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_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
  67                              const float *win, float add_bias, int len);
  68 void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
  69
  70 /* encoding scans */
  71 extern const uint8_t ff_alternate_horizontal_scan[64];
  72 extern const uint8_t ff_alternate_vertical_scan[64];
  73 extern const uint8_t ff_zigzag_direct[64];
  74 extern const uint8_t ff_zigzag248_direct[64];
  75
  76 /* pixel operations */
  77 #define MAX_NEG_CROP 1024
  78
  79 /* temporary */
  80 extern uint32_t ff_squareTbl[512];
  81 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
  82
  83 /* VP3 DSP functions */
  84 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
  85 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  86 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  87
  88 /* 1/2^n downscaling functions from imgconvert.c */
  89 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  90 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  91 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  92 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  93
  94 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
  95               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
  96
  97 /* minimum alignment rules ;)
  98 If you notice errors in the align stuff, need more alignment for some ASM code
  99 for some CPU or need to use a function with less aligned data then send a mail
 100 to the ffmpeg-devel mailing list, ...
 101
 102 !warning These alignments might not match reality, (missing attribute((align))
 103 stuff somewhere possible).
 104 I (Michael) did not check them, these are just the alignments which I think
 105 could be reached easily ...
 106
 107 !future video codecs might need functions with less strict alignment
 108 */
 109
 110 /*
 111 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
 112 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
 113 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 114 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 115 void clear_blocks_c(DCTELEM *blocks);
 116 */
 117
 118 /* add and put pixel (decoding) */
 119 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 120 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
 121 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
 122 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);
 123 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 124 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 125 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
 126 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
 127
 128 #define DEF_OLD_QPEL(name)\
 129 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 130 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 131 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 132
 133 DEF_OLD_QPEL(qpel16_mc11_old_c)
 134 DEF_OLD_QPEL(qpel16_mc31_old_c)
 135 DEF_OLD_QPEL(qpel16_mc12_old_c)
 136 DEF_OLD_QPEL(qpel16_mc32_old_c)
 137 DEF_OLD_QPEL(qpel16_mc13_old_c)
 138 DEF_OLD_QPEL(qpel16_mc33_old_c)
 139 DEF_OLD_QPEL(qpel8_mc11_old_c)
 140 DEF_OLD_QPEL(qpel8_mc31_old_c)
 141 DEF_OLD_QPEL(qpel8_mc12_old_c)
 142 DEF_OLD_QPEL(qpel8_mc32_old_c)
 143 DEF_OLD_QPEL(qpel8_mc13_old_c)
 144 DEF_OLD_QPEL(qpel8_mc33_old_c)
 145
 146 #define CALL_2X_PIXELS(a, b, n)\
 147 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 148     b(block  , pixels  , line_size, h);\
 149     b(block+n, pixels+n, line_size, h);\
 150 }
 151
 152 /* motion estimation */
 153 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 154 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
 155 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))*/;
 156
 157
 158 // for snow slices
 159 typedef struct slice_buffer_s slice_buffer;
 160
 161 /**
 162  * Scantable.
 163  */
 164 typedef struct ScanTable{
 165     const uint8_t *scantable;
 166     uint8_t permutated[64];
 167     uint8_t raster_end[64];
 168 #ifdef ARCH_POWERPC
 169                 /** Used by dct_quantize_altivec to find last-non-zero */
 170     DECLARE_ALIGNED(16, uint8_t, inverse[64]);
 171 #endif
 172 } ScanTable;
 173
 174 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
 175
 176 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
 177                          int block_w, int block_h,
 178                          int src_x, int src_y, int w, int h);
 179
 180 /**
 181  * DSPContext.
 182  */
 183 typedef struct DSPContext {
 184     /* pixel ops : interface with DCT */
 185     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 186     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 187     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 188     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 189     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 190     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
 191     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
 192     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
 193     /**
 194      * translational global motion compensation.
 195      */
 196     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 197     /**
 198      * global motion compensation.
