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 */ 39typedef short DCTELEM; 40 41void fdct_ifast (DCTELEM *data); 42void fdct_ifast248 (DCTELEM *data); 43void ff_jpeg_fdct_islow (DCTELEM *data); 44void ff_fdct248_islow (DCTELEM *data); 45 46void j_rev_dct (DCTELEM *data); 47void j_rev_dct4 (DCTELEM *data); 48void j_rev_dct2 (DCTELEM *data); 49void j_rev_dct1 (DCTELEM *data); 50void ff_wmv2_idct_c(DCTELEM *data); 51 52void ff_fdct_mmx(DCTELEM *block); 53void ff_fdct_mmx2(DCTELEM *block); 54void ff_fdct_sse2(DCTELEM *block); 55 56void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride); 57void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride); 58void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); 59void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); 60void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block); 61void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block); 62void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); 63void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); 64void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); 65void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); 66 67void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1, 68 const float *win, float add_bias, int len); 69void ff_float_to_int16_c(int16_t *dst, const float *src, long len); 70void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels); 71 72/* encoding scans */ 73extern const uint8_t ff_alternate_horizontal_scan[64]; 74extern const uint8_t ff_alternate_vertical_scan[64]; 75extern const uint8_t ff_zigzag_direct[64]; 76extern const uint8_t ff_zigzag248_direct[64]; 77 78/* pixel operations */ 79#define MAX_NEG_CROP 1024 80 81/* temporary */ 82extern uint32_t ff_squareTbl[512]; 83extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP]; 84 85/* VP3 DSP functions */ 86void ff_vp3_idct_c(DCTELEM *block/* align 16*/); 87void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); 88void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); 89void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/); 90 91void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values); 92void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values); 93 94/* VP6 DSP functions */ 95void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride, 96 const int16_t *h_weights, const int16_t *v_weights); 97 98/* Bink functions */ 99void ff_bink_idct_c (DCTELEM *block); 100void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block); 101void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block); 102 103/* CAVS functions */ 104void ff_put_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride); 105void ff_avg_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride); 106void ff_put_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride); 107void ff_avg_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride); 108 109/* VC1 functions */ 110void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd); 111void ff_avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd); 112 113/* EA functions */ 114void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block); 115 116/* 1/2^n downscaling functions from imgconvert.c */ 117void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); 118void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); 119void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); 120void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); 121 122void 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 ;) 126If you notice errors in the align stuff, need more alignment for some ASM code 127for some CPU or need to use a function with less aligned data then send a mail 128to the ffmpeg-devel mailing list, ... 129 130!warning These alignments might not match reality, (missing attribute((align)) 131stuff somewhere possible). 132I (Michael) did not check them, these are just the alignments which I think 133could be reached easily ... 134 135!future video codecs might need functions with less strict alignment 136*/ 137 138/* 139void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); 140void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); 141void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); 142void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); 143void 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 then 4 149typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h); 150typedef 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); 151typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); 152typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); 153 154typedef 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)\ 157void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ 158void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ 159void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); 160 161DEF_OLD_QPEL(qpel16_mc11_old_c) 162DEF_OLD_QPEL(qpel16_mc31_old_c) 163DEF_OLD_QPEL(qpel16_mc12_old_c) 164DEF_OLD_QPEL(qpel16_mc32_old_c) 165DEF_OLD_QPEL(qpel16_mc13_old_c) 166DEF_OLD_QPEL(qpel16_mc33_old_c) 167DEF_OLD_QPEL(qpel8_mc11_old_c) 168DEF_OLD_QPEL(qpel8_mc31_old_c) 169DEF_OLD_QPEL(qpel8_mc12_old_c) 170DEF_OLD_QPEL(qpel8_mc32_old_c) 171DEF_OLD_QPEL(qpel8_mc13_old_c) 172DEF_OLD_QPEL(qpel8_mc33_old_c) 173 174#define CALL_2X_PIXELS(a, b, n)\ 175static 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 then 2 182// although currently h<4 is not used as functions with width <8 are neither used nor implemented 183typedef 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 */ 188typedef struct ScanTable{ 189 const uint8_t *scantable; 190 uint8_t permutated[64]; 191 uint8_t raster_end[64]; 192#if ARCH_PPC 193 /** Used by dct_quantize_altivec to find last-non-zero */ 194 DECLARE_ALIGNED(16, uint8_t, inverse)[64]; 195#endif 196} ScanTable; 197 198void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); 199 200void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, 201 int block_w, int block_h, 202 int src_x, int src_y, int w, int h); 203 204/** 205 * DSPContext. 