3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
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.
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.
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
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
30 #ifndef AVCODEC_DSPUTIL_H
31 #define AVCODEC_DSPUTIL_H
33 #include "libavutil/intreadwrite.h"
39 typedef short DCTELEM;
41 void ff_fdct_ifast (DCTELEM *data);
42 void ff_fdct_ifast248 (DCTELEM *data);
43 void ff_jpeg_fdct_islow_8(DCTELEM *data);
44 void ff_jpeg_fdct_islow_10(DCTELEM *data);
45 void ff_fdct248_islow_8(DCTELEM *data);
46 void ff_fdct248_islow_10(DCTELEM *data);
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);
52 void ff_wmv2_idct_c(DCTELEM *data);
54 void ff_fdct_mmx(DCTELEM *block);
55 void ff_fdct_mmx2(DCTELEM *block);
56 void ff_fdct_sse2(DCTELEM *block);
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);
76 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp);
77 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
80 extern const uint8_t ff_alternate_horizontal_scan[64];
81 extern const uint8_t ff_alternate_vertical_scan[64];
82 extern const uint8_t ff_zigzag_direct[64];
83 extern const uint8_t ff_zigzag248_direct[64];
85 /* pixel operations */
86 #define MAX_NEG_CROP 1024
89 extern uint32_t ff_squareTbl[512];
90 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
92 #define PUTAVG_PIXELS(depth)\
93 void ff_put_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
94 void ff_avg_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
95 void ff_put_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
96 void ff_avg_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);
102 #define ff_put_pixels8x8_c ff_put_pixels8x8_8_c
103 #define ff_avg_pixels8x8_c ff_avg_pixels8x8_8_c
104 #define ff_put_pixels16x16_c ff_put_pixels16x16_8_c
105 #define ff_avg_pixels16x16_c ff_avg_pixels16x16_8_c
107 /* VP3 DSP functions */
108 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
109 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
110 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
111 void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
113 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
114 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
117 void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
120 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
121 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
122 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
123 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
125 /* 1/2^n downscaling functions from imgconvert.c */
126 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
127 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
128 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
130 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
131 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
133 /* minimum alignment rules ;)
134 If you notice errors in the align stuff, need more alignment for some ASM code
135 for some CPU or need to use a function with less aligned data then send a mail
136 to the ffmpeg-devel mailing list, ...
138 !warning These alignments might not match reality, (missing attribute((align))
139 stuff somewhere possible).
140 I (Michael) did not check them, these are just the alignments which I think
141 could be reached easily ...
143 !future video codecs might need functions with less strict alignment
147 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
148 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
149 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
150 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
151 void clear_blocks_c(DCTELEM *blocks);
154 /* add and put pixel (decoding) */
155 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
156 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4
157 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
158 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);
159 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
160 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
162 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
164 #define DEF_OLD_QPEL(name)\
165 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
166 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
167 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
169 DEF_OLD_QPEL(qpel16_mc11_old_c)
170 DEF_OLD_QPEL(qpel16_mc31_old_c)
171 DEF_OLD_QPEL(qpel16_mc12_old_c)
172 DEF_OLD_QPEL(qpel16_mc32_old_c)
173 DEF_OLD_QPEL(qpel16_mc13_old_c)
174 DEF_OLD_QPEL(qpel16_mc33_old_c)
175 DEF_OLD_QPEL(qpel8_mc11_old_c)
176 DEF_OLD_QPEL(qpel8_mc31_old_c)
177 DEF_OLD_QPEL(qpel8_mc12_old_c)
178 DEF_OLD_QPEL(qpel8_mc32_old_c)
179 DEF_OLD_QPEL(qpel8_mc13_old_c)
180 DEF_OLD_QPEL(qpel8_mc33_old_c)
182 #define CALL_2X_PIXELS(a, b, n)\
183 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
184 b(block , pixels , line_size, h);\
185 b(block+n, pixels+n, line_size, h);\
188 /* motion estimation */
189 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
190 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
191 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))*/;
196 typedef struct ScanTable{
197 const uint8_t *scantable;
198 uint8_t permutated[64];
199 uint8_t raster_end[64];
202 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
203 void ff_init_scantable_permutation(uint8_t *idct_permutation,
204 int idct_permutation_type);
206 #define EMULATED_EDGE(depth) \
207 void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, int linesize,\
208 int block_w, int block_h,\
209 int src_x, int src_y, int w, int h);
215 void ff_add_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
216 void ff_put_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
217 void ff_put_signed_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
222 typedef struct DSPContext {
224 * Size of DCT coefficients.
