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);
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);
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];
87 /* pixel operations */
88 #define MAX_NEG_CROP 1024
91 extern uint32_t ff_squareTbl[512];
92 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
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);
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
111 /* VP3 DSP functions */
112 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
113 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
114 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115 void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
117 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
118 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
121 void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
124 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
125 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride);
126 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
127 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride);
129 /* 1/2^n downscaling functions from imgconvert.c */
130 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
131 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
132 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
134 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
135 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
137 /* minimum alignment rules ;)
138 If you notice errors in the align stuff, need more alignment for some ASM code
139 for some CPU or need to use a function with less aligned data then send a mail
140 to the ffmpeg-devel mailing list, ...
142 !warning These alignments might not match reality, (missing attribute((align))
143 stuff somewhere possible).
144 I (Michael) did not check them, these are just the alignments which I think
145 could be reached easily ...
147 !future video codecs might need functions with less strict alignment
151 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
152 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
153 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
154 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
155 void clear_blocks_c(DCTELEM *blocks);
158 /* add and put pixel (decoding) */
159 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
160 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4
161 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
162 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);
163 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
164 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
166 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
168 #define DEF_OLD_QPEL(name)\
169 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
170 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
171 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
173 DEF_OLD_QPEL(qpel16_mc11_old_c)
174 DEF_OLD_QPEL(qpel16_mc31_old_c)
175 DEF_OLD_QPEL(qpel16_mc12_old_c)
176 DEF_OLD_QPEL(qpel16_mc32_old_c)
177 DEF_OLD_QPEL(qpel16_mc13_old_c)
178 DEF_OLD_QPEL(qpel16_mc33_old_c)
179 DEF_OLD_QPEL(qpel8_mc11_old_c)
180 DEF_OLD_QPEL(qpel8_mc31_old_c)
181 DEF_OLD_QPEL(qpel8_mc12_old_c)
182 DEF_OLD_QPEL(qpel8_mc32_old_c)
183 DEF_OLD_QPEL(qpel8_mc13_old_c)
184 DEF_OLD_QPEL(qpel8_mc33_old_c)
186 #define CALL_2X_PIXELS(a, b, n)\
187 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
188 b(block , pixels , line_size, h);\
189 b(block+n, pixels+n, line_size, h);\
192 /* motion estimation */
193 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
194 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
195 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))*/;
200 typedef struct ScanTable{
201 const uint8_t *scantable;
202 uint8_t permutated[64];
203 uint8_t raster_end[64];
206 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
207 void ff_init_scantable_permutation(uint8_t *idct_permutation,
208 int idct_permutation_type);
210 #define EMULATED_EDGE(depth) \
211 void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, int linesize,\
212 int block_w, int block_h,\
213 int src_x, int src_y, int w, int h);
221 void ff_add_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
222 void ff_put_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
223 void ff_put_signed_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
228 typedef struct DSPContext {
230 * Size of DCT coefficients.
234 /* pixel ops : interface with DCT */
235 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
236 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
237 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
238 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
239 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
240 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
241 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
242 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
244 * Motion estimation with emulated edge values.
245 * @param buf pointer to destination buffer (unaligned)
246 * @param src pointer to pixel source (unaligned)
247 * @param linesize width (in pixels) for src/buf
248 * @param block_w number of pixels (per row) to copy to buf
249 * @param block_h nummber of pixel rows to copy to buf
250 * @param src_x offset of src to start of row - this may be negative
251 * @param src_y offset of src to top of image - this may be negative
252 * @param w width of src in pixels
253 * @param h height of src in pixels
255 void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
256 int block_w, int block_h,
257 int src_x, int src_y, int w, int h);
259 * translational global motion compensation.
261 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
263 * global motion compensation.
