FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
dsputil.h
Go to the documentation of this file.
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 ff_fdct_ifast (DCTELEM *data);
47 
48 void ff_j_rev_dct (DCTELEM *data);
49 void ff_j_rev_dct4 (DCTELEM *data);
50 void ff_j_rev_dct2 (DCTELEM *data);
51 void ff_j_rev_dct1 (DCTELEM *data);
53 
57 
58 #define H264_IDCT(depth) \
59 void ff_h264_idct8_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
60 void ff_h264_idct_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
61 void ff_h264_idct8_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
62 void ff_h264_idct_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
63 void ff_h264_idct_add16_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
64 void ff_h264_idct_add16intra_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
65 void ff_h264_idct8_add4_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
66 void ff_h264_idct_add8_422_ ## depth ## _c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
67 void ff_h264_idct_add8_ ## depth ## _c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
68 void ff_h264_luma_dc_dequant_idct_ ## depth ## _c(DCTELEM *output, DCTELEM *input, int qmul);\
69 void ff_h264_chroma422_dc_dequant_idct_ ## depth ## _c(DCTELEM *block, int qmul);\
70 void ff_h264_chroma_dc_dequant_idct_ ## depth ## _c(DCTELEM *block, int qmul);
71 
72 H264_IDCT( 8)
73 H264_IDCT( 9)
74 H264_IDCT(10)
75 H264_IDCT(12)
76 H264_IDCT(14)
77 
78 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp);
79 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
80 
81 /* encoding scans */
82 extern const uint8_t ff_alternate_horizontal_scan[64];
83 extern const uint8_t ff_alternate_vertical_scan[64];
84 extern const uint8_t ff_zigzag_direct[64];
85 extern const uint8_t ff_zigzag248_direct[64];
86 
87 /* pixel operations */
88 #define MAX_NEG_CROP 1024
89 
90 /* temporary */
91 extern uint32_t ff_squareTbl[512];
92 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
93 
94 #define PUTAVG_PIXELS(depth)\
95 void ff_put_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
96 void ff_avg_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
97 void ff_put_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
98 void ff_avg_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);
99 
100 PUTAVG_PIXELS( 8)
101 PUTAVG_PIXELS( 9)
102 PUTAVG_PIXELS(10)
103 PUTAVG_PIXELS(12)
104 PUTAVG_PIXELS(14)
105 
106 #define ff_put_pixels8x8_c ff_put_pixels8x8_8_c
107 #define ff_avg_pixels8x8_c ff_avg_pixels8x8_8_c
108 #define ff_put_pixels16x16_c ff_put_pixels16x16_8_c
109 #define ff_avg_pixels16x16_c ff_avg_pixels16x16_8_c
110 
111 /* RV40 functions */
112 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
113 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
114 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
115 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
116 
117 /* 1/2^n downscaling functions from imgconvert.c */
118 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
119 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
120 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
121 
122 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
123  int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
124 
125 /* minimum alignment rules ;)
126 If you notice errors in the align stuff, need more alignment for some ASM code
127 for some CPU or need to use a function with less aligned data then send a mail
128 to the ffmpeg-devel mailing list, ...
129 
130 !warning These alignments might not match reality, (missing attribute((align))
131 stuff somewhere possible).
132 I (Michael) did not check them, these are just the alignments which I think
133 could be reached easily ...