228 /* pixel ops : interface with DCT */
229 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
230 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
231 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
232 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
233 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
234 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
235 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
236 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
238 * Motion estimation with emulated edge values.
239 * @param buf pointer to destination buffer (unaligned)
240 * @param src pointer to pixel source (unaligned)
241 * @param linesize width (in pixels) for src/buf
242 * @param block_w number of pixels (per row) to copy to buf
243 * @param block_h nummber of pixel rows to copy to buf
244 * @param src_x offset of src to start of row - this may be negative
245 * @param src_y offset of src to top of image - this may be negative
246 * @param w width of src in pixels
247 * @param h height of src in pixels
249 void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
250 int block_w, int block_h,
251 int src_x, int src_y, int w, int h);
253 * translational global motion compensation.
255 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
257 * global motion compensation.
259 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
260 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
261 void (*clear_block)(DCTELEM *block/*align 16*/);
262 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
263 int (*pix_sum)(uint8_t * pix, int line_size);
264 int (*pix_norm1)(uint8_t * pix, int line_size);
265 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
267 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
269 me_cmp_func hadamard8_diff[6];
270 me_cmp_func dct_sad[6];
271 me_cmp_func quant_psnr[6];
279 me_cmp_func dct_max[6];
280 me_cmp_func dct264_sad[6];
282 me_cmp_func me_pre_cmp[6];
283 me_cmp_func me_cmp[6];
284 me_cmp_func me_sub_cmp[6];
285 me_cmp_func mb_cmp[6];
286 me_cmp_func ildct_cmp[6]; //only width 16 used
287 me_cmp_func frame_skip_cmp[6]; //only width 8 used
289 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
293 * Halfpel motion compensation with rounding (a+b+1)>>1.
294 * this is an array[4][4] of motion compensation functions for 4
295 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
296 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
297 * @param block destination where the result is stored
298 * @param pixels source
299 * @param line_size number of bytes in a horizontal line of block
302 op_pixels_func put_pixels_tab[4][4];
305 * Halfpel motion compensation with rounding (a+b+1)>>1.
306 * This is an array[4][4] of motion compensation functions for 4
307 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
308 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
309 * @param block destination into which the result is averaged (a+b+1)>>1
310 * @param pixels source
311 * @param line_size number of bytes in a horizontal line of block
314 op_pixels_func avg_pixels_tab[4][4];
317 * Halfpel motion compensation with no rounding (a+b)>>1.
318 * this is an array[2][4] of motion compensation functions for 2
319 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
320 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
321 * @param block destination where the result is stored
322 * @param pixels source
323 * @param line_size number of bytes in a horizontal line of block
326 op_pixels_func put_no_rnd_pixels_tab[4][4];
329 * Halfpel motion compensation with no rounding (a+b)>>1.
330 * this is an array[2][4] of motion compensation functions for 2
331 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
332 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
333 * @param block destination into which the result is averaged (a+b)>>1
334 * @param pixels source
335 * @param line_size number of bytes in a horizontal line of block
338 op_pixels_func avg_no_rnd_pixels_tab[4][4];
340 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);
343 * Thirdpel motion compensation with rounding (a+b+1)>>1.
344 * this is an array[12] of motion compensation functions for the 9 thirdpe
346 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
347 * @param block destination where the result is stored
348 * @param pixels source
349 * @param line_size number of bytes in a horizontal line of block
352 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
353 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
355 qpel_mc_func put_qpel_pixels_tab[2][16];
356 qpel_mc_func avg_qpel_pixels_tab[2][16];
357 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
358 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
359 qpel_mc_func put_mspel_pixels_tab[8];
364 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
365 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
367 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
368 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
370 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
371 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
373 me_cmp_func pix_abs[2][4];
375 /* huffyuv specific */
376 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
377 void (*diff_bytes)(uint8_t *dst/*align 16*/, const uint8_t *src1/*align 16*/, const uint8_t *src2/*align 1*/,int w);
379 * subtract huffyuv's variant of median prediction
380 * note, this might read from src1[-1], src2[-1]
382 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
383 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
384 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
385 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
386 /* this might write to dst[w] */
387 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
388 void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
390 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
391 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
393 void (*h261_loop_filter)(uint8_t *src, int stride);
395 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
396 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
398 void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
399 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
400 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
402 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
403 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
404 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
405 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
406 void (*vector_fmul)(float *dst, const float *src0, const float *src1, int len);
407 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
408 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
409 void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
410 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
411 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
412 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
413 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
415 * Multiply a vector of floats by a scalar float. Source and
416 * destination vectors must overlap exactly or not at all.