265 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
266 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
267 void (*clear_block)(DCTELEM *block/*align 16*/);
268 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
269 int (*pix_sum)(uint8_t * pix, int line_size);
270 int (*pix_norm1)(uint8_t * pix, int line_size);
271 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
273 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
275 me_cmp_func hadamard8_diff[6];
276 me_cmp_func dct_sad[6];
277 me_cmp_func quant_psnr[6];
285 me_cmp_func dct_max[6];
286 me_cmp_func dct264_sad[6];
288 me_cmp_func me_pre_cmp[6];
289 me_cmp_func me_cmp[6];
290 me_cmp_func me_sub_cmp[6];
291 me_cmp_func mb_cmp[6];
292 me_cmp_func ildct_cmp[6]; //only width 16 used
293 me_cmp_func frame_skip_cmp[6]; //only width 8 used
295 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
299 * Halfpel motion compensation with rounding (a+b+1)>>1.
300 * this is an array[4][4] of motion compensation functions for 4
301 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
302 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
303 * @param block destination where the result is stored
304 * @param pixels source
305 * @param line_size number of bytes in a horizontal line of block
308 op_pixels_func put_pixels_tab[4][4];
311 * Halfpel motion compensation with rounding (a+b+1)>>1.
312 * This is an array[4][4] of motion compensation functions for 4
313 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
314 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
315 * @param block destination into which the result is averaged (a+b+1)>>1
316 * @param pixels source
317 * @param line_size number of bytes in a horizontal line of block
320 op_pixels_func avg_pixels_tab[4][4];
323 * Halfpel motion compensation with no rounding (a+b)>>1.
324 * this is an array[2][4] of motion compensation functions for 2
325 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
326 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
327 * @param block destination where the result is stored
328 * @param pixels source
329 * @param line_size number of bytes in a horizontal line of block
332 op_pixels_func put_no_rnd_pixels_tab[4][4];
335 * Halfpel motion compensation with no rounding (a+b)>>1.
336 * this is an array[2][4] of motion compensation functions for 2
337 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
338 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
339 * @param block destination into which the result is averaged (a+b)>>1
340 * @param pixels source
341 * @param line_size number of bytes in a horizontal line of block
344 op_pixels_func avg_no_rnd_pixels_tab[4][4];
346 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);
349 * Thirdpel motion compensation with rounding (a+b+1)>>1.
350 * this is an array[12] of motion compensation functions for the 9 thirdpe
352 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
353 * @param block destination where the result is stored
354 * @param pixels source
355 * @param line_size number of bytes in a horizontal line of block
358 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
359 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
361 qpel_mc_func put_qpel_pixels_tab[2][16];
362 qpel_mc_func avg_qpel_pixels_tab[2][16];
363 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
364 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
365 qpel_mc_func put_mspel_pixels_tab[8];
370 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
371 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
373 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
374 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
376 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
377 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
379 me_cmp_func pix_abs[2][4];
381 /* huffyuv specific */
382 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
383 void (*diff_bytes)(uint8_t *dst/*align 16*/, const uint8_t *src1/*align 16*/, const uint8_t *src2/*align 1*/,int w);
385 * subtract huffyuv's variant of median prediction
386 * note, this might read from src1[-1], src2[-1]
388 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
389 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
390 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
391 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
392 /* this might write to dst[w] */
393 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
394 void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
396 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
397 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
399 void (*h261_loop_filter)(uint8_t *src, int stride);
401 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
402 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
404 void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
405 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
406 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
408 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
409 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
410 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
411 /* assume len is a multiple of 16, and arrays are 32-byte aligned */
412 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
413 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
414 void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
415 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
416 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
417 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
418 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
420 * Multiply a vector of floats by a scalar float. Source and
421 * destination vectors must overlap exactly or not at all.
422 * @param dst result vector, 16-byte aligned
423 * @param src input vector, 16-byte aligned
424 * @param mul scalar value
425 * @param len length of vector, multiple of 4
427 void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
430 * Calculate the scalar product of two vectors of floats.
