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 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);
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);
52 void ff_fdct_mmx(DCTELEM *block);
53 void ff_fdct_mmx2(DCTELEM *block);
54 void ff_fdct_sse2(DCTELEM *block);
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]);
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_chroma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qmul);
70 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
72 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
73 const float *win, float add_bias, int len);
74 void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
75 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
78 extern const uint8_t ff_alternate_horizontal_scan[64];
79 extern const uint8_t ff_alternate_vertical_scan[64];
80 extern const uint8_t ff_zigzag_direct[64];
81 extern const uint8_t ff_zigzag248_direct[64];
83 /* pixel operations */
84 #define MAX_NEG_CROP 1024
87 extern uint32_t ff_squareTbl[512];
88 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
90 void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
91 void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
92 void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
93 void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
95 /* VP3 DSP functions */
96 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
97 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
98 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
99 void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
101 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
102 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
105 void ff_bink_idct_c (DCTELEM *block);
106 void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
107 void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
110 void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
112 /* 1/2^n downscaling functions from imgconvert.c */
113 #if LIBAVCODEC_VERSION_MAJOR < 53
115 * @deprecated Use av_image_copy_plane() instead.
118 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
121 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
122 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
123 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
125 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
126 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
128 /* minimum alignment rules ;)
129 If you notice errors in the align stuff, need more alignment for some ASM code
130 for some CPU or need to use a function with less aligned data then send a mail
131 to the ffmpeg-devel mailing list, ...
133 !warning These alignments might not match reality, (missing attribute((align))
134 stuff somewhere possible).
135 I (Michael) did not check them, these are just the alignments which I think
136 could be reached easily ...
138 !future video codecs might need functions with less strict alignment
142 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
143 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
144 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
145 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
146 void clear_blocks_c(DCTELEM *blocks);
149 /* add and put pixel (decoding) */
150 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
151 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
152 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
153 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);
154 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
155 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
157 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
159 #define DEF_OLD_QPEL(name)\
160 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
161 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
162 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
164 DEF_OLD_QPEL(qpel16_mc11_old_c)
165 DEF_OLD_QPEL(qpel16_mc31_old_c)
166 DEF_OLD_QPEL(qpel16_mc12_old_c)
167 DEF_OLD_QPEL(qpel16_mc32_old_c)
168 DEF_OLD_QPEL(qpel16_mc13_old_c)
169 DEF_OLD_QPEL(qpel16_mc33_old_c)
170 DEF_OLD_QPEL(qpel8_mc11_old_c)
171 DEF_OLD_QPEL(qpel8_mc31_old_c)
172 DEF_OLD_QPEL(qpel8_mc12_old_c)
173 DEF_OLD_QPEL(qpel8_mc32_old_c)
174 DEF_OLD_QPEL(qpel8_mc13_old_c)
175 DEF_OLD_QPEL(qpel8_mc33_old_c)
177 #define CALL_2X_PIXELS(a, b, n)\
178 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
179 b(block , pixels , line_size, h);\
180 b(block+n, pixels+n, line_size, h);\
183 /* motion estimation */
184 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
185 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
186 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))*/;
191 typedef struct ScanTable{
192 const uint8_t *scantable;
193 uint8_t permutated[64];
194 uint8_t raster_end[64];
196 /** Used by dct_quantize_altivec to find last-non-zero */
197 DECLARE_ALIGNED(16, uint8_t, inverse)[64];
201 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
203 void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
204 int block_w, int block_h,
205 int src_x, int src_y, int w, int h);
210 typedef struct DSPContext {
211 /* pixel ops : interface with DCT */
212 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
213 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
214 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
215 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
216 void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
217 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
218 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
219 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
220 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
222 * translational global motion compensation.
224 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
226 * global motion compensation.
228 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
229 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
230 void (*clear_block)(DCTELEM *block/*align 16*/);
231 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
232 int (*pix_sum)(uint8_t * pix, int line_size);
233 int (*pix_norm1)(uint8_t * pix, int line_size);
234 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
236 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
238 me_cmp_func hadamard8_diff[6];
239 me_cmp_func dct_sad[6];
240 me_cmp_func quant_psnr[6];
248 me_cmp_func dct_max[6];
249 me_cmp_func dct264_sad[6];
251 me_cmp_func me_pre_cmp[6];
252 me_cmp_func me_cmp[6];
253 me_cmp_func me_sub_cmp[6];
254 me_cmp_func mb_cmp[6];
255 me_cmp_func ildct_cmp[6]; //only width 16 used
256 me_cmp_func frame_skip_cmp[6]; //only width 8 used
258 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
262 * Halfpel motion compensation with rounding (a+b+1)>>1.
263 * this is an array[4][4] of motion compensation functions for 4
264 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
265 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
266 * @param block destination where the result is stored
267 * @param pixels source
268 * @param line_size number of bytes in a horizontal line of block
271 op_pixels_func put_pixels_tab[4][4];
274 * Halfpel motion compensation with rounding (a+b+1)>>1.
