libavcodec/dsputil.h

   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 dsputil.h
  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 FFMPEG_DSPUTIL_H
  31 #define FFMPEG_DSPUTIL_H
  32
  33 #include "avcodec.h"
  34
  35
  36 //#define DEBUG
  37 /* dct code */
  38 typedef short DCTELEM;
  39 typedef int DWTELEM;
  40 typedef short IDWTELEM;
  41
  42 void fdct_ifast (DCTELEM *data);
  43 void fdct_ifast248 (DCTELEM *data);
  44 void ff_jpeg_fdct_islow (DCTELEM *data);
  45 void ff_fdct248_islow (DCTELEM *data);
  46
  47 void j_rev_dct (DCTELEM *data);
  48 void j_rev_dct4 (DCTELEM *data);
  49 void j_rev_dct2 (DCTELEM *data);
  50 void j_rev_dct1 (DCTELEM *data);
  51 void ff_wmv2_idct_c(DCTELEM *data);
  52
  53 void ff_fdct_mmx(DCTELEM *block);
  54 void ff_fdct_mmx2(DCTELEM *block);
  55 void ff_fdct_sse2(DCTELEM *block);
  56
  57 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
  58 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
  59 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
  60 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
  61 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
  62 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
  63
  64 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
  65                               const float *src2, int src3, int blocksize, int step);
  66 void ff_float_to_int16_c(int16_t *dst, const float *src, int len);
  67
  68 /* encoding scans */
  69 extern const uint8_t ff_alternate_horizontal_scan[64];
  70 extern const uint8_t ff_alternate_vertical_scan[64];
  71 extern const uint8_t ff_zigzag_direct[64];
  72 extern const uint8_t ff_zigzag248_direct[64];
  73
  74 /* pixel operations */
  75 #define MAX_NEG_CROP 1024
  76
  77 /* temporary */
  78 extern uint32_t ff_squareTbl[512];
  79 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
  80
  81 /* VP3 DSP functions */
  82 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
  83 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  84 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  85
  86 /* 1/2^n downscaling functions from imgconvert.c */
  87 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  88 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  89 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  90 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  91
  92 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
  93               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
  94
  95 /* minimum alignment rules ;)
  96 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
  97 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
  98
  99 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
 100 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
 101
 102 !future video codecs might need functions with less strict alignment
 103 */
 104
 105 /*
 106 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
 107 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
 108 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 109 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 110 void clear_blocks_c(DCTELEM *blocks);
 111 */
 112
 113 /* add and put pixel (decoding) */
 114 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 115 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
 116 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
 117 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);
 118 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 119 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 120 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
 121 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
 122
 123 #define DEF_OLD_QPEL(name)\
 124 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 125 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 126 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 127
 128 DEF_OLD_QPEL(qpel16_mc11_old_c)
 129 DEF_OLD_QPEL(qpel16_mc31_old_c)
 130 DEF_OLD_QPEL(qpel16_mc12_old_c)
 131 DEF_OLD_QPEL(qpel16_mc32_old_c)
 132 DEF_OLD_QPEL(qpel16_mc13_old_c)
 133 DEF_OLD_QPEL(qpel16_mc33_old_c)
 134 DEF_OLD_QPEL(qpel8_mc11_old_c)
 135 DEF_OLD_QPEL(qpel8_mc31_old_c)
 136 DEF_OLD_QPEL(qpel8_mc12_old_c)
 137 DEF_OLD_QPEL(qpel8_mc32_old_c)
 138 DEF_OLD_QPEL(qpel8_mc13_old_c)
 139 DEF_OLD_QPEL(qpel8_mc33_old_c)
 140
 141 #define CALL_2X_PIXELS(a, b, n)\
 142 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 143     b(block  , pixels  , line_size, h);\
 144     b(block+n, pixels+n, line_size, h);\
 145 }
 146
 147 /* motion estimation */
 148 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 149 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
 150 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))*/;
 151
 152
 153 // for snow slices
 154 typedef struct slice_buffer_s slice_buffer;
 155
 156 /**
 157  * DSPContext.