431 * @param v1 first vector, 16-byte aligned
432 * @param v2 second vector, 16-byte aligned
433 * @param len length of vectors, multiple of 4
435 float (*scalarproduct_float)(const float *v1, const float *v2, int len);
437 * Calculate the sum and difference of two vectors of floats.
438 * @param v1 first input vector, sum output, 16-byte aligned
439 * @param v2 second input vector, difference output, 16-byte aligned
440 * @param len length of vectors, multiple of 4
442 void (*butterflies_float)(float *av_restrict v1, float *av_restrict v2, int len);
445 * Calculate the sum and difference of two vectors of floats and interleave
446 * results into a separate output vector of floats, with each sum
447 * positioned before the corresponding difference.
449 * @param dst output vector
450 * constraints: 16-byte aligned
451 * @param src0 first input vector
452 * constraints: 32-byte aligned
453 * @param src1 second input vector
454 * constraints: 32-byte aligned
455 * @param len number of elements in the input
456 * constraints: multiple of 8
458 void (*butterflies_float_interleave)(float *dst, const float *src0,
459 const float *src1, int len);
462 void (*fdct)(DCTELEM *block/* align 16*/);
463 void (*fdct248)(DCTELEM *block/* align 16*/);
466 void (*idct)(DCTELEM *block/* align 16*/);
469 * block -> idct -> clip to unsigned 8 bit -> dest.
470 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
471 * @param line_size size in bytes of a horizontal line of dest
473 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
476 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
477 * @param line_size size in bytes of a horizontal line of dest
479 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
482 * idct input permutation.
483 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
485 * this permutation must be performed before the idct_put/add, note, normally this can be merged
486 * with the zigzag/alternate scan<br>
487 * an example to avoid confusion:
488 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
489 * - (x -> reference dct -> reference idct -> x)
490 * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
491 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
493 uint8_t idct_permutation[64];
494 int idct_permutation_type;
495 #define FF_NO_IDCT_PERM 1
496 #define FF_LIBMPEG2_IDCT_PERM 2
497 #define FF_SIMPLE_IDCT_PERM 3
498 #define FF_TRANSPOSE_IDCT_PERM 4
499 #define FF_PARTTRANS_IDCT_PERM 5
500 #define FF_SSE2_IDCT_PERM 6
502 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
503 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
504 #define BASIS_SHIFT 16
505 #define RECON_SHIFT 6
507 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides);
508 #define EDGE_WIDTH 16
510 #define EDGE_BOTTOM 2
512 void (*prefetch)(void *mem, int stride, int h);
514 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
516 /* mlp/truehd functions */
517 void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
518 int firorder, int iirorder,
519 unsigned int filter_shift, int32_t mask, int blocksize,
520 int32_t *sample_buffer);
522 /* intrax8 functions */
523 void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
524 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
525 int * range, int * sum, int edges);
528 * Calculate scalar product of two vectors.
529 * @param len length of vectors, should be multiple of 16
531 int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len);
534 * Calculate scalar product of v1 and v2,
535 * and v1[i] += v3[i] * mul
536 * @param len length of vectors, should be multiple of 16
538 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
541 * Apply symmetric window in 16-bit fixed-point.
542 * @param output destination array
543 * constraints: 16-byte aligned
544 * @param input source array
545 * constraints: 16-byte aligned
546 * @param window window array
547 * constraints: 16-byte aligned, at least len/2 elements
548 * @param len full window length
549 * constraints: multiple of ? greater than zero
551 void (*apply_window_int16)(int16_t *output, const int16_t *input,
552 const int16_t *window, unsigned int len);
555 * Clip each element in an array of int32_t to a given minimum and maximum value.