417 * @param dst result vector, 16-byte aligned
418 * @param src input vector, 16-byte aligned
419 * @param mul scalar value
420 * @param len length of vector, multiple of 4
422 void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
425 * Multiply a vector of floats by a scalar float and add to
426 * destination vector. Source and destination vectors must
427 * overlap exactly or not at all.
428 * @param dst result vector, 16-byte aligned
429 * @param src input vector, 16-byte aligned
430 * @param mul scalar value
431 * @param len length of vector, multiple of 4
433 void (*vector_fmac_scalar)(float *dst, const float *src, float mul,
436 * Calculate the scalar product of two vectors of floats.
437 * @param v1 first vector, 16-byte aligned
438 * @param v2 second vector, 16-byte aligned
439 * @param len length of vectors, multiple of 4
441 float (*scalarproduct_float)(const float *v1, const float *v2, int len);
443 * Calculate the sum and difference of two vectors of floats.
444 * @param v1 first input vector, sum output, 16-byte aligned
445 * @param v2 second input vector, difference output, 16-byte aligned
446 * @param len length of vectors, multiple of 4
448 void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
451 * Calculate the sum and difference of two vectors of floats and interleave
452 * results into a separate output vector of floats, with each sum
453 * positioned before the corresponding difference.
455 * @param dst output vector
456 * constraints: 16-byte aligned
457 * @param src0 first input vector
458 * constraints: 32-byte aligned
459 * @param src1 second input vector
460 * constraints: 32-byte aligned
461 * @param len number of elements in the input
462 * constraints: multiple of 8
464 void (*butterflies_float_interleave)(float *dst, const float *src0,
465 const float *src1, int len);
468 void (*fdct)(DCTELEM *block/* align 16*/);
469 void (*fdct248)(DCTELEM *block/* align 16*/);
472 void (*idct)(DCTELEM *block/* align 16*/);
475 * block -> idct -> clip to unsigned 8 bit -> dest.
476 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
477 * @param line_size size in bytes of a horizontal line of dest
479 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
482 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
483 * @param line_size size in bytes of a horizontal line of dest
485 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
488 * idct input permutation.
489 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
491 * this permutation must be performed before the idct_put/add, note, normally this can be merged
492 * with the zigzag/alternate scan<br>
493 * an example to avoid confusion:
494 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
495 * - (x -> reference dct -> reference idct -> x)
496 * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
497 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
499 uint8_t idct_permutation[64];
500 int idct_permutation_type;
501 #define FF_NO_IDCT_PERM 1
502 #define FF_LIBMPEG2_IDCT_PERM 2
503 #define FF_SIMPLE_IDCT_PERM 3
504 #define FF_TRANSPOSE_IDCT_PERM 4
505 #define FF_PARTTRANS_IDCT_PERM 5
506 #define FF_SSE2_IDCT_PERM 6
508 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
509 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
510 #define BASIS_SHIFT 16
511 #define RECON_SHIFT 6
513 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides);
514 #define EDGE_WIDTH 16
516 #define EDGE_BOTTOM 2
518 void (*prefetch)(void *mem, int stride, int h);
520 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
522 /* mlp/truehd functions */
523 void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
524 int firorder, int iirorder,
525 unsigned int filter_shift, int32_t mask, int blocksize,
526 int32_t *sample_buffer);
528 /* intrax8 functions */
529 void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
530 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
531 int * range, int * sum, int edges);
534 * Calculate scalar product of two vectors.
535 * @param len length of vectors, should be multiple of 16
537 int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len);
540 * Calculate scalar product of v1 and v2,
541 * and v1[i] += v3[i] * mul
542 * @param len length of vectors, should be multiple of 16
544 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
547 * Apply symmetric window in 16-bit fixed-point.
548 * @param output destination array
549 * constraints: 16-byte aligned
550 * @param input source array
551 * constraints: 16-byte aligned
552 * @param window window array
553 * constraints: 16-byte aligned, at least len/2 elements
554 * @param len full window length
555 * constraints: multiple of ? greater than zero
557 void (*apply_window_int16)(int16_t *output, const int16_t *input,
558 const int16_t *window, unsigned int len);
561 * Clip each element in an array of int32_t to a given minimum and maximum value.
