16*/); |
449 |
|
450 |
/**
|
451 |
* block -> idct -> clip to unsigned 8 bit -> dest.
|
452 |
* (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
|
453 |
* @param line_size size in bytes of a horizontal line of dest
|
454 |
*/
|
455 |
void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
456 |
|
457 |
/**
|
458 |
* block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
|
459 |
* @param line_size size in bytes of a horizontal line of dest
|
460 |
*/
|
461 |
void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
462 |
|
463 |
/**
|
464 |
* idct input permutation.
|
465 |
* several optimized IDCTs need a permutated input (relative to the normal order of the reference
|
466 |
* IDCT)
|
467 |
* this permutation must be performed before the idct_put/add, note, normally this can be merged
|
468 |
* with the zigzag/alternate scan<br>
|
469 |
* an example to avoid confusion:
|
470 |
* - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
|
471 |
* - (x -> referece dct -> reference idct -> x)
|
472 |
* - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
|
473 |
* - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
|
474 |
*/
|
475 |
uint8_t idct_permutation[64];
|
476 |
int idct_permutation_type;
|
477 |
#define FF_NO_IDCT_PERM 1 |
478 |
#define FF_LIBMPEG2_IDCT_PERM 2 |
479 |
#define FF_SIMPLE_IDCT_PERM 3 |
480 |
#define FF_TRANSPOSE_IDCT_PERM 4 |
481 |
#define FF_PARTTRANS_IDCT_PERM 5 |
482 |
#define FF_SSE2_IDCT_PERM 6 |
483 |
|
484 |
int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); |
485 |
void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); |
486 |
#define BASIS_SHIFT 16 |
487 |
#define RECON_SHIFT 6 |
488 |
|
489 |
void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w); |
490 |
#define EDGE_WIDTH 16 |
491 |
|
492 |
void (*prefetch)(void *mem, int stride, int h); |
493 |
|
494 |
void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); |
495 |
|
496 |
/* mlp/truehd functions */
|
497 |
void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff, |
498 |
int firorder, int iirorder, |
499 |
unsigned int filter_shift, int32_t mask, int blocksize, |
500 |
int32_t *sample_buffer); |
501 |
|
502 |
/* vc1 functions */
|
503 |
void (*vc1_inv_trans_8x8)(DCTELEM *b);
|
504 |
void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block); |
505 |
void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block); |
506 |
void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block); |
507 |
void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
508 |
void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
509 |
void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
510 |
void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block); |
511 |
void (*vc1_v_overlap)(uint8_t* src, int stride); |
512 |
void (*vc1_h_overlap)(uint8_t* src, int stride); |
513 |
void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq); |
514 |
void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq); |
515 |
void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq); |
516 |
void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq); |
517 |
void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq); |
518 |
void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq); |
519 |
/* put 8x8 block with bicubic interpolation and quarterpel precision
|
520 |
* last argument is actually round value instead of height
|
521 |
*/
|
522 |
op_pixels_func put_vc1_mspel_pixels_tab[16];
|
523 |
op_pixels_func avg_vc1_mspel_pixels_tab[16];
|
524 |
|
525 |
/* intrax8 functions */
|
526 |
void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize); |
527 |
void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize, |
528 |
int * range, int * sum, int edges); |
529 |
|
530 |
/**
|
531 |
* Calculate scalar product of two vectors.
|
532 |
* @param len length of vectors, should be multiple of 16
|
533 |
* @param shift number of bits to discard from product
|
534 |
*/
|
535 |
int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift); |
536 |
/* ape functions */
|
537 |
/**
|
538 |
* Calculate scalar product of v1 and v2,
|
539 |
* and v1[i] += v3[i] * mul
|
540 |
* @param len length of vectors, should be multiple of 16
|
541 |
*/
|
542 |
int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul); |
543 |
|
544 |
/* rv30 functions */
|
545 |
qpel_mc_func put_rv30_tpel_pixels_tab[4][16]; |
546 |
qpel_mc_func avg_rv30_tpel_pixels_tab[4][16]; |
547 |
|
548 |
/* rv40 functions */
|
549 |
qpel_mc_func put_rv40_qpel_pixels_tab[4][16]; |
550 |
qpel_mc_func avg_rv40_qpel_pixels_tab[4][16]; |
551 |
h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
|
552 |
h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
|
553 |
|
554 |
/* bink functions */
|
555 |
op_fill_func fill_block_tab[2];
|
556 |
void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize); |
557 |
} DSPContext; |
558 |
|
559 |
void dsputil_static_init(void); |
560 |
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
|
561 |
|
562 |
int ff_check_alignment(void); |
563 |
|
564 |
/**
|
565 |
* permute block according to permuatation.