556 * @param dst destination array
557 * constraints: 16-byte aligned
558 * @param src source array
559 * constraints: 16-byte aligned
560 * @param min minimum value
561 * constraints: must be in the range [-(1 << 24), 1 << 24]
562 * @param max maximum value
563 * constraints: must be in the range [-(1 << 24), 1 << 24]
564 * @param len number of elements in the array
565 * constraints: multiple of 32 greater than zero
567 void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
568 int32_t max, unsigned int len);
570 op_fill_func fill_block_tab[2];
573 void ff_dsputil_static_init(void);
574 void ff_dsputil_init(DSPContext* p, AVCodecContext *avctx);
575 attribute_deprecated void dsputil_init(DSPContext* c, AVCodecContext *avctx);
577 int ff_check_alignment(void);
580 * permute block according to permuatation.
581 * @param last last non zero element in scantable order
583 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
585 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
587 #define BYTE_VEC32(c) ((c)*0x01010101UL)
588 #define BYTE_VEC64(c) ((c)*0x0001000100010001UL)
590 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
592 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
595 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
597 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
600 static inline uint64_t rnd_avg64(uint64_t a, uint64_t b)
602 return (a | b) - (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
605 static inline uint64_t no_rnd_avg64(uint64_t a, uint64_t b)
607 return (a & b) + (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
610 static inline int get_penalty_factor(int lambda, int lambda2, int type){
614 return lambda>>FF_LAMBDA_SHIFT;
616 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
618 return (4*lambda)>>(FF_LAMBDA_SHIFT);
620 return (2*lambda)>>(FF_LAMBDA_SHIFT);
623 return (2*lambda)>>FF_LAMBDA_SHIFT;
628 return lambda2>>FF_LAMBDA_SHIFT;
634 void ff_dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
635 void ff_dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
636 void ff_dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
637 void ff_dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
638 void ff_dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
639 void ff_dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
640 void ff_dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
641 void ff_dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
643 void ff_dsputil_init_dwt(DSPContext *c);
644 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
645 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
646 void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
648 #if (ARCH_ARM && HAVE_NEON) || ARCH_PPC || HAVE_MMI || HAVE_MMX
649 # define STRIDE_ALIGN 16
651 # define STRIDE_ALIGN 8
654 #define LOCAL_ALIGNED_A(a, t, v, s, o, ...) \
655 uint8_t la_##v[sizeof(t s o) + (a)]; \
656 t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
658 #define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
660 #define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
662 #if HAVE_LOCAL_ALIGNED_8
663 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
665 # define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
668 #if HAVE_LOCAL_ALIGNED_16
669 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
671 # define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
674 #define WRAPPER8_16(name8, name16)\
675 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
676 return name8(s, dst , src , stride, h)\
677 +name8(s, dst+8 , src+8 , stride, h);\
680 #define WRAPPER8_16_SQ(name8, name16)\
681 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
683 score +=name8(s, dst , src , stride, 8);\
684 score +=name8(s, dst+8 , src+8 , stride, 8);\
688 score +=name8(s, dst , src , stride, 8);\
689 score +=name8(s, dst+8 , src+8 , stride, 8);\
695 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
700 AV_WN16(dst , AV_RN16(src ));
706 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
711 AV_WN32(dst , AV_RN32(src ));
717 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
722 AV_WN32(dst , AV_RN32(src ));
723 AV_WN32(dst+4 , AV_RN32(src+4 ));
729 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
734 AV_WN32(dst , AV_RN32(src ));
735 AV_WN32(dst+4 , AV_RN32(src+4 ));
742 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
747 AV_WN32(dst , AV_RN32(src ));
748 AV_WN32(dst+4 , AV_RN32(src+4 ));
749 AV_WN32(dst+8 , AV_RN32(src+8 ));
750 AV_WN32(dst+12, AV_RN32(src+12));
756 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
761 AV_WN32(dst , AV_RN32(src ));
762 AV_WN32(dst+4 , AV_RN32(src+4 ));
763 AV_WN32(dst+8 , AV_RN32(src+8 ));
764 AV_WN32(dst+12, AV_RN32(src+12));
771 #endif /* AVCODEC_DSPUTIL_H */