134 
135 !future video codecs might need functions with less strict alignment
136 */
137 
138 /*
139 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
140 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
141 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
142 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
143 void clear_blocks_c(DCTELEM *blocks);
144 */
145 
146 /* add and put pixel (decoding) */
147 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
148 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4
149 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
150 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);
151 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
152 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
153 
154 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
155 
156 #define DEF_OLD_QPEL(name)\
157 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
158 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
159 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
160 
161 DEF_OLD_QPEL(qpel16_mc11_old_c)
162 DEF_OLD_QPEL(qpel16_mc31_old_c)
163 DEF_OLD_QPEL(qpel16_mc12_old_c)
164 DEF_OLD_QPEL(qpel16_mc32_old_c)
165 DEF_OLD_QPEL(qpel16_mc13_old_c)
166 DEF_OLD_QPEL(qpel16_mc33_old_c)
167 DEF_OLD_QPEL(qpel8_mc11_old_c)
168 DEF_OLD_QPEL(qpel8_mc31_old_c)
169 DEF_OLD_QPEL(qpel8_mc12_old_c)
170 DEF_OLD_QPEL(qpel8_mc32_old_c)
171 DEF_OLD_QPEL(qpel8_mc13_old_c)
172 DEF_OLD_QPEL(qpel8_mc33_old_c)
173 
174 #define CALL_2X_PIXELS(a, b, n)\
175 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
176  b(block , pixels , line_size, h);\
177  b(block+n, pixels+n, line_size, h);\
178 }
179 
180 /* motion estimation */
181 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
182 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
183 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))*/;
184 
185 /**
186  * Scantable.
187  */
188 typedef struct ScanTable{
192 } ScanTable;
193 
194 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
195 void ff_init_scantable_permutation(uint8_t *idct_permutation,
196  int idct_permutation_type);
197 
198 #define EMULATED_EDGE(depth) \
199 void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, ptrdiff_t linesize,\
200  int block_w, int block_h,\
201  int src_x, int src_y, int w, int h);
202 
203 EMULATED_EDGE(8)
204 EMULATED_EDGE(16)
205 
206 /**
207  * DSPContext.
208  */
209 typedef struct DSPContext {
210  /**
211  * Size of DCT coefficients.
212  */
213  int dct_bits;
214 
215  /* pixel ops : interface with DCT */
216  void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
217  void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
218  void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
219  void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
220  void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
221  void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
222  void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
223  int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
224  /**
225  * translational global motion compensation.
226  */
227  void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
228  /**
229  * global motion compensation.
230  */
231  void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
232  int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
233  void (*clear_block)(DCTELEM *block/*align 16*/);
234  void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
235  int (*pix_sum)(uint8_t * pix, int line_size);
236  int (*pix_norm1)(uint8_t * pix, int line_size);
237 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
238 
239  me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
241  me_cmp_func hadamard8_diff[6];
242  me_cmp_func dct_sad[6];
243  me_cmp_func quant_psnr[6];
244  me_cmp_func bit[6];
245  me_cmp_func rd[6];
246  me_cmp_func vsad[6];
247  me_cmp_func vsse[6];
248  me_cmp_func nsse[6];
249  me_cmp_func w53[6];
250  me_cmp_func w97[6];
251  me_cmp_func dct_max[6];
252  me_cmp_func dct264_sad[6];
253 
254  me_cmp_func me_pre_cmp[6];
255  me_cmp_func me_cmp[6];
256  me_cmp_func me_sub_cmp[6];
257  me_cmp_func mb_cmp[6];
258  me_cmp_func ildct_cmp[6]; //only width 16 used
259  me_cmp_func frame_skip_cmp[6]; //only width 8 used
260 
261  int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
262  int size);
263 
264  /**
265  * Halfpel motion compensation with rounding (a+b+1)>>1.
266  * this is an array[4][4] of motion compensation functions for 4
267  * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
268  * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
269  * @param block destination where the result is stored
270  * @param pixels source
271  * @param line_size number of bytes in a horizontal line of block
272  * @param h height
273  */
274  op_pixels_func put_pixels_tab[4][4];
275 
276  /**
277  * Halfpel motion compensation with rounding (a+b+1)>>1.
278  * This is an array[4][4] of motion compensation functions for 4
279  * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
280  * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
281  * @param block destination into which the result is averaged (a+b+1)>>1
282  * @param pixels source
283  * @param line_size number of bytes in a horizontal line of block
284  * @param h height
285  */
286  op_pixels_func avg_pixels_tab[4][4];
287 
288  /**
289  * Halfpel motion compensation with no rounding (a+b)>>1.