 158  */
 159 typedef struct DSPContext {
 160     /* pixel ops : interface with DCT */
 161     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 162     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 163     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 164     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 165     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 166     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
 167     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
 168     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
 169     /**
 170      * translational global motion compensation.
 171      */
 172     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 173     /**
 174      * global motion compensation.
 175      */
 176     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 177                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 178     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 179     int (*pix_sum)(uint8_t * pix, int line_size);
 180     int (*pix_norm1)(uint8_t * pix, int line_size);
 181 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 182
 183     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
 184     me_cmp_func sse[5];
 185     me_cmp_func hadamard8_diff[5];
 186     me_cmp_func dct_sad[5];
 187     me_cmp_func quant_psnr[5];
 188     me_cmp_func bit[5];
 189     me_cmp_func rd[5];
 190     me_cmp_func vsad[5];
 191     me_cmp_func vsse[5];
 192     me_cmp_func nsse[5];
 193     me_cmp_func w53[5];
 194     me_cmp_func w97[5];
 195     me_cmp_func dct_max[5];
 196     me_cmp_func dct264_sad[5];
 197
 198     me_cmp_func me_pre_cmp[5];
 199     me_cmp_func me_cmp[5];
 200     me_cmp_func me_sub_cmp[5];
 201     me_cmp_func mb_cmp[5];
 202     me_cmp_func ildct_cmp[5]; //only width 16 used
 203     me_cmp_func frame_skip_cmp[5]; //only width 8 used
 204
 205     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
 206                              int size);
 207
 208     /**
 209      * Halfpel motion compensation with rounding (a+b+1)>>1.
 210      * this is an array[4][4] of motion compensation functions for 4
 211      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 212      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 213      * @param block destination where the result is stored
 214      * @param pixels source
 215      * @param line_size number of bytes in a horizontal line of block
 216      * @param h height
 217      */
 218     op_pixels_func put_pixels_tab[4][4];
 219
 220     /**
 221      * Halfpel motion compensation with rounding (a+b+1)>>1.
 222      * This is an array[4][4] of motion compensation functions for 4
 223      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 224      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 225      * @param block destination into which the result is averaged (a+b+1)>>1
 226      * @param pixels source
 227      * @param line_size number of bytes in a horizontal line of block
 228      * @param h height
 229      */
 230     op_pixels_func avg_pixels_tab[4][4];
 231
 232     /**
 233      * Halfpel motion compensation with no rounding (a+b)>>1.
 234      * this is an array[2][4] of motion compensation functions for 2
 235      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 236      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 237      * @param block destination where the result is stored
 238      * @param pixels source
 239      * @param line_size number of bytes in a horizontal line of block
 240      * @param h height
 241      */
 242     op_pixels_func put_no_rnd_pixels_tab[4][4];
 243
 244     /**
 245      * Halfpel motion compensation with no rounding (a+b)>>1.