275 * This is an array[4][4] of motion compensation functions for 4
276 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
277 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
278 * @param block destination into which the result is averaged (a+b+1)>>1
279 * @param pixels source
280 * @param line_size number of bytes in a horizontal line of block
283 op_pixels_func avg_pixels_tab[4][4];
286 * Halfpel motion compensation with no rounding (a+b)>>1.
287 * this is an array[2][4] of motion compensation functions for 2
288 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
289 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
290 * @param block destination where the result is stored
291 * @param pixels source
292 * @param line_size number of bytes in a horizontal line of block
295 op_pixels_func put_no_rnd_pixels_tab[4][4];
298 * Halfpel motion compensation with no rounding (a+b)>>1.
299 * this is an array[2][4] of motion compensation functions for 2
300 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
301 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
302 * @param block destination into which the result is averaged (a+b)>>1
303 * @param pixels source
304 * @param line_size number of bytes in a horizontal line of block
307 op_pixels_func avg_no_rnd_pixels_tab[4][4];
309 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);
312 * Thirdpel motion compensation with rounding (a+b+1)>>1.
313 * this is an array[12] of motion compensation functions for the 9 thirdpe
315 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
316 * @param block destination where the result is stored
317 * @param pixels source
318 * @param line_size number of bytes in a horizontal line of block
321 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
322 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
324 qpel_mc_func put_qpel_pixels_tab[2][16];
325 qpel_mc_func avg_qpel_pixels_tab[2][16];
326 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
327 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
328 qpel_mc_func put_mspel_pixels_tab[8];
333 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
334 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
335 /* This is really one func used in VC-1 decoding */
336 h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
337 h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
339 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
340 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
342 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
343 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
345 me_cmp_func pix_abs[2][4];
347 /* huffyuv specific */
348 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
349 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
350 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
352 * subtract huffyuv's variant of median prediction
353 * note, this might read from src1[-1], src2[-1]
355 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
356 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
357 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
358 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
359 /* this might write to dst[w] */
360 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
361 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
363 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
364 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
366 void (*h261_loop_filter)(uint8_t *src, int stride);
368 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
369 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
371 void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
372 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
373 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
375 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
376 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
377 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
378 /* no alignment needed */
379 void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
380 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
381 void (*vector_fmul)(float *dst, const float *src, int len);
382 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
383 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
384 void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
385 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
386 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
387 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
388 void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
389 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
391 * Multiply a vector of floats by a scalar float. Source and
392 * destination vectors must overlap exactly or not at all.
393 * @param dst result vector, 16-byte aligned
394 * @param src input vector, 16-byte aligned
395 * @param mul scalar value
396 * @param len length of vector, multiple of 4
398 void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
401 * Multiply a vector of floats by concatenated short vectors of
402 * floats and by a scalar float. Source and destination vectors
403 * must overlap exactly or not at all.
404 * [0]: short vectors of length 2, 8-byte aligned
405 * [1]: short vectors of length 4, 16-byte aligned
406 * @param dst output vector, 16-byte aligned
407 * @param src input vector, 16-byte aligned
408 * @param sv array of pointers to short vectors
409 * @param mul scalar value
410 * @param len number of elements in src and dst, multiple of 4
412 void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
413 const float **sv, float mul, int len);
415 * Multiply short vectors of floats by a scalar float, store
416 * concatenated result.
417 * [0]: short vectors of length 2, 8-byte aligned
418 * [1]: short vectors of length 4, 16-byte aligned
419 * @param dst output vector, 16-byte aligned
420 * @param sv array of pointers to short vectors
421 * @param mul scalar value
422 * @param len number of output elements, multiple of 4
424 void (*sv_fmul_scalar[2])(float *dst, const float **sv,
427 * Calculate the scalar product of two vectors of floats.
428 * @param v1 first vector, 16-byte aligned
429 * @param v2 second vector, 16-byte aligned
430 * @param len length of vectors, multiple of 4
432 float (*scalarproduct_float)(const float *v1, const float *v2, int len);
434 * Calculate the sum and difference of two vectors of floats.
435 * @param v1 first input vector, sum output, 16-byte aligned
436 * @param v2 second input vector, difference output, 16-byte aligned
437 * @param len length of vectors, multiple of 4
439 void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
441 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
442 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
443 void (*float_to_int16)(int16_t *dst, const float *src, long len);
444 void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
447 void (*fdct)(DCTELEM *block/* align 16*/);
448 void (*fdct248)(DCTELEM *block/* align 16*/);
451 void (*idct)(DCTELEM *block/* align 16*/);
454 * block -> idct -> clip to unsigned 8 bit -> dest.
455 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
456 * @param line_size size in bytes of a horizontal line of dest
458 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
461 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
462 * @param line_size size in bytes of a horizontal line of dest
464 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
467 * idct input permutation.