290  * this is an array[2][4] of motion compensation functions for 2
291  * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
292  * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
293  * @param block destination where the result is stored
294  * @param pixels source
295  * @param line_size number of bytes in a horizontal line of block
296  * @param h height
297  */
298  op_pixels_func put_no_rnd_pixels_tab[4][4];
299 
300  /**
301  * Halfpel motion compensation with no rounding (a+b)>>1.
302  * this is an array[2][4] of motion compensation functions for 2
303  * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
304  * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
305  * @param block destination into which the result is averaged (a+b)>>1
306  * @param pixels source
307  * @param line_size number of bytes in a horizontal line of block
308  * @param h height
309  */
310  op_pixels_func avg_no_rnd_pixels_tab[4][4];
311 
312  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);
313 
314  /**
315  * Thirdpel motion compensation with rounding (a+b+1)>>1.
316  * this is an array[12] of motion compensation functions for the 9 thirdpe
317  * positions<br>
318  * *pixels_tab[ xthirdpel + 4*ythirdpel ]
319  * @param block destination where the result is stored
320  * @param pixels source
321  * @param line_size number of bytes in a horizontal line of block
322  * @param h height
323  */
324  tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
325  tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
326 
327  qpel_mc_func put_qpel_pixels_tab[2][16];
328  qpel_mc_func avg_qpel_pixels_tab[2][16];
329  qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
330  qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
331  qpel_mc_func put_mspel_pixels_tab[8];
332 
333  /**
334  * h264 Chroma MC
335  */
336  h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
337  h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
338 
339  qpel_mc_func put_h264_qpel_pixels_tab[4][16];
340  qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
341 
342  qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
343  qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
344 
345  me_cmp_func pix_abs[2][4];
346 
347  /* huffyuv specific */
348  void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
349  void (*diff_bytes)(uint8_t *dst/*align 16*/, const uint8_t *src1/*align 16*/, const 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 (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
360  void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
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  /* assume len is a multiple of 4, and arrays are 16-byte aligned */
368  void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
369  /* assume len is a multiple of 16, and arrays are 32-byte aligned */
370  void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
371  /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
372  void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
373  /* assume len is a multiple of 4, and arrays are 16-byte aligned */
374  void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
375  /* assume len is a multiple of 8, and arrays are 16-byte aligned */
376  void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
377  /**
378  * Calculate the scalar product of two vectors of floats.
379  * @param v1 first vector, 16-byte aligned
380  * @param v2 second vector, 16-byte aligned
381  * @param len length of vectors, multiple of 4
382  */
383  float (*scalarproduct_float)(const float *v1, const float *v2, int len);
384  /**
385  * Calculate the sum and difference of two vectors of floats.
386  * @param v1 first input vector, sum output, 16-byte aligned
387  * @param v2 second input vector, difference output, 16-byte aligned
388  * @param len length of vectors, multiple of 4
389  */
390  void (*butterflies_float)(float *av_restrict v1, float *av_restrict v2, int len);
391 
392  /**
393  * Calculate the sum and difference of two vectors of floats and interleave
394  * results into a separate output vector of floats, with each sum
395  * positioned before the corresponding difference.
396  *
397  * @param dst output vector
398  * constraints: 16-byte aligned
399  * @param src0 first input vector
400  * constraints: 32-byte aligned
401  * @param src1 second input vector
402  * constraints: 32-byte aligned
403  * @param len number of elements in the input
404  * constraints: multiple of 8
405  */
406  void (*butterflies_float_interleave)(float *dst, const float *src0,
407  const float *src1, int len);
408 
409  /* (I)DCT */
410  void (*fdct)(DCTELEM *block/* align 16*/);
411  void (*fdct248)(DCTELEM *block/* align 16*/);
412 
413  /* IDCT really*/
414  void (*idct)(DCTELEM *block/* align 16*/);
415 
416  /**
417  * block -> idct -> clip to unsigned 8 bit -> dest.