|
566 |
* @param last last non zero element in scantable order
|
567 |
*/
|
568 |
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
569 |
|
570 |
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); |
571 |
|
572 |
#define BYTE_VEC32(c) ((c)*0x01010101UL) |
573 |
|
574 |
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) |
575 |
{ |
576 |
return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); |
577 |
} |
578 |
|
579 |
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) |
580 |
{ |
581 |
return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); |
582 |
} |
583 |
|
584 |
static inline int get_penalty_factor(int lambda, int lambda2, int type){ |
585 |
switch(type&0xFF){ |
586 |
default:
|
587 |
case FF_CMP_SAD:
|
588 |
return lambda>>FF_LAMBDA_SHIFT;
|
589 |
case FF_CMP_DCT:
|
590 |
return (3*lambda)>>(FF_LAMBDA_SHIFT+1); |
591 |
case FF_CMP_W53:
|
592 |
return (4*lambda)>>(FF_LAMBDA_SHIFT); |
593 |
case FF_CMP_W97:
|
594 |
return (2*lambda)>>(FF_LAMBDA_SHIFT); |
595 |
case FF_CMP_SATD:
|
596 |
case FF_CMP_DCT264:
|
597 |
return (2*lambda)>>FF_LAMBDA_SHIFT; |
598 |
case FF_CMP_RD:
|
599 |
case FF_CMP_PSNR:
|
600 |
case FF_CMP_SSE:
|
601 |
case FF_CMP_NSSE:
|
602 |
return lambda2>>FF_LAMBDA_SHIFT;
|
603 |
case FF_CMP_BIT:
|
604 |
return 1; |
605 |
} |
606 |
} |
607 |
|
608 |
/**
|
609 |
* Empty mmx state.
|
610 |
* this must be called between any dsp function and float/double code.
|
611 |
* for example sin(); dsp->idct_put(); emms_c(); cos()
|
612 |
*/
|
613 |
#define emms_c()
|
614 |
|
615 |
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
|
616 |
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
|
617 |
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
|
618 |
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
|
619 |
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
|
620 |
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
|
621 |
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
|
622 |
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
|
623 |
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
|
624 |
|
625 |
void ff_dsputil_init_dwt(DSPContext *c);
|
626 |
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
|
627 |
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
|
628 |
void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
|
629 |
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
|
630 |
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
|
631 |
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
|
632 |
|
633 |
#if HAVE_MMX
|
634 |
|
635 |
#undef emms_c
|
636 |
|
637 |
static inline void emms(void) |
638 |
{ |
639 |
__asm__ volatile ("emms;":::"memory"); |
640 |
} |
641 |
|
642 |
#define emms_c() emms()
|
643 |
|
644 |
#elif ARCH_ARM
|
645 |
|
646 |
#if HAVE_NEON
|
647 |
# define STRIDE_ALIGN 16 |
648 |
#endif
|
649 |
|
650 |
#elif ARCH_PPC
|
651 |
|
652 |
#define STRIDE_ALIGN 16 |
653 |
|
654 |
#elif HAVE_MMI
|
655 |
|
656 |
#define STRIDE_ALIGN 16 |
657 |
|
658 |
#endif
|
659 |
|
660 |
#ifndef STRIDE_ALIGN
|
661 |
# define STRIDE_ALIGN 8 |
662 |
#endif
|
663 |
|
664 |
#define LOCAL_ALIGNED_A(a, t, v, s, o, ...) \
|
665 |
uint8_t la_##v[sizeof(t s o) + (a)]; \ |
666 |
t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a) |
667 |
|
668 |