206 */ 207typedef struct DSPContext { 208 /* pixel ops : interface with DCT */ 209 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); 210 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); 211 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); 212 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); 213 void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); 214 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); 215 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size); 216 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size); 217 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/); 218 /** 219 * translational global motion compensation. 220 */ 221 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); 222 /** 223 * global motion compensation. 224 */ 225 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, 226 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); 227 void (*clear_block)(DCTELEM *block/*align 16*/); 228 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); 229 int (*pix_sum)(uint8_t * pix, int line_size); 230 int (*pix_norm1)(uint8_t * pix, int line_size); 231// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 232 233 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */ 234 me_cmp_func sse[6]; 235 me_cmp_func hadamard8_diff[6]; 236 me_cmp_func dct_sad[6]; 237 me_cmp_func quant_psnr[6]; 238 me_cmp_func bit[6]; 239 me_cmp_func rd[6]; 240 me_cmp_func vsad[6]; 241 me_cmp_func vsse[6]; 242 me_cmp_func nsse[6]; 243 me_cmp_func w53[6]; 244 me_cmp_func w97[6]; 245 me_cmp_func dct_max[6]; 246 me_cmp_func dct264_sad[6]; 247 248 me_cmp_func me_pre_cmp[6]; 249 me_cmp_func me_cmp[6]; 250 me_cmp_func me_sub_cmp[6]; 251 me_cmp_func mb_cmp[6]; 252 me_cmp_func ildct_cmp[6]; //only width 16 used 253 me_cmp_func frame_skip_cmp[6]; //only width 8 used 254 255 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, 256 int size); 257 258 /** 259 * Halfpel motion compensation with rounding (a+b+1)>>1. 260 * this is an array[4][4] of motion compensation functions for 4 261 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> 262 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] 263 * @param block destination where the result is stored 264 * @param pixels source 265 * @param line_size number of bytes in a horizontal line of block 266 * @param h height 267 */ 268 op_pixels_func put_pixels_tab[4][4]; 269 270 /** 271 * Halfpel motion compensation with rounding (a+b+1)>>1. 272 * This is an array[4][4] of motion compensation functions for 4 273 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> 274 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] 275 * @param block destination into which the result is averaged (a+b+1)>>1 276 * @param pixels source 277 * @param line_size number of bytes in a horizontal line of block 278 * @param h height 279 */ 280 op_pixels_func avg_pixels_tab[4][4]; 281 282 /** 283 * Halfpel motion compensation with no rounding (a+b)>>1. 284 * this is an array[2][4] of motion compensation functions for 2 285 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> 286 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] 287 * @param block destination where the result is stored 288 * @param pixels source 289 * @param line_size number of bytes in a horizontal line of block 290 * @param h height 291 */ 292 op_pixels_func put_no_rnd_pixels_tab[4][4]; 293 294 /** 295 * Halfpel motion compensation with no rounding (a+b)>>1. 296 * this is an array[2][4] of motion compensation functions for 2 297 * horizontal blocksizes (8,16) and the 4 halfpel positions<br> 298 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] 299 * @param block destination into which the result is averaged (a+b)>>1 300 * @param pixels source 301 * @param line_size number of bytes in a horizontal line of block 302 * @param h height 303 */ 304 op_pixels_func avg_no_rnd_pixels_tab[4][4]; 305 306 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); 307 308 /** 309 * Thirdpel motion compensation with rounding (a+b+1)>>1. 310 * this is an array[12] of motion compensation functions for the 9 thirdpe 311 * positions<br> 312 * *pixels_tab[ xthirdpel + 4*ythirdpel ] 313 * @param block destination where the result is stored 314 * @param pixels source 315 * @param line_size number of bytes in a horizontal line of block 316 * @param h height 317 */ 318 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? 319 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? 