 199      */
 200     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 201                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 202     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 203     int (*pix_sum)(uint8_t * pix, int line_size);
 204     int (*pix_norm1)(uint8_t * pix, int line_size);
 205 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 206
 207     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
 208     me_cmp_func sse[5];
 209     me_cmp_func hadamard8_diff[5];
 210     me_cmp_func dct_sad[5];
 211     me_cmp_func quant_psnr[5];
 212     me_cmp_func bit[5];
 213     me_cmp_func rd[5];
 214     me_cmp_func vsad[5];
 215     me_cmp_func vsse[5];
 216     me_cmp_func nsse[5];
 217     me_cmp_func w53[5];
 218     me_cmp_func w97[5];
 219     me_cmp_func dct_max[5];
 220     me_cmp_func dct264_sad[5];
 221
 222     me_cmp_func me_pre_cmp[5];
 223     me_cmp_func me_cmp[5];
 224     me_cmp_func me_sub_cmp[5];
 225     me_cmp_func mb_cmp[5];
 226     me_cmp_func ildct_cmp[5]; //only width 16 used
 227     me_cmp_func frame_skip_cmp[5]; //only width 8 used
 228
 229     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
 230                              int size);
 231
 232     /**
 233      * Halfpel motion compensation with rounding (a+b+1)>>1.
 234      * this is an array[4][4] of motion compensation functions for 4
 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_pixels_tab[4][4];
 243
 244     /**
 245      * Halfpel motion compensation with rounding (a+b+1)>>1.
 246      * This is an array[4][4] of motion compensation functions for 4
 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)>>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_pixels_tab[4][4];
 255
 256     /**
 257      * Halfpel motion compensation with no rounding (a+b)>>1.
 258      * this is an array[2][4] of motion compensation functions for 2
 259      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 260      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 261      * @param block destination where the result is stored
 262      * @param pixels source
 263      * @param line_size number of bytes in a horizontal line of block
 264      * @param h height
 265      */
 266     op_pixels_func put_no_rnd_pixels_tab[4][4];
 267
 268     /**
 269      * Halfpel motion compensation with no rounding (a+b)>>1.
 270      * this is an array[2][4] of motion compensation functions for 2
 271      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 272      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 273      * @param block destination into which the result is averaged (a+b)>>1
 274      * @param pixels source
 275      * @param line_size number of bytes in a horizontal line of block
 276      * @param h height
 277      */
 278     op_pixels_func avg_no_rnd_pixels_tab[4][4];
 279
 280     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);
 281
 282     /**
 283      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 284      * this is an array[12] of motion compensation functions for the 9 thirdpe
 285      * positions<br>
 286      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 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     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 293     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 294
 295     qpel_mc_func put_qpel_pixels_tab[2][16];
 296     qpel_mc_func avg_qpel_pixels_tab[2][16];
 297     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 298     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 299     qpel_mc_func put_mspel_pixels_tab[8];
 300
 301     /**
 302      * h264 Chroma MC
 303      */
 304     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 305     /* This is really one func used in VC-1 decoding */
 306     h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
 307     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 308
 309     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
 310     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
 311
 312     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
 313     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
 314
 315     h264_weight_func weight_h264_pixels_tab[10];
 316     h264_biweight_func biweight_h264_pixels_tab[10];
 317
 318     /* AVS specific */
 319     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
 320     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
 321     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 322     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 323     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 324     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 325     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 326
 327     me_cmp_func pix_abs[2][4];
 328
 329     /* huffyuv specific */
 330     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 331     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
 332     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 333     /**
 334      * subtract huffyuv's variant of median prediction
 335      * note, this might read from src1[-1], src2[-1]
 336      */
 337     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
 338     /* this might write to dst[w] */
 339     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
 340     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
 341
 342     void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 343     void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 344     void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 345     void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 346     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 347     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 348     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
 349     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],
 350                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1);
 351
 352     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 353     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 354
 355     void (*h261_loop_filter)(uint8_t *src, int stride);
 356
 357     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
 358     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
 359
 360     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 361     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
 362     /* no alignment needed */
 363     void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
 364     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 365     void (*vector_fmul)(float *dst, const float *src, int len);
 366     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
 367     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
 368     void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
 369     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 370     void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
 371
 372     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
 373      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
 374     void (*float_to_int16)(int16_t *dst, const float *src, long len);
 375     void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
 376
 377     /* (I)DCT */
 378     void (*fdct)(DCTELEM *block/* align 16*/);
 379     void (*fdct248)(DCTELEM *block/* align 16*/);
 380
 381     /* IDCT really*/
 382     void (*idct)(DCTELEM *block/* align 16*/);
 383
 384     /**
 385      * block -> idct -> clip to unsigned 8 bit -> dest.
