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 libavcodec/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 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 */
  39 typedef short DCTELEM;
  40 typedef int DWTELEM;
  41 typedef short IDWTELEM;
  42
  43 void fdct_ifast (DCTELEM *data);
  44 void fdct_ifast248 (DCTELEM *data);
  45 void ff_jpeg_fdct_islow (DCTELEM *data);
  46 void ff_fdct248_islow (DCTELEM *data);
  47
  48 void j_rev_dct (DCTELEM *data);
  49 void j_rev_dct4 (DCTELEM *data);
  50 void j_rev_dct2 (DCTELEM *data);
  51 void j_rev_dct1 (DCTELEM *data);
  52 void ff_wmv2_idct_c(DCTELEM *data);
  53
  54 void ff_fdct_mmx(DCTELEM *block);
  55 void ff_fdct_mmx2(DCTELEM *block);
  56 void ff_fdct_sse2(DCTELEM *block);
  57
  58 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
  59 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
  60 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
  61 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
  62 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
  63 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
  64 void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  65 void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  66 void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  67 void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  68
  69 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
  70                              const float *win, float add_bias, int len);
  71 void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
  72 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
  73
  74 /* encoding scans */
  75 extern const uint8_t ff_alternate_horizontal_scan[64];
  76 extern const uint8_t ff_alternate_vertical_scan[64];
  77 extern const uint8_t ff_zigzag_direct[64];
  78 extern const uint8_t ff_zigzag248_direct[64];
  79
  80 /* pixel operations */
  81 #define MAX_NEG_CROP 1024
  82
  83 /* temporary */
  84 extern uint32_t ff_squareTbl[512];
  85 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
  86
  87 /* VP3 DSP functions */
  88 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
  89 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  90 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  91
  92 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
  93 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
  94
  95 /* VP6 DSP functions */
  96 void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride,
  97                            const int16_t *h_weights, const int16_t *v_weights);
  98
  99 /* 1/2^n downscaling functions from imgconvert.c */
 100 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 101 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 102 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 103 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 104
 105 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
 106               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 107
 108 /* minimum alignment rules ;)
 109 If you notice errors in the align stuff, need more alignment for some ASM code
 110 for some CPU or need to use a function with less aligned data then send a mail
 111 to the ffmpeg-devel mailing list, ...
 112
 113 !warning These alignments might not match reality, (missing attribute((align))
 114 stuff somewhere possible).
 115 I (Michael) did not check them, these are just the alignments which I think
 116 could be reached easily ...
 117
 118 !future video codecs might need functions with less strict alignment
 119 */
 120
 121 /*
 122 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
 123 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
 124 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 125 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 126 void clear_blocks_c(DCTELEM *blocks);
 127 */
 128
 129 /* add and put pixel (decoding) */
 130 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 131 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
 132 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
 133 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);
 134 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 135 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 136 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
 137 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
 138
 139 #define DEF_OLD_QPEL(name)\
 140 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 141 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 142 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 143
 144 DEF_OLD_QPEL(qpel16_mc11_old_c)
 145 DEF_OLD_QPEL(qpel16_mc31_old_c)
 146 DEF_OLD_QPEL(qpel16_mc12_old_c)
 147 DEF_OLD_QPEL(qpel16_mc32_old_c)
 148 DEF_OLD_QPEL(qpel16_mc13_old_c)
 149 DEF_OLD_QPEL(qpel16_mc33_old_c)
 150 DEF_OLD_QPEL(qpel8_mc11_old_c)
 151 DEF_OLD_QPEL(qpel8_mc31_old_c)
 152 DEF_OLD_QPEL(qpel8_mc12_old_c)
 153 DEF_OLD_QPEL(qpel8_mc32_old_c)
 154 DEF_OLD_QPEL(qpel8_mc13_old_c)
 155 DEF_OLD_QPEL(qpel8_mc33_old_c)
 156
 157 #define CALL_2X_PIXELS(a, b, n)\
 158 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 159     b(block  , pixels  , line_size, h);\
 160     b(block+n, pixels+n, line_size, h);\
 161 }
 162
 163 /* motion estimation */
 164 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 165 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
 166 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))*/;
 167
 168
 169 // for snow slices
 170 typedef struct slice_buffer_s slice_buffer;
 171
 172 /**
 173  * Scantable.