418  * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
419  * @param line_size size in bytes of a horizontal line of dest
420  */
421  void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
422 
423  /**
424  * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
425  * @param line_size size in bytes of a horizontal line of dest
426  */
427  void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
428 
429  /**
430  * idct input permutation.
431  * several optimized IDCTs need a permutated input (relative to the normal order of the reference
432  * IDCT)
433  * this permutation must be performed before the idct_put/add, note, normally this can be merged
434  * with the zigzag/alternate scan<br>
435  * an example to avoid confusion:
436  * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
437  * - (x -> reference dct -> reference idct -> x)
438  * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
439  * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
440  */
441  uint8_t idct_permutation[64];
443 #define FF_NO_IDCT_PERM 1
444 #define FF_LIBMPEG2_IDCT_PERM 2
445 #define FF_SIMPLE_IDCT_PERM 3
446 #define FF_TRANSPOSE_IDCT_PERM 4
447 #define FF_PARTTRANS_IDCT_PERM 5
448 #define FF_SSE2_IDCT_PERM 6
449 
450  int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
451  void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
452 #define BASIS_SHIFT 16
453 #define RECON_SHIFT 6
454 
455  void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides);
456 #define EDGE_WIDTH 16
457 #define EDGE_TOP 1
458 #define EDGE_BOTTOM 2
459 
460  void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
461 
462  /**
463  * Calculate scalar product of two vectors.
464  * @param len length of vectors, should be multiple of 16
465  */
466  int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len);
467  /* ape functions */
468  /**
469  * Calculate scalar product of v1 and v2,
470  * and v1[i] += v3[i] * mul
471  * @param len length of vectors, should be multiple of 16
472  */
473  int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
474 
475  /**
476  * Apply symmetric window in 16-bit fixed-point.
477  * @param output destination array
478  * constraints: 16-byte aligned
479  * @param input source array
480  * constraints: 16-byte aligned
481  * @param window window array
482  * constraints: 16-byte aligned, at least len/2 elements
483  * @param len full window length
484  * constraints: multiple of ? greater than zero
485  */
486  void (*apply_window_int16)(int16_t *output, const int16_t *input,
487  const int16_t *window, unsigned int len);
488 
489  /**
490  * Clip each element in an array of int32_t to a given minimum and maximum value.
491  * @param dst destination array
492  * constraints: 16-byte aligned
493  * @param src source array
494  * constraints: 16-byte aligned
495  * @param min minimum value
496  * constraints: must be in the range [-(1 << 24), 1 << 24]
497  * @param max maximum value
498  * constraints: must be in the range [-(1 << 24), 1 << 24]
499  * @param len number of elements in the array
500  * constraints: multiple of 32 greater than zero
501  */
502  void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
503  int32_t max, unsigned int len);
504 
505  op_fill_func fill_block_tab[2];
506 } DSPContext;
507 
508 void ff_dsputil_static_init(void);
509 void ff_dsputil_init(DSPContext* p, AVCodecContext *avctx);
511 
512 int ff_check_alignment(void);
513 
514 /**
515  * Return the scalar product of two vectors.
516  *
517  * @param v1 first input vector
518  * @param v2 first input vector
519  * @param len number of elements
520  *
521  * @return sum of elementwise products
522  */
523 float ff_scalarproduct_float_c(const float *v1, const float *v2, int len);
524 
525 /**
526  * permute block according to permuatation.