 386      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 387      * @param line_size size in bytes of a horizontal line of dest
 388      */
 389     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 390
 391     /**
 392      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 393      * @param line_size size in bytes of a horizontal line of dest
 394      */
 395     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 396
 397     /**
 398      * idct input permutation.
 399      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 400      * IDCT)
 401      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 402      * with the zigzag/alternate scan<br>
 403      * an example to avoid confusion:
 404      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 405      * - (x -> referece dct -> reference idct -> x)
 406      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 407      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 408      */
 409     uint8_t idct_permutation[64];
 410     int idct_permutation_type;
 411 #define FF_NO_IDCT_PERM 1
 412 #define FF_LIBMPEG2_IDCT_PERM 2
 413 #define FF_SIMPLE_IDCT_PERM 3
 414 #define FF_TRANSPOSE_IDCT_PERM 4
 415 #define FF_PARTTRANS_IDCT_PERM 5
 416 #define FF_SSE2_IDCT_PERM 6
 417
 418     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 419     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 420 #define BASIS_SHIFT 16
 421 #define RECON_SHIFT 6
 422
 423     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
 424 #define EDGE_WIDTH 16
 425
 426     /* h264 functions */
 427     void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
 428     void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 429     void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
 430     void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
 431     void (*h264_dct)(DCTELEM block[4][4]);
 432
 433     /* snow wavelet */
 434     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
 435     void (*horizontal_compose97i)(IDWTELEM *b, int width);
 436     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);
 437
 438     void (*prefetch)(void *mem, int stride, int h);
 439
 440     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 441
 442     /* vc1 functions */
 443     void (*vc1_inv_trans_8x8)(DCTELEM *b);
 444     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
 445     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
 446     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
 447     void (*vc1_v_overlap)(uint8_t* src, int stride);
 448     void (*vc1_h_overlap)(uint8_t* src, int stride);
 449     /* put 8x8 block with bicubic interpolation and quarterpel precision
 450      * last argument is actually round value instead of height
 451      */
 452     op_pixels_func put_vc1_mspel_pixels_tab[16];
 453
 454     /* intrax8 functions */
 455     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
 456     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
 457            int * range, int * sum,  int edges);
 458
 459     /* ape functions */
 460     /**
 461      * Add contents of the second vector to the first one.
 462      * @param len length of vectors, should be multiple of 16
 463      */
 464     void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
 465     /**
 466      * Add contents of the second vector to the first one.
 467      * @param len length of vectors, should be multiple of 16
 468      */
 469     void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
 470     /**
 471      * Calculate scalar product of two vectors.
 472      * @param len length of vectors, should be multiple of 16
 473      * @param shift number of bits to discard from product
 474      */
 475     int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
 476 } DSPContext;
 477
 478 void dsputil_static_init(void);
 479 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 480
 481 int ff_check_alignment(void);
 482
 483 /**
 484  * permute block according to permuatation.
 485  * @param last last non zero element in scantable order
 486  */
 487 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 488
 489 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 490
 491 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 492
 493 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 494 {
 495     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 496 }
 497
 498 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 499 {
 500     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 501 }
 502
 503 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 504     switch(type&0xFF){
 505     default:
 506     case FF_CMP_SAD:
 507         return lambda>>FF_LAMBDA_SHIFT;
 508     case FF_CMP_DCT:
 509         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 510     case FF_CMP_W53:
 511         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 512     case FF_CMP_W97:
 513         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 514     case FF_CMP_SATD:
 515     case FF_CMP_DCT264:
 516         return (2*lambda)>>FF_LAMBDA_SHIFT;
 517     case FF_CMP_RD:
 518     case FF_CMP_PSNR:
 519     case FF_CMP_SSE:
 520     case FF_CMP_NSSE:
 521         return lambda2>>FF_LAMBDA_SHIFT;
 522     case FF_CMP_BIT:
 523         return 1;
 524     }
 525 }
 526
 527 /**
 528  * Empty mmx state.