 174  */
 175 typedef struct ScanTable{
 176     const uint8_t *scantable;
 177     uint8_t permutated[64];
 178     uint8_t raster_end[64];
 179 #if ARCH_PPC
 180                 /** Used by dct_quantize_altivec to find last-non-zero */
 181     DECLARE_ALIGNED(16, uint8_t, inverse[64]);
 182 #endif
 183 } ScanTable;
 184
 185 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
 186
 187 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
 188                          int block_w, int block_h,
 189                          int src_x, int src_y, int w, int h);
 190
 191 /**
 192  * DSPContext.
 193  */
 194 typedef struct DSPContext {
 195     /* pixel ops : interface with DCT */
 196     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 197     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 198     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 199     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 200     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 201     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
 202     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
 203     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
 204     /**
 205      * translational global motion compensation.
 206      */
 207     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 208     /**
 209      * global motion compensation.
 210      */
 211     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 212                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 213     void (*clear_block)(DCTELEM *block/*align 16*/);
 214     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 215     int (*pix_sum)(uint8_t * pix, int line_size);
 216     int (*pix_norm1)(uint8_t * pix, int line_size);
 217 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 218
 219     me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
 220     me_cmp_func sse[6];
 221     me_cmp_func hadamard8_diff[6];
 222     me_cmp_func dct_sad[6];
 223     me_cmp_func quant_psnr[6];
 224     me_cmp_func bit[6];
 225     me_cmp_func rd[6];
 226     me_cmp_func vsad[6];
 227     me_cmp_func vsse[6];
 228     me_cmp_func nsse[6];
 229     me_cmp_func w53[6];
 230     me_cmp_func w97[6];
 231     me_cmp_func dct_max[6];
 232     me_cmp_func dct264_sad[6];
 233
 234     me_cmp_func me_pre_cmp[6];
 235     me_cmp_func me_cmp[6];
 236     me_cmp_func me_sub_cmp[6];
 237     me_cmp_func mb_cmp[6];
 238     me_cmp_func ildct_cmp[6]; //only width 16 used
 239     me_cmp_func frame_skip_cmp[6]; //only width 8 used
 240
 241     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
 242                              int size);
 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 where the result is stored
 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 put_pixels_tab[4][4];
 255
 256     /**
 257      * Halfpel motion compensation with rounding (a+b+1)>>1.
 258      * This is an array[4][4] of motion compensation functions for 4
 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 into which the result is averaged (a+b+1)>>1
 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 avg_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 where the result is stored
 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 put_no_rnd_pixels_tab[4][4];
 279
 280     /**
 281      * Halfpel motion compensation with no rounding (a+b)>>1.
 282      * this is an array[2][4] of motion compensation functions for 2
 283      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 284      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 285      * @param block destination into which the result is averaged (a+b)>>1
 286      * @param pixels source
 287      * @param line_size number of bytes in a horizontal line of block
 288      * @param h height
 289      */
 290     op_pixels_func avg_no_rnd_pixels_tab[4][4];
 291
 292     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);
 293
 294     /**
 295      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 296      * this is an array[12] of motion compensation functions for the 9 thirdpe
 297      * positions<br>
 298      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 299      * @param block destination where the result is stored
 300      * @param pixels source
 301      * @param line_size number of bytes in a horizontal line of block
 302      * @param h height
 303      */
 304     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 305     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 306
 307     qpel_mc_func put_qpel_pixels_tab[2][16];
 308     qpel_mc_func avg_qpel_pixels_tab[2][16];
 309     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 310     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 311     qpel_mc_func put_mspel_pixels_tab[8];
 312
 313     /**
 314      * h264 Chroma MC
 315      */
 316     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 317     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 318     /* This is really one func used in VC-1 decoding */
 319     h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
 320     h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
 321
 322     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
 323     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
 324
 325     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
 326     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
 327
 328     h264_weight_func weight_h264_pixels_tab[10];
 329     h264_biweight_func biweight_h264_pixels_tab[10];
 330
 331     /* AVS specific */
 332     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
 333     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
 334     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 335     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 336     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 337     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 338     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 339
 340     me_cmp_func pix_abs[2][4];
 341
 342     /* huffyuv specific */
 343     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 344     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
 345     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 346     /**
 347      * subtract huffyuv's variant of median prediction
 348      * note, this might read from src1[-1], src2[-1]