 246      * this is an array[2][4] of motion compensation functions for 2
 247      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 248      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 249      * @param block destination into which the result is averaged (a+b)>>1
 250      * @param pixels source
 251      * @param line_size number of bytes in a horizontal line of block
 252      * @param h height
 253      */
 254     op_pixels_func avg_no_rnd_pixels_tab[4][4];
 255
 256     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);
 257
 258     /**
 259      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 260      * this is an array[12] of motion compensation functions for the 9 thirdpe
 261      * positions<br>
 262      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 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     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 269     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 270
 271     qpel_mc_func put_qpel_pixels_tab[2][16];
 272     qpel_mc_func avg_qpel_pixels_tab[2][16];
 273     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 274     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 275     qpel_mc_func put_mspel_pixels_tab[8];
 276
 277     /**
 278      * h264 Chroma MC
 279      */
 280     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 281     /* This is really one func used in VC-1 decoding */
 282     h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
 283     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 284
 285     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
 286     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
 287
 288     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
 289     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
 290
 291     h264_weight_func weight_h264_pixels_tab[10];
 292     h264_biweight_func biweight_h264_pixels_tab[10];
 293
 294     /* AVS specific */
 295     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
 296     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
 297     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 298     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 299     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 300     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 301     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 302
 303     me_cmp_func pix_abs[2][4];
 304
 305     /* huffyuv specific */
 306     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 307     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 308     /**
 309      * subtract huffyuv's variant of median prediction
 310      * note, this might read from src1[-1], src2[-1]
 311      */
 312     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
 313     void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
 314
 315     void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 316     void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 317     void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 318     void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 319     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 320     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 321     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
 322     void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
 323                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1);
 324
 325     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 326     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 327
 328     void (*h261_loop_filter)(uint8_t *src, int stride);
 329
 330     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
 331     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
 332
 333     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 334     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
 335     /* no alignment needed */
 336     void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
 337     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 338     void (*vector_fmul)(float *dst, const float *src, int len);
 339     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
 340     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
 341     void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
 342
 343     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
 344      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
 345     void (*float_to_int16)(int16_t *dst, const float *src, int len);
 346
 347     /* (I)DCT */
 348     void (*fdct)(DCTELEM *block/* align 16*/);
 349     void (*fdct248)(DCTELEM *block/* align 16*/);
 350
 351     /* IDCT really*/
 352     void (*idct)(DCTELEM *block/* align 16*/);
 353
 354     /**
 355      * block -> idct -> clip to unsigned 8 bit -> dest.
 356      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 357      * @param line_size size in bytes of a horizontal line of dest
 358      */
 359     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 360
 361     /**
 362      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 363      * @param line_size size in bytes of a horizontal line of dest
 364      */
 365     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 366
 367     /**
 368      * idct input permutation.
 369      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 370      * IDCT)
 371      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 372      * with the zigzag/alternate scan<br>
 373      * an example to avoid confusion:
 374      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 375      * - (x -> referece dct -> reference idct -> x)
 376      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 377      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 378      */
 379     uint8_t idct_permutation[64];
 380     int idct_permutation_type;
 381 #define FF_NO_IDCT_PERM 1
 382 #define FF_LIBMPEG2_IDCT_PERM 2
 383 #define FF_SIMPLE_IDCT_PERM 3
 384 #define FF_TRANSPOSE_IDCT_PERM 4
 385 #define FF_PARTTRANS_IDCT_PERM 5
 386
 387     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 388     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 389 #define BASIS_SHIFT 16
 390 #define RECON_SHIFT 6
 391
 392     /* h264 functions */
 393     void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
 394     void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 395     void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
 396     void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
 397     void (*h264_dct)(DCTELEM block[4][4]);
 398
 399     /* snow wavelet */
 400     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
 401     void (*horizontal_compose97i)(IDWTELEM *b, int width);
 402     void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
 403
 404     void (*prefetch)(void *mem, int stride, int h);
 405
 406     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 407
 408     /* vc1 functions */
 409     void (*vc1_inv_trans_8x8)(DCTELEM *b);
 410     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
 411     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
 412     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
 413     void (*vc1_v_overlap)(uint8_t* src, int stride);
 414     void (*vc1_h_overlap)(uint8_t* src, int stride);
 415     /* put 8x8 block with bicubic interpolation and quarterpel precision
 416      * last argument is actually round value instead of height
 417      */
 418     op_pixels_func put_vc1_mspel_pixels_tab[16];
 419
 420     /* intrax8 functions */
 421     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
 422     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
 423            int * range, int * sum,  int edges);
 424
 425 } DSPContext;
 426
 427 void dsputil_static_init(void);
 428 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 429
 430 int ff_check_alignment(void);
 431
 432 /**
 433  * permute block according to permuatation.