 529  * this must be called between any dsp function and float/double code.
 530  * for example sin(); dsp->idct_put(); emms_c(); cos()
 531  */
 532 #define emms_c()
 533
 534 /* should be defined by architectures supporting
 535    one or more MultiMedia extension */
 536 int mm_support(void);
 537
 538 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 539 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
 540 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
 541 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 542 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 543 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 544 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 545 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 546 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 547
 548 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
 549
 550 #if defined(HAVE_MMX)
 551
 552 #undef emms_c
 553
 554 #define MM_MMX    0x0001 /* standard MMX */
 555 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
 556 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
 557 #define MM_SSE    0x0008 /* SSE functions */
 558 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
 559 #define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
 560 #define MM_SSE3   0x0040 /* Prescott SSE3 functions */
 561 #define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
 562
 563 extern int mm_flags;
 564
 565 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 566 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 567 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 568
 569 static inline void emms(void)
 570 {
 571     asm volatile ("emms;":::"memory");
 572 }
 573
 574
 575 #define emms_c() \
 576 {\
 577     if (mm_flags & MM_MMX)\
 578         emms();\
 579 }
 580
 581 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
 582
 583 #elif defined(ARCH_ARMV4L)
 584
 585 #define MM_IWMMXT    0x0100 /* XScale IWMMXT */
 586
 587 extern int mm_flags;
 588
 589 #elif defined(ARCH_POWERPC)
 590
 591 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
 592
 593 extern int mm_flags;
 594
 595 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 596 #define STRIDE_ALIGN 16
 597
 598 #elif defined(HAVE_MMI)
 599
 600 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 601 #define STRIDE_ALIGN 16
 602
 603 #else
 604
 605 #define mm_flags 0
 606 #define mm_support() 0
 607
 608 #endif
 609
 610 #ifndef DECLARE_ALIGNED_8
 611 #   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 612 #endif
 613
 614 #ifndef STRIDE_ALIGN
 615 #   define STRIDE_ALIGN 8
 616 #endif
 617
 618 /* PSNR */
 619 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 620               int orig_linesize[3], int coded_linesize,
 621               AVCodecContext *avctx);
 622
 623 /* FFT computation */
 624
 625 /* NOTE: soon integer code will be added, so you must use the
 626    FFTSample type */
 627 typedef float FFTSample;
 628
 629 struct MDCTContext;
 630
 631 typedef struct FFTComplex {
 632     FFTSample re, im;
 633 } FFTComplex;
 634
 635 typedef struct FFTContext {
 636     int nbits;
 637     int inverse;
 638     uint16_t *revtab;
 639     FFTComplex *exptab;
 640     FFTComplex *exptab1; /* only used by SSE code */
 641     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 642     void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
 643                        const FFTSample *input, FFTSample *tmp);
 644     void (*imdct_half)(struct MDCTContext *s, FFTSample *output,
 645                        const FFTSample *input, FFTSample *tmp);
 646 } FFTContext;
 647
 648 int ff_fft_init(FFTContext *s, int nbits, int inverse);
 649 void ff_fft_permute(FFTContext *s, FFTComplex *z);
 650 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 651 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
 652 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
 653 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
 654 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
 655
 656 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
 657 {
 658     s->fft_calc(s, z);
 659 }
 660 void ff_fft_end(FFTContext *s);
 661
 662 /* MDCT computation */
 663
 664 typedef struct MDCTContext {
 665     int n;  /* size of MDCT (i.e. number of input data * 2) */
 666     int nbits; /* n = 2^nbits */
 667     /* pre/post rotation tables */
 668     FFTSample *tcos;
 669     FFTSample *tsin;
 670     FFTContext fft;
 671 } MDCTContext;
 672
 673 /**
 674  * Generate a Kaiser-Bessel Derived Window.