468 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
470 * this permutation must be performed before the idct_put/add, note, normally this can be merged
471 * with the zigzag/alternate scan<br>
472 * an example to avoid confusion:
473 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
474 * - (x -> referece dct -> reference idct -> x)
475 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
476 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
478 uint8_t idct_permutation[64];
479 int idct_permutation_type;
480 #define FF_NO_IDCT_PERM 1
481 #define FF_LIBMPEG2_IDCT_PERM 2
482 #define FF_SIMPLE_IDCT_PERM 3
483 #define FF_TRANSPOSE_IDCT_PERM 4
484 #define FF_PARTTRANS_IDCT_PERM 5
485 #define FF_SSE2_IDCT_PERM 6
487 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
488 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
489 #define BASIS_SHIFT 16
490 #define RECON_SHIFT 6
492 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
493 #define EDGE_WIDTH 16
495 void (*prefetch)(void *mem, int stride, int h);
497 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
499 /* mlp/truehd functions */
500 void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
501 int firorder, int iirorder,
502 unsigned int filter_shift, int32_t mask, int blocksize,
503 int32_t *sample_buffer);
506 void (*vc1_inv_trans_8x8)(DCTELEM *b);
507 void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
508 void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
509 void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
510 void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
511 void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
512 void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
513 void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
514 void (*vc1_v_overlap)(uint8_t* src, int stride);
515 void (*vc1_h_overlap)(uint8_t* src, int stride);
516 void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
517 void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
518 void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
519 void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
520 void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
521 void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
522 /* put 8x8 block with bicubic interpolation and quarterpel precision
523 * last argument is actually round value instead of height
525 op_pixels_func put_vc1_mspel_pixels_tab[16];
526 op_pixels_func avg_vc1_mspel_pixels_tab[16];
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
536 * @param shift number of bits to discard from product
538 int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
541 * Calculate scalar product of v1 and v2,
542 * and v1[i] += v3[i] * mul
543 * @param len length of vectors, should be multiple of 16
545 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
548 qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
549 qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
552 qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
553 qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
554 h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
555 h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
558 op_fill_func fill_block_tab[2];
559 void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
562 void dsputil_static_init(void);
563 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
565 int ff_check_alignment(void);
568 * permute block according to permuatation.
569 * @param last last non zero element in scantable order
571 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
573 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
575 #define BYTE_VEC32(c) ((c)*0x01010101UL)
577 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
579 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
582 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
584 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
587 static inline int get_penalty_factor(int lambda, int lambda2, int type){
591 return lambda>>FF_LAMBDA_SHIFT;
593 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
595 return (4*lambda)>>(FF_LAMBDA_SHIFT);
597 return (2*lambda)>>(FF_LAMBDA_SHIFT);
600 return (2*lambda)>>FF_LAMBDA_SHIFT;
605 return lambda2>>FF_LAMBDA_SHIFT;
613 * this must be called between any dsp function and float/double code.
614 * for example sin(); dsp->idct_put(); emms_c(); cos()
618 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
619 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
620 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
621 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
622 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
623 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
624 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
625 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
626 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
628 void ff_dsputil_init_dwt(DSPContext *c);
629 void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
630 void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
631 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
632 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
633 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
634 void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
640 static inline void emms(void)
642 __asm__ volatile ("emms;":::"memory");
645 #define emms_c() emms()
650 # define STRIDE_ALIGN 16
655 #define STRIDE_ALIGN 16
659 #define STRIDE_ALIGN 16
664 # define STRIDE_ALIGN 8
667 #define LOCAL_ALIGNED(a, t, v, s, ...) \
668 uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)]; \
669 t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a)
671 #if HAVE_LOCAL_ALIGNED_8
672 # define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__
674 # define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__)
677 #if HAVE_LOCAL_ALIGNED_16
678 # define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__
680 # define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__)
684 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
685 int orig_linesize[3], int coded_linesize,
686 AVCodecContext *avctx);
688 #define WRAPPER8_16(name8, name16)\
689 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
690 return name8(s, dst , src , stride, h)\
691 +name8(s, dst+8 , src+8 , stride, h);\
694 #define WRAPPER8_16_SQ(name8, name16)\
695 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
697 score +=name8(s, dst , src , stride, 8);\
698 score +=name8(s, dst+8 , src+8 , stride, 8);\
702 score +=name8(s, dst , src , stride, 8);\
703 score +=name8(s, dst+8 , src+8 , stride, 8);\
709 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
714 AV_WN16(dst , AV_RN16(src ));
720 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
725 AV_WN32(dst , AV_RN32(src ));
731 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
736 AV_WN32(dst , AV_RN32(src ));
737 AV_WN32(dst+4 , AV_RN32(src+4 ));
743 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
748 AV_WN32(dst , AV_RN32(src ));
749 AV_WN32(dst+4 , AV_RN32(src+4 ));
756 static inline void copy_block16(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));
770 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
775 AV_WN32(dst , AV_RN32(src ));
776 AV_WN32(dst+4 , AV_RN32(src+4 ));
777 AV_WN32(dst+8 , AV_RN32(src+8 ));
778 AV_WN32(dst+12, AV_RN32(src+12));
785 #endif /* AVCODEC_DSPUTIL_H */