 349      */
 350     void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
 351     void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
 352     int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
 353     void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue);
 354     /* this might write to dst[w] */
 355     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
 356     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
 357
 358     void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
 359     void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
 360     /* v/h_loop_filter_luma_intra: align 16 */
 361     void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 362     void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 363     void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
 364     void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
 365     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
 366     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
 367     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
 368     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],
 369                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
 370
 371     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 372     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 373
 374     void (*h261_loop_filter)(uint8_t *src, int stride);
 375
 376     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
 377     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
 378
 379     void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
 380     void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
 381
 382     void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride,
 383                              const int16_t *h_weights,const int16_t *v_weights);
 384
 385     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 386     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
 387     void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
 388     /* no alignment needed */
 389     void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
 390     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 391     void (*vector_fmul)(float *dst, const float *src, int len);
 392     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
 393     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
 394     void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
 395     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 396     void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
 397     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 398     void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
 399     void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
 400     /**
 401      * Multiply a vector of floats by a scalar float.  Source and
 402      * destination vectors must overlap exactly or not at all.
 403      * @param dst result vector, 16-byte aligned
 404      * @param src input vector, 16-byte aligned
 405      * @param mul scalar value
 406      * @param len length of vector, multiple of 4
 407      */
 408     void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
 409                                int len);
 410     /**
 411      * Multiply a vector of floats by concatenated short vectors of
 412      * floats and by a scalar float.  Source and destination vectors
 413      * must overlap exactly or not at all.
 414      * [0]: short vectors of length 2, 8-byte aligned
 415      * [1]: short vectors of length 4, 16-byte aligned
 416      * @param dst output vector, 16-byte aligned
 417      * @param src input vector, 16-byte aligned
 418      * @param sv  array of pointers to short vectors
 419      * @param mul scalar value
 420      * @param len number of elements in src and dst, multiple of 4
 421      */
 422     void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
 423                                      const float **sv, float mul, int len);
 424     /**
 425      * Multiply short vectors of floats by a scalar float, store
 426      * concatenated result.
 427      * [0]: short vectors of length 2, 8-byte aligned
 428      * [1]: short vectors of length 4, 16-byte aligned
 429      * @param dst output vector, 16-byte aligned
 430      * @param sv  array of pointers to short vectors
 431      * @param mul scalar value
 432      * @param len number of output elements, multiple of 4
 433      */
 434     void (*sv_fmul_scalar[2])(float *dst, const float **sv,
 435                               float mul, int len);
 436     /**
 437      * Calculate the scalar product of two vectors of floats.
 438      * @param v1  first vector, 16-byte aligned
 439      * @param v2  second vector, 16-byte aligned
 440      * @param len length of vectors, multiple of 4
 441      */
 442     float (*scalarproduct_float)(const float *v1, const float *v2, int len);
 443     /**
 444      * Calculate the sum and difference of two vectors of floats.
 445      * @param v1  first input vector, sum output, 16-byte aligned
 446      * @param v2  second input vector, difference output, 16-byte aligned
 447      * @param len length of vectors, multiple of 4
 448      */
 449     void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
 450
 451     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
 452      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
 453     void (*float_to_int16)(int16_t *dst, const float *src, long len);
 454     void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
 455
 456     /* (I)DCT */
 457     void (*fdct)(DCTELEM *block/* align 16*/);
 458     void (*fdct248)(DCTELEM *block/* align 16*/);
 459
 460     /* IDCT really*/
 461     void (*idct)(DCTELEM *block/* align 16*/);
 462
 463     /**
 464      * block -> idct -> clip to unsigned 8 bit -> dest.
 465      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 466      * @param line_size size in bytes of a horizontal line of dest
 467      */
 468     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 469
 470     /**
 471      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 472      * @param line_size size in bytes of a horizontal line of dest
 473      */
 474     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 475
 476     /**
 477      * idct input permutation.