320 321 qpel_mc_func put_qpel_pixels_tab[2][16]; 322 qpel_mc_func avg_qpel_pixels_tab[2][16]; 323 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; 324 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; 325 qpel_mc_func put_mspel_pixels_tab[8]; 326 327 /** 328 * h264 Chroma MC 329 */ 330 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; 331 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3]; 332 /* This is really one func used in VC-1 decoding */ 333 h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3]; 334 h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3]; 335 336 qpel_mc_func put_h264_qpel_pixels_tab[4][16]; 337 qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; 338 339 qpel_mc_func put_2tap_qpel_pixels_tab[4][16]; 340 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16]; 341 342 /* AVS specific */ 343 qpel_mc_func put_cavs_qpel_pixels_tab[2][16]; 344 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16]; 345 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); 346 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); 347 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); 348 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2); 349 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); 350 351 me_cmp_func pix_abs[2][4]; 352 353 /* huffyuv specific */ 354 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); 355 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w); 356 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); 357 /** 358 * subtract huffyuv's variant of median prediction 359 * note, this might read from src1[-1], src2[-1] 360 */ 361 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top); 362 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top); 363 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left); 364 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha); 365 /* this might write to dst[w] */ 366 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); 367 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); 368 369 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); 370 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); 371 372 void (*h261_loop_filter)(uint8_t *src, int stride); 373 374 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale); 375 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale); 376 377 void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/); 378 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values); 379 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values); 380 381 void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride, 382 const int16_t *h_weights,const int16_t *v_weights); 383 384 /* assume len is a multiple of 4, and arrays are 16-byte aligned */ 385 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); 386 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len); 387 /* no alignment needed */ 388 void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc); 389 /* assume len is a multiple of 8, and arrays are 16-byte aligned */ 390 void (*vector_fmul)(float *dst, const float *src, int len); 391 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len); 392 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */ 393 void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len); 394 /* assume len is a multiple of 4, and arrays are 16-byte aligned */ 395 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len); 396 /* assume len is a multiple of 8, and arrays are 16-byte aligned */ 397 void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len); 398 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */); 399 /** 400 * Multiply a vector of floats by a scalar float. Source and 401 * destination vectors must overlap exactly or not at all. 402 * @param dst result vector, 16-byte aligned 403 * @param src input vector, 16-byte aligned 404 * @param mul scalar value 405 * @param len length of vector, multiple of 4 406 */ 407 void (*vector_fmul_scalar)(float *dst, const float *src, float mul, 408 int len); 409 /** 410 * Multiply a vector of floats by concatenated short vectors of 411 * floats and by a scalar float. Source and destination vectors 412 * must overlap exactly or not at all. 413 * [0]: short vectors of length 2, 8-byte aligned 414 * [1]: short vectors of length 4, 16-byte aligned 415 * @param dst output vector, 16-byte aligned 416 * @param src input vector, 16-byte aligned 417 * @param sv array of pointers to short vectors 418 * @param mul scalar value 419 * @param len number of elements in src and dst, multiple of 4 420 */ 421 void (*vector_fmul_sv_scalar[2])(float *dst, const float *src, 422 const float **sv, float mul, int len); 423 /** 424 * Multiply short vectors of floats by a scalar float, store 425 * concatenated result. 426 * [0]: short vectors of length 2, 8-byte aligned 427 * [1]: short vectors of length 4, 16-byte aligned 428 * @param dst output vector, 16-byte aligned 429 * @param sv array of pointers to short vectors 430 * @param mul scalar value 431 * @param len number of output elements, multiple of 4 432 */ 433 void (*sv_fmul_scalar[2])(float *dst, const float **sv, 434 float mul, int len); 435 /** 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 4 440 */ 441 float (*scalarproduct_float)(const float *v1, const float *v2, int len); 442 /** 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 4 447 */ 448 void (*butterflies_float)(float *restrict v1, float *restrict v2, int len); 449 450 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767] 451 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */ 452 void (*float_to_int16)(int16_t *dst, const float *src, long len); 453 void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels); 454 455 /* (I)DCT */ 456 void (*fdct)(DCTELEM *block/* align 16*/); 457 void (*fdct248)(DCTELEM *block/* align 16*/); 458 459 /* IDCT really*/ 460 void (*idct)(DCTELEM *block/* align 16*/); 461 462 /** 463 * block -> idct -> clip to unsigned 8 bit -> dest. 464 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) 465 * @param line_size size in bytes of a horizontal line of dest 466 */ 467 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); 468 469 /** 470 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. 