 675  * @param   window  pointer to half window
 676  * @param   alpha   determines window shape
 677  * @param   n       size of half window
 678  */
 679 void ff_kbd_window_init(float *window, float alpha, int n);
 680
 681 /**
 682  * Generate a sine window.
 683  * @param   window  pointer to half window
 684  * @param   n       size of half window
 685  */
 686 void ff_sine_window_init(float *window, int n);
 687
 688 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
 689 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
 690                 const FFTSample *input, FFTSample *tmp);
 691 void ff_imdct_half(MDCTContext *s, FFTSample *output,
 692                    const FFTSample *input, FFTSample *tmp);
 693 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
 694                         const FFTSample *input, FFTSample *tmp);
 695 void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output,
 696                         const FFTSample *input, FFTSample *tmp);
 697 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
 698                        const FFTSample *input, FFTSample *tmp);
 699 void ff_imdct_half_sse(MDCTContext *s, FFTSample *output,
 700                        const FFTSample *input, FFTSample *tmp);
 701 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
 702                const FFTSample *input, FFTSample *tmp);
 703 void ff_mdct_end(MDCTContext *s);
 704
 705 #define WRAPPER8_16(name8, name16)\
 706 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 707     return name8(s, dst           , src           , stride, h)\
 708           +name8(s, dst+8         , src+8         , stride, h);\
 709 }
 710
 711 #define WRAPPER8_16_SQ(name8, name16)\
 712 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 713     int score=0;\
 714     score +=name8(s, dst           , src           , stride, 8);\
 715     score +=name8(s, dst+8         , src+8         , stride, 8);\
 716     if(h==16){\
 717         dst += 8*stride;\
 718         src += 8*stride;\
 719         score +=name8(s, dst           , src           , stride, 8);\
 720         score +=name8(s, dst+8         , src+8         , stride, 8);\
 721     }\
 722     return score;\
 723 }
 724
 725
 726 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 727 {
 728     int i;
 729     for(i=0; i<h; i++)
 730     {
 731         AV_WN16(dst   , AV_RN16(src   ));
 732         dst+=dstStride;
 733         src+=srcStride;
 734     }
 735 }
 736
 737 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 738 {
 739     int i;
 740     for(i=0; i<h; i++)
 741     {
 742         AV_WN32(dst   , AV_RN32(src   ));
 743         dst+=dstStride;
 744         src+=srcStride;
 745     }
 746 }
 747
 748 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 749 {
 750     int i;
 751     for(i=0; i<h; i++)
 752     {
 753         AV_WN32(dst   , AV_RN32(src   ));
 754         AV_WN32(dst+4 , AV_RN32(src+4 ));
 755         dst+=dstStride;
 756         src+=srcStride;
 757     }
 758 }
 759
 760 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 761 {
 762     int i;
 763     for(i=0; i<h; i++)
 764     {
 765         AV_WN32(dst   , AV_RN32(src   ));
 766         AV_WN32(dst+4 , AV_RN32(src+4 ));
 767         dst[8]= src[8];
 768         dst+=dstStride;
 769         src+=srcStride;
 770     }
 771 }
 772
 773 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 774 {
 775     int i;
 776     for(i=0; i<h; i++)
 777     {
 778         AV_WN32(dst   , AV_RN32(src   ));
 779         AV_WN32(dst+4 , AV_RN32(src+4 ));
 780         AV_WN32(dst+8 , AV_RN32(src+8 ));
 781         AV_WN32(dst+12, AV_RN32(src+12));
 782         dst+=dstStride;
 783         src+=srcStride;
 784     }
 785 }
 786
 787 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 788 {
 789     int i;
 790     for(i=0; i<h; i++)
 791     {
 792         AV_WN32(dst   , AV_RN32(src   ));
 793         AV_WN32(dst+4 , AV_RN32(src+4 ));
 794         AV_WN32(dst+8 , AV_RN32(src+8 ));
 795         AV_WN32(dst+12, AV_RN32(src+12));
 796         dst[16]= src[16];
 797         dst+=dstStride;
 798         src+=srcStride;
 799     }
 800 }
 801
 802 #endif /* FFMPEG_DSPUTIL_H */