 478      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 479      * IDCT)
 480      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 481      * with the zigzag/alternate scan<br>
 482      * an example to avoid confusion:
 483      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 484      * - (x -> referece dct -> reference idct -> x)
 485      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 486      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 487      */
 488     uint8_t idct_permutation[64];
 489     int idct_permutation_type;
 490 #define FF_NO_IDCT_PERM 1
 491 #define FF_LIBMPEG2_IDCT_PERM 2
 492 #define FF_SIMPLE_IDCT_PERM 3
 493 #define FF_TRANSPOSE_IDCT_PERM 4
 494 #define FF_PARTTRANS_IDCT_PERM 5
 495 #define FF_SSE2_IDCT_PERM 6
 496
 497     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 498     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 499 #define BASIS_SHIFT 16
 500 #define RECON_SHIFT 6
 501
 502     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
 503 #define EDGE_WIDTH 16
 504
 505     /* h264 functions */
 506     /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them
 507        NOTE!!! if you implement any of h264_idct_add, h264_idct_add16, h264_idct_add16intra, h264_idct_add8 then you must implement all of them
 508         The reason for above, is that no 2 out of one list may use a different permutation.
 509     */
 510     void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
 511     void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
 512     void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
 513     void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
 514     void (*h264_dct)(DCTELEM block[4][4]);
 515     void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 516     void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 517     void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 518     void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 519
 520     /* snow wavelet */
 521     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
 522     void (*horizontal_compose97i)(IDWTELEM *b, int width);
 523     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);
 524
 525     void (*prefetch)(void *mem, int stride, int h);
 526
 527     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 528
 529     /* mlp/truehd functions */
 530     void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
 531                                int firorder, int iirorder,
 532                                unsigned int filter_shift, int32_t mask, int blocksize,
 533                                int32_t *sample_buffer);
 534
 535     /* vc1 functions */
 536     void (*vc1_inv_trans_8x8)(DCTELEM *b);
 537     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
 538     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
 539     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
 540     void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 541     void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 542     void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 543     void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 544     void (*vc1_v_overlap)(uint8_t* src, int stride);
 545     void (*vc1_h_overlap)(uint8_t* src, int stride);
 546     void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
 547     void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
 548     void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
 549     void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
 550     void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
 551     void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
 552     /* put 8x8 block with bicubic interpolation and quarterpel precision
 553      * last argument is actually round value instead of height
 554      */
 555     op_pixels_func put_vc1_mspel_pixels_tab[16];
 556     op_pixels_func avg_vc1_mspel_pixels_tab[16];
 557
 558     /* intrax8 functions */
 559     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
 560     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
 561            int * range, int * sum,  int edges);
 562
 563     /* ape functions */
 564     /**
 565      * Add contents of the second vector to the first one.
 566      * @param len length of vectors, should be multiple of 16
 567      */
 568     void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
 569     /**
 570      * Add contents of the second vector to the first one.
 571      * @param len length of vectors, should be multiple of 16
 572      */
 573     void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
 574     /**
 575      * Calculate scalar product of two vectors.
 576      * @param len length of vectors, should be multiple of 16
 577      * @param shift number of bits to discard from product
 578      */
 579     int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
 580
 581     /* rv30 functions */
 582     qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
 583     qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
 584
 585     /* rv40 functions */
 586     qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
 587     qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
 588     h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
 589     h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
 590 } DSPContext;
 591
 592 void dsputil_static_init(void);
 593 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 594
 595 int ff_check_alignment(void);
 596
 597 /**
 598  * permute block according to permuatation.
 599  * @param last last non zero element in scantable order
 600  */
 601 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 602
 603 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 604
 605 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 606
 607 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 608 {
 609     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 610 }
 611
 612 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 613 {
 614     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 615 }
 616
 617 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 618     switch(type&0xFF){
 619     default:
 620     case FF_CMP_SAD:
 621         return lambda>>FF_LAMBDA_SHIFT;
 622     case FF_CMP_DCT:
 623         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 624     case FF_CMP_W53:
 625         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 626     case FF_CMP_W97:
 627         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 628     case FF_CMP_SATD:
 629     case FF_CMP_DCT264:
 630         return (2*lambda)>>FF_LAMBDA_SHIFT;
 631     case FF_CMP_RD:
 632     case FF_CMP_PSNR:
 633     case FF_CMP_SSE:
 634     case FF_CMP_NSSE:
 635         return lambda2>>FF_LAMBDA_SHIFT;
 636     case FF_CMP_BIT:
 637         return 1;
 638     }
 639 }
 640
 641 /**
 642  * Empty mmx state.