#define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
|
669 |
|
670 |
#define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
|
671 |
|
672 |
#if HAVE_LOCAL_ALIGNED_8
|
673 |
# define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,) |
674 |
#else
|
675 |
# define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__) |
676 |
#endif
|
677 |
|
678 |
#if HAVE_LOCAL_ALIGNED_16
|
679 |
# define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,) |
680 |
#else
|
681 |
# define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__) |
682 |
#endif
|
683 |
|
684 |
/* PSNR */
|
685 |
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3], |
686 |
int orig_linesize[3], int coded_linesize, |
687 |
AVCodecContext *avctx); |
688 |
|
689 |
#define WRAPPER8_16(name8, name16)\
|
690 |
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
691 |
return name8(s, dst , src , stride, h)\
|
692 |
+name8(s, dst+8 , src+8 , stride, h);\ |
693 |
} |
694 |
|
695 |
#define WRAPPER8_16_SQ(name8, name16)\
|
696 |
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ |
697 |
int score=0;\ |
698 |
score +=name8(s, dst , src , stride, 8);\
|
699 |
score +=name8(s, dst+8 , src+8 , stride, 8);\ |
700 |
if(h==16){\ |
701 |
dst += 8*stride;\
|
702 |
src += 8*stride;\
|
703 |
score +=name8(s, dst , src , stride, 8);\
|
704 |
score +=name8(s, dst+8 , src+8 , stride, 8);\ |
705 |
}\ |
706 |
return score;\
|
707 |
} |
708 |
|
709 |
|
710 |
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
711 |
{ |
712 |
int i;
|
713 |
for(i=0; i<h; i++) |
714 |
{ |
715 |
AV_WN16(dst , AV_RN16(src )); |
716 |
dst+=dstStride; |
717 |
src+=srcStride; |
718 |
} |
719 |
} |
720 |
|
721 |
static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
722 |
{ |
723 |
int i;
|
724 |
for(i=0; i<h; i++) |
725 |
{ |
726 |
AV_WN32(dst , AV_RN32(src )); |
727 |
dst+=dstStride; |
728 |
src+=srcStride; |
729 |
} |
730 |
} |
731 |
|
732 |
static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
733 |
{ |
734 |
int i;
|
735 |
for(i=0; i<h; i++) |
736 |
{ |
737 |
AV_WN32(dst , AV_RN32(src )); |
738 |
AV_WN32(dst+4 , AV_RN32(src+4 )); |
739 |
dst+=dstStride; |
740 |
src+=srcStride; |
741 |
} |
742 |
} |
743 |
|
744 |
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
745 |
{ |
746 |
int i;
|
747 |
for(i=0; i<h; i++) |
748 |
{ |
749 |
AV_WN32(dst , AV_RN32(src )); |
750 |
AV_WN32(dst+4 , AV_RN32(src+4 )); |
751 |
dst[8]= src[8]; |
752 |
dst+=dstStride; |
753 |
src+=srcStride; |
754 |
} |
755 |
} |
756 |
|
757 |
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
758 |
{ |
759 |
int i;
|
760 |
for(i=0; i<h; i++) |
761 |
{ |
762 |
AV_WN32(dst , AV_RN32(src )); |
763 |
AV_WN32(dst+4 , AV_RN32(src+4 )); |
764 |
AV_WN32(dst+8 , AV_RN32(src+8 )); |
765 |
AV_WN32(dst+12, AV_RN32(src+12)); |
766 |
dst+=dstStride; |
767 |
src+=srcStride; |
768 |
} |
769 |
} |
770 |
|
771 |
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) |
772 |
{ |
773 |
int i;
|
774 |
for(i=0; i<h; i++) |
775 |
{ |
776 |
AV_WN32(dst , AV_RN32(src )); |
777 |
AV_WN32(dst+4 , AV_RN32(src+4 )); |
778 |
AV_WN32(dst+8 , AV_RN32(src+8 )); |
779 |
AV_WN32(dst+12, AV_RN32(src+12)); |
780 |
dst[16]= src[16]; |
781 |
dst+=dstStride; |
782 |
src+=srcStride; |
783 |
} |
784 |
} |
785 |
|
786 |
#endif /* AVCODEC_DSPUTIL_H */ |