471 * @param line_size size in bytes of a horizontal line of dest 472 */ 473 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); 474 475 /** 476 * idct input permutation. 477 * several optimized IDCTs need a permutated input (relative to the normal order of the reference 478 * IDCT) 479 * this permutation must be performed before the idct_put/add, note, normally this can be merged 480 * with the zigzag/alternate scan<br> 481 * an example to avoid confusion: 482 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) 483 * - (x -> referece dct -> reference idct -> x) 484 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) 485 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) 486 */ 487 uint8_t idct_permutation[64]; 488 int idct_permutation_type; 489#define FF_NO_IDCT_PERM 1 490#define FF_LIBMPEG2_IDCT_PERM 2 491#define FF_SIMPLE_IDCT_PERM 3 492#define FF_TRANSPOSE_IDCT_PERM 4 493#define FF_PARTTRANS_IDCT_PERM 5 494#define FF_SSE2_IDCT_PERM 6 495 496 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); 497 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); 498#define BASIS_SHIFT 16 499#define RECON_SHIFT 6 500 501 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w); 502#define EDGE_WIDTH 16 503 504 void (*prefetch)(void *mem, int stride, int h); 505 506 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); 507 508 /* mlp/truehd functions */ 509 void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff, 510 int firorder, int iirorder, 511 unsigned int filter_shift, int32_t mask, int blocksize, 512 int32_t *sample_buffer); 513 514 /* vc1 functions */ 515 void (*vc1_inv_trans_8x8)(DCTELEM *b); 516 void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block); 517 void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block); 518 void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block); 519 void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block); 520 void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block); 521 void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block); 522 void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block); 523 void (*vc1_v_overlap)(uint8_t* src, int stride); 524 void (*vc1_h_overlap)(uint8_t* src, int stride); 525 void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq); 526 void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq); 527 void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq); 528 void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq); 529 void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq); 530 void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq); 531 /* put 8x8 block with bicubic interpolation and quarterpel precision 532 * last argument is actually round value instead of height 533 */ 534 op_pixels_func put_vc1_mspel_pixels_tab[16]; 535 op_pixels_func avg_vc1_mspel_pixels_tab[16]; 536 537 /* intrax8 functions */ 538 void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize); 539 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize, 540 int * range, int * sum, int edges); 541 542 /** 543 * Calculate scalar product of two vectors. 544 * @param len length of vectors, should be multiple of 16 545 * @param shift number of bits to discard from product 546 */ 547 int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift); 548 /* ape functions */ 549 /** 550 * Calculate scalar product of v1 and v2, 551 * and v1[i] += v3[i] * mul 552 * @param len length of vectors, should be multiple of 16 553 */ 554 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, int16_t *v2, int16_t *v3, int len, int mul); 555 556 /* rv30 functions */ 557 qpel_mc_func put_rv30_tpel_pixels_tab[4][16]; 558 qpel_mc_func avg_rv30_tpel_pixels_tab[4][16]; 559 560 /* rv40 functions */ 561 qpel_mc_func put_rv40_qpel_pixels_tab[4][16]; 562 qpel_mc_func avg_rv40_qpel_pixels_tab[4][16]; 563 h264_chroma_mc_func put_rv40_chroma_pixels_tab[3]; 564 h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3]; 565 566 /* bink functions */ 567 op_fill_func fill_block_tab[2]; 568 void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize); 569} DSPContext; 570 571void dsputil_static_init(void); 572void dsputil_init(DSPContext* p, AVCodecContext *avctx); 573 574int ff_check_alignment(void); 575 576/** 577 * permute block according to permuatation. 578 * @param last last non zero element in scantable order 579 */ 580void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); 581 582void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); 583 584#define BYTE_VEC32(c) ((c)*0x01010101UL) 585 586static inline uint32_t rnd_avg32(uint32_t a, uint32_t b) 587{ 588 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); 589} 590 591static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b) 592{ 593 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1); 594} 595 596static inline int get_penalty_factor(int lambda, int lambda2, int type){ 597 switch(type&0xFF){ 598 default: 599 case FF_CMP_SAD: 600 return lambda>>FF_LAMBDA_SHIFT; 601 case FF_CMP_DCT: 602 return (3*lambda)>>(FF_LAMBDA_SHIFT+1); 603 case FF_CMP_W53: 604 return (4*lambda)>>(FF_LAMBDA_SHIFT); 605 case FF_CMP_W97: 606 return (2*lambda)>>(FF_LAMBDA_SHIFT); 607 case FF_CMP_SATD: 608 case FF_CMP_DCT264: 609 return (2*lambda)>>FF_LAMBDA_SHIFT; 610 case FF_CMP_RD: 611 case FF_CMP_PSNR: 612 case FF_CMP_SSE: 613 case FF_CMP_NSSE: 614 return lambda2>>FF_LAMBDA_SHIFT; 615 case FF_CMP_BIT: 616 return 1; 617 } 618} 619 620/** 621 * Empty mmx state. 622 * this must be called between any dsp function and float/double code. 623 * for example sin(); dsp->idct_put(); emms_c(); cos() 624 */ 625#define emms_c() 626 627/* should be defined by architectures supporting 628 one or more MultiMedia extension */ 629int mm_support(void); 630extern int mm_flags; 631 632void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); 633void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx); 634void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx); 635void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); 636void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); 637void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); 638void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); 639void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx); 640void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); 641 642void ff_dsputil_init_dwt(DSPContext *c); 643void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx); 644void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx); 645void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx); 646void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx); 647void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx); 648void ff_mlp_init(DSPContext* c, AVCodecContext *avctx); 649void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx); 650 651#if HAVE_MMX 652 653#undef emms_c 654 655static inline void emms(void) 656{ 657 __asm__ volatile ("emms;":::"memory"); 658} 659 660 661#define emms_c() \ 662{\ 663 if (mm_flags & FF_MM_MMX)\ 664 emms();\ 665} 666 667#elif ARCH_ARM 668 669#if HAVE_NEON 670# define STRIDE_ALIGN 16 671#endif 672 673#elif ARCH_PPC 674 675#define STRIDE_ALIGN 16 676 677#elif HAVE_MMI 678 679#define STRIDE_ALIGN 16 680 681#else 682 683#define mm_flags 0 684#define mm_support() 0 685 686#endif 687 688#ifndef STRIDE_ALIGN 689# define STRIDE_ALIGN 8 690#endif 691 692#define LOCAL_ALIGNED(a, t, v, s, ...) \ 693 uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)]; \ 694 t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a) 695 696#if HAVE_LOCAL_ALIGNED_8 697# define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__ 698#else 699# define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__) 700#endif 701 702#if HAVE_LOCAL_ALIGNED_16 703# define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__ 704#else 705# define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__) 706#endif 707 708/* PSNR */ 709void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3], 710 int orig_linesize[3], int coded_linesize, 711 AVCodecContext *avctx); 712 713#define WRAPPER8_16(name8, name16)\ 714static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ 715 return name8(s, dst , src , stride, h)\ 716 +name8(s, dst+8 , src+8 , stride, h);\ 717} 718 719#define WRAPPER8_16_SQ(name8, name16)\ 720static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\ 721 int score=0;\ 722 score +=name8(s, dst , src , stride, 8);\ 723 score +=name8(s, dst+8 , src+8 , stride, 8);\ 724 if(h==16){\ 725 dst += 8*stride;\ 726 src += 8*stride;\ 727 score +=name8(s, dst , src , stride, 8);\ 728 score +=name8(s, dst+8 , src+8 , stride, 8);\ 729 }\ 730 return score;\ 731} 732 733 734static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) 735{ 736 int i; 737 for(i=0; i<h; i++) 738 { 739 AV_WN16(dst , AV_RN16(src )); 740 dst+=dstStride; 741 src+=srcStride; 742 } 743} 744 745static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) 746{ 747 int i; 748 for(i=0; i<h; i++) 749 { 750 AV_WN32(dst , AV_RN32(src )); 751 dst+=dstStride; 752 src+=srcStride; 753 } 754} 755 756static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) 757{ 758 int i; 759 for(i=0; i<h; i++) 760 { 761 AV_WN32(dst , AV_RN32(src )); 762 AV_WN32(dst+4 , AV_RN32(src+4 )); 763 dst+=dstStride; 764 src+=srcStride; 765 } 766} 767 768static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) 769{ 770 int i; 771 for(i=0; i<h; i++) 772 { 773 AV_WN32(dst , AV_RN32(src )); 774 AV_WN32(dst+4 , AV_RN32(src+4 )); 775 dst[8]= src[8]; 776 dst+=dstStride; 777 src+=srcStride; 778 } 779} 780 781static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) 782{ 783 int i; 784 for(i=0; i<h; i++) 785 { 786 AV_WN32(dst , AV_RN32(src )); 787 AV_WN32(dst+4 , AV_RN32(src+4 )); 788 AV_WN32(dst+8 , AV_RN32(src+8 )); 789 AV_WN32(dst+12, AV_RN32(src+12)); 790 dst+=dstStride; 791 src+=srcStride; 792 } 793} 794 795static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) 796{ 797 int i; 798 for(i=0; i<h; i++) 799 { 800 AV_WN32(dst , AV_RN32(src )); 801 AV_WN32(dst+4 , AV_RN32(src+4 )); 802 AV_WN32(dst+8 , AV_RN32(src+8 )); 803 AV_WN32(dst+12, AV_RN32(src+12)); 804 dst[16]= src[16]; 805 dst+=dstStride; 806 src+=srcStride; 807 } 808} 809 810#endif /* AVCODEC_DSPUTIL_H */ 811