562 * @param dst destination array
563 * constraints: 16-byte aligned
564 * @param src source array
565 * constraints: 16-byte aligned
566 * @param min minimum value
567 * constraints: must in the the range [-(1<<24), 1<<24]
568 * @param max maximum value
569 * constraints: must in the the range [-(1<<24), 1<<24]
570 * @param len number of elements in the array
571 * constraints: multiple of 32 greater than zero
573 void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
574 int32_t max, unsigned int len);
576 op_fill_func fill_block_tab[2];
579 void ff_dsputil_static_init(void);
580 void ff_dsputil_init(DSPContext* p, AVCodecContext *avctx);
581 attribute_deprecated void dsputil_init(DSPContext* c, AVCodecContext *avctx);
583 int ff_check_alignment(void);
586 * permute block according to permuatation.
587 * @param last last non zero element in scantable order
589 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
591 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
593 #define BYTE_VEC32(c) ((c)*0x01010101UL)
594 #define BYTE_VEC64(c) ((c)*0x0001000100010001UL)
596 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
598 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
601 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
603 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
606 static inline uint64_t rnd_avg64(uint64_t a, uint64_t b)
608 return (a | b) - (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
611 static inline uint64_t no_rnd_avg64(uint64_t a, uint64_t b)
613 return (a & b) + (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
616 static inline int get_penalty_factor(int lambda, int lambda2, int type){
620 return lambda>>FF_LAMBDA_SHIFT;
622 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
624 return (4*lambda)>>(FF_LAMBDA_SHIFT);
626 return (2*lambda)>>(FF_LAMBDA_SHIFT);
629 return (2*lambda)>>FF_LAMBDA_SHIFT;
634 return lambda2>>FF_LAMBDA_SHIFT;
640 void ff_dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
641 void ff_dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
642 void ff_dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
643 void ff_dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
644 void ff_dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
645 void ff_dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
646 void ff_dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
647 void ff_dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
649 void ff_dsputil_init_dwt(DSPContext *c);
650 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
651 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
652 void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
654 #if (ARCH_ARM && HAVE_NEON) || ARCH_PPC || HAVE_MMI || HAVE_MMX
655 # define STRIDE_ALIGN 16
657 # define STRIDE_ALIGN 8
660 #define LOCAL_ALIGNED_A(a, t, v, s, o, ...) \
661 uint8_t la_##v[sizeof(t s o) + (a)]; \
662 t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
664 #define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
666 #define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
668 #if HAVE_LOCAL_ALIGNED_8
669 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
671 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
674 #if HAVE_LOCAL_ALIGNED_16
675 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
677 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
680 #define WRAPPER8_16(name8, name16)\
681 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
682 return name8(s, dst , src , stride, h)\
683 +name8(s, dst+8 , src+8 , stride, h);\
686 #define WRAPPER8_16_SQ(name8, name16)\
687 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
689 score +=name8(s, dst , src , stride, 8);\
690 score +=name8(s, dst+8 , src+8 , stride, 8);\
694 score +=name8(s, dst , src , stride, 8);\
695 score +=name8(s, dst+8 , src+8 , stride, 8);\
701 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
706 AV_WN16(dst , AV_RN16(src ));
712 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
717 AV_WN32(dst , AV_RN32(src ));
723 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
728 AV_WN32(dst , AV_RN32(src ));
729 AV_WN32(dst+4 , AV_RN32(src+4 ));
735 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
740 AV_WN32(dst , AV_RN32(src ));
741 AV_WN32(dst+4 , AV_RN32(src+4 ));
748 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
753 AV_WN32(dst , AV_RN32(src ));
754 AV_WN32(dst+4 , AV_RN32(src+4 ));
755 AV_WN32(dst+8 , AV_RN32(src+8 ));
756 AV_WN32(dst+12, AV_RN32(src+12));
762 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
767 AV_WN32(dst , AV_RN32(src ));
768 AV_WN32(dst+4 , AV_RN32(src+4 ));
769 AV_WN32(dst+8 , AV_RN32(src+8 ));
770 AV_WN32(dst+12, AV_RN32(src+12));
777 #endif /* AVCODEC_DSPUTIL_H */