 434  * @param last last non zero element in scantable order
 435  */
 436 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 437
 438 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 439
 440 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 441
 442 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 443 {
 444     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 445 }
 446
 447 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 448 {
 449     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 450 }
 451
 452 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 453     switch(type&0xFF){
 454     default:
 455     case FF_CMP_SAD:
 456         return lambda>>FF_LAMBDA_SHIFT;
 457     case FF_CMP_DCT:
 458         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 459     case FF_CMP_W53:
 460         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 461     case FF_CMP_W97:
 462         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 463     case FF_CMP_SATD:
 464     case FF_CMP_DCT264:
 465         return (2*lambda)>>FF_LAMBDA_SHIFT;
 466     case FF_CMP_RD:
 467     case FF_CMP_PSNR:
 468     case FF_CMP_SSE:
 469     case FF_CMP_NSSE:
 470         return lambda2>>FF_LAMBDA_SHIFT;
 471     case FF_CMP_BIT:
 472         return 1;
 473     }
 474 }
 475
 476 /**
 477  * Empty mmx state.
 478  * this must be called between any dsp function and float/double code.
 479  * for example sin(); dsp->idct_put(); emms_c(); cos()
 480  */
 481 #define emms_c()
 482
 483 /* should be defined by architectures supporting
 484    one or more MultiMedia extension */
 485 int mm_support(void);
 486
 487 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 488 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
 489 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
 490 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 491 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 492 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 493 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 494 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 495 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 496
 497 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
 498
 499 #if defined(HAVE_MMX)
 500
 501 #undef emms_c
 502
 503 #define MM_MMX    0x0001 /* standard MMX */
 504 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
 505 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
 506 #define MM_SSE    0x0008 /* SSE functions */
 507 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
 508 #define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
 509 #define MM_SSE3   0x0040 /* Prescott SSE3 functions */
 510 #define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
 511
 512 extern int mm_flags;
 513
 514 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 515 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 516 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 517
 518 static inline void emms(void)
 519 {
 520     __asm __volatile ("emms;":::"memory");
 521 }
 522
 523
 524 #define emms_c() \
 525 {\
 526     if (mm_flags & MM_MMX)\
 527         emms();\
 528 }
 529
 530 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 531
 532 #define STRIDE_ALIGN 8
 533
 534 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
 535
 536 #elif defined(ARCH_ARMV4L)
 537
 538 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
 539    line optimizations */
 540 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(4, t, v)
 541 #define STRIDE_ALIGN 4
 542
 543 #define MM_IWMMXT    0x0100 /* XScale IWMMXT */
 544
 545 extern int mm_flags;
 546
 547 #elif defined(HAVE_MLIB)
 548
 549 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 550 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 551 #define STRIDE_ALIGN 8
 552
 553 #elif defined(HAVE_VIS)
 554
 555 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 556 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 557 #define STRIDE_ALIGN 8
 558
 559 #elif defined(ARCH_ALPHA)
 560
 561 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 562 #define STRIDE_ALIGN 8
 563
 564 #elif defined(ARCH_POWERPC)
 565
 566 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
 567
 568 extern int mm_flags;
 569
 570 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 571 #define STRIDE_ALIGN 16
 572
 573 #elif defined(HAVE_MMI)
 574
 575 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 576 #define STRIDE_ALIGN 16
 577
 578 #elif defined(ARCH_SH4)
 579
 580 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 581 #define STRIDE_ALIGN 8
 582
 583 #elif defined(ARCH_BFIN)
 584
 585 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 586 #define STRIDE_ALIGN 8
 587
 588 #else
 589
 590 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 591 #define STRIDE_ALIGN 8
 592
 593 #endif
 594
 595 /* PSNR */
 596 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 597               int orig_linesize[3], int coded_linesize,