527  * @param last last non zero element in scantable order
528  */
529 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
530 
531 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
532 
533 #define BYTE_VEC32(c) ((c)*0x01010101UL)
534 #define BYTE_VEC64(c) ((c)*0x0001000100010001UL)
535 
536 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
537 {
538  return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
539 }
540 
541 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
542 {
543  return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
544 }
545 
546 static inline uint64_t rnd_avg64(uint64_t a, uint64_t b)
547 {
548  return (a | b) - (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
549 }
550 
551 static inline uint64_t no_rnd_avg64(uint64_t a, uint64_t b)
552 {
553  return (a & b) + (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
554 }
555 
556 static inline int get_penalty_factor(int lambda, int lambda2, int type){
557  switch(type&0xFF){
558  default:
559  case FF_CMP_SAD:
560  return lambda>>FF_LAMBDA_SHIFT;
561  case FF_CMP_DCT:
562  return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
563  case FF_CMP_W53:
564  return (4*lambda)>>(FF_LAMBDA_SHIFT);
565  case FF_CMP_W97:
566  return (2*lambda)>>(FF_LAMBDA_SHIFT);
567  case FF_CMP_SATD:
568  case FF_CMP_DCT264:
569  return (2*lambda)>>FF_LAMBDA_SHIFT;
570  case FF_CMP_RD:
571  case FF_CMP_PSNR:
572  case FF_CMP_SSE:
573  case FF_CMP_NSSE:
574  return lambda2>>FF_LAMBDA_SHIFT;
575  case FF_CMP_BIT:
576  return 1;
577  }
578 }
579 
588 
590 
591 #if (ARCH_ARM && HAVE_NEON) || ARCH_PPC || HAVE_MMX
592 # define STRIDE_ALIGN 16
593 #else
594 # define STRIDE_ALIGN 8
595 #endif
596 
597 // Some broken preprocessors need a second expansion
598 // to be forced to tokenize __VA_ARGS__
599 #define E(x) x
600 
601 #define LOCAL_ALIGNED_A(a, t, v, s, o, ...) \
602  uint8_t la_##v[sizeof(t s o) + (a)]; \
603  t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
604 
605 #define LOCAL_ALIGNED_D(a, t, v, s, o, ...) \
606  DECLARE_ALIGNED(a, t, la_##v) s o; \
607  t (*v) o = la_##v
608 
609 #define LOCAL_ALIGNED(a, t, v, ...) E(LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,))
610 
611 #if HAVE_LOCAL_ALIGNED_8
612 # define LOCAL_ALIGNED_8(t, v, ...) E(LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,))
613 #else
614 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
615 #endif
616 
617 #if HAVE_LOCAL_ALIGNED_16
618 # define LOCAL_ALIGNED_16(t, v, ...) E(LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,))
619 #else
620 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
621 #endif
622 
623 #define WRAPPER8_16_SQ(name8, name16)\
624 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
625  int score=0;\
626  score +=name8(s, dst , src , stride, 8);\
627  score +=name8(s, dst+8 , src+8 , stride, 8);\
628  if(h==16){\
629  dst += 8*stride;\
630  src += 8*stride;\
631  score +=name8(s, dst , src , stride, 8);\
632  score +=name8(s, dst+8 , src+8 , stride, 8);\
633  }\
634  return score;\
635 }
636 
637 
638 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
639 {
640  int i;
641  for(i=0; i<h; i++)
642  {
643  AV_COPY16U(dst, src);
644  dst+=dstStride;
645  src+=srcStride;
646  }
647 }
648 
649 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
650 {
651  int i;
652  for(i=0; i<h; i++)
653  {
654  AV_COPY32U(dst, src);
655  dst+=dstStride;
656  src+=srcStride;
657  }
658 }
659 
660 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
661 {
662  int i;
663  for(i=0; i<h; i++)
664  {
665  AV_COPY64U(dst, src);
666  dst+=dstStride;
667  src+=srcStride;
668  }
669 }
670 
671 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
672 {
673  int i;
674  for(i=0; i<h; i++)
675  {
676  AV_COPY64U(dst, src);
677  dst[8]= src[8];
678  dst+=dstStride;
679  src+=srcStride;
680  }
681 }
682 
683 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
684 {
685  int i;
686  for(i=0; i<h; i++)
687  {
688  AV_COPY128U(dst, src);
689  dst+=dstStride;
690  src+=srcStride;
691  }
692 }
693 
694 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
695 {
696  int i;
697  for(i=0; i<h; i++)
698  {
699  AV_COPY128U(dst, src);
700  dst[16]= src[16];
701  dst+=dstStride;
702  src+=srcStride;
703  }
704 }
705 
706 #endif /* AVCODEC_DSPUTIL_H */