 643  * this must be called between any dsp function and float/double code.
 644  * for example sin(); dsp->idct_put(); emms_c(); cos()
 645  */
 646 #define emms_c()
 647
 648 /* should be defined by architectures supporting
 649    one or more MultiMedia extension */
 650 int mm_support(void);
 651 extern int mm_flags;
 652
 653 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 654 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
 655 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
 656 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 657 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 658 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 659 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 660 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 661 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 662
 663 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
 664 #define DECLARE_ALIGNED_8(t, v)  DECLARE_ALIGNED(8, t, v)
 665
 666 #if HAVE_MMX
 667
 668 #undef emms_c
 669
 670 static inline void emms(void)
 671 {
 672     __asm__ volatile ("emms;":::"memory");
 673 }
 674
 675
 676 #define emms_c() \
 677 {\
 678     if (mm_flags & FF_MM_MMX)\
 679         emms();\
 680 }
 681
 682 #elif ARCH_ARM
 683
 684 #if HAVE_NEON
 685 #   define STRIDE_ALIGN 16
 686 #endif
 687
 688 #elif ARCH_PPC
 689
 690 #define STRIDE_ALIGN 16
 691
 692 #elif HAVE_MMI
 693
 694 #define STRIDE_ALIGN 16
 695
 696 #else
 697
 698 #define mm_flags 0
 699 #define mm_support() 0
 700
 701 #endif
 702
 703 #ifndef STRIDE_ALIGN
 704 #   define STRIDE_ALIGN 8
 705 #endif
 706
 707 /* PSNR */
 708 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 709               int orig_linesize[3], int coded_linesize,
 710               AVCodecContext *avctx);
 711
 712 /* FFT computation */
 713
 714 /* NOTE: soon integer code will be added, so you must use the
 715    FFTSample type */
 716 typedef float FFTSample;
 717
 718 typedef struct FFTComplex {
 719     FFTSample re, im;
 720 } FFTComplex;
 721
 722 typedef struct FFTContext {
 723     int nbits;
 724     int inverse;
 725     uint16_t *revtab;
 726     FFTComplex *exptab;
 727     FFTComplex *exptab1; /* only used by SSE code */
 728     FFTComplex *tmp_buf;
 729     int mdct_size; /* size of MDCT (i.e. number of input data * 2) */
 730     int mdct_bits; /* n = 2^nbits */
 731     /* pre/post rotation tables */
 732     FFTSample *tcos;
 733     FFTSample *tsin;
 734     void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
 735     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 736     void (*imdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
 737     void (*imdct_half)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
 738     void (*mdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
 739     int split_radix;
 740     int permutation;
 741 #define FF_MDCT_PERM_NONE       0
 742 #define FF_MDCT_PERM_INTERLEAVE 1
 743 } FFTContext;
 744
 745 #if CONFIG_HARDCODED_TABLES
 746 #define COSTABLE_CONST const
 747 #define SINTABLE_CONST const
 748 #else
 749 #define COSTABLE_CONST
 750 #define SINTABLE_CONST
 751 #endif
 752
 753 #define COSTABLE(size) \
 754     COSTABLE_CONST DECLARE_ALIGNED_16(FFTSample, ff_cos_##size[size/2])
 755 #define SINTABLE(size) \
 756     SINTABLE_CONST DECLARE_ALIGNED_16(FFTSample, ff_sin_##size[size/2])
 757 extern COSTABLE(16);
 758 extern COSTABLE(32);
 759 extern COSTABLE(64);
 760 extern COSTABLE(128);
 761 extern COSTABLE(256);
 762 extern COSTABLE(512);
 763 extern COSTABLE(1024);
 764 extern COSTABLE(2048);
 765 extern COSTABLE(4096);
 