 598               AVCodecContext *avctx);
 599
 600 /* FFT computation */
 601
 602 /* NOTE: soon integer code will be added, so you must use the
 603    FFTSample type */
 604 typedef float FFTSample;
 605
 606 struct MDCTContext;
 607
 608 typedef struct FFTComplex {
 609     FFTSample re, im;
 610 } FFTComplex;
 611
 612 typedef struct FFTContext {
 613     int nbits;
 614     int inverse;
 615     uint16_t *revtab;
 616     FFTComplex *exptab;
 617     FFTComplex *exptab1; /* only used by SSE code */
 618     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 619     void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
 620                        const FFTSample *input, FFTSample *tmp);
 621 } FFTContext;
 622
 623 int ff_fft_init(FFTContext *s, int nbits, int inverse);
 624 void ff_fft_permute(FFTContext *s, FFTComplex *z);
 625 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 626 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
 627 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
 628 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
 629 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
 630
 631 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
 632 {
 633     s->fft_calc(s, z);
 634 }
 635 void ff_fft_end(FFTContext *s);
 636
 637 /* MDCT computation */
 638
 639 typedef struct MDCTContext {
 640     int n;  /* size of MDCT (i.e. number of input data * 2) */
 641     int nbits; /* n = 2^nbits */
 642     /* pre/post rotation tables */
 643     FFTSample *tcos;
 644     FFTSample *tsin;
 645     FFTContext fft;
 646 } MDCTContext;
 647
 648 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
 649 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
 650                 const FFTSample *input, FFTSample *tmp);
 651 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
 652                         const FFTSample *input, FFTSample *tmp);
 653 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
 654                        const FFTSample *input, FFTSample *tmp);
 655 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
 656                const FFTSample *input, FFTSample *tmp);
 657 void ff_mdct_end(MDCTContext *s);
 658
 659 #define WARPER8_16(name8, name16)\
 660 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 661     return name8(s, dst           , src           , stride, h)\
 662           +name8(s, dst+8         , src+8         , stride, h);\
 663 }
 664
 665 #define WARPER8_16_SQ(name8, name16)\
 666 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 667     int score=0;\
 668     score +=name8(s, dst           , src           , stride, 8);\
 669     score +=name8(s, dst+8         , src+8         , stride, 8);\
 670     if(h==16){\
 671         dst += 8*stride;\
 672         src += 8*stride;\
 673         score +=name8(s, dst           , src           , stride, 8);\
 674         score +=name8(s, dst+8         , src+8         , stride, 8);\
 675     }\
 676     return score;\
 677 }
 678
 679
 680 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 681 {
 682     int i;
 683     for(i=0; i<h; i++)
 684     {
 685         AV_WN16(dst   , AV_RN16(src   ));
 686         dst+=dstStride;
 687         src+=srcStride;
 688     }
 689 }
 690
 691 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 692 {
 693     int i;
 694     for(i=0; i<h; i++)
 695     {
 696         AV_WN32(dst   , AV_RN32(src   ));
 697         dst+=dstStride;
 698         src+=srcStride;
 699     }
 700 }
 701
 702 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 703 {
 704     int i;
 705     for(i=0; i<h; i++)
 706     {
 707         AV_WN32(dst   , AV_RN32(src   ));
 708         AV_WN32(dst+4 , AV_RN32(src+4 ));
 709         dst+=dstStride;
 710         src+=srcStride;
 711     }
 712 }
 713
 714 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 715 {
 716     int i;
 717     for(i=0; i<h; i++)
 718     {
 719         AV_WN32(dst   , AV_RN32(src   ));
 720         AV_WN32(dst+4 , AV_RN32(src+4 ));
 721         dst[8]= src[8];
 722         dst+=dstStride;
 723         src+=srcStride;
 724     }
 725 }
 726
 727 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 728 {
 729     int i;
 730     for(i=0; i<h; i++)
 731     {
 732         AV_WN32(dst   , AV_RN32(src   ));
 733         AV_WN32(dst+4 , AV_RN32(src+4 ));
 734         AV_WN32(dst+8 , AV_RN32(src+8 ));
 735         AV_WN32(dst+12, AV_RN32(src+12));
 736         dst+=dstStride;
 737         src+=srcStride;
 738     }
 739 }
 740
 741 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 742 {
 743     int i;
 744     for(i=0; i<h; i++)
 745     {
 746         AV_WN32(dst   , AV_RN32(src   ));
 747         AV_WN32(dst+4 , AV_RN32(src+4 ));
 748         AV_WN32(dst+8 , AV_RN32(src+8 ));
 749         AV_WN32(dst+12, AV_RN32(src+12));
 750         dst[16]= src[16];
 751         dst+=dstStride;
 752         src+=srcStride;
 753     }
 754 }
 755
 756 #endif /* FFMPEG_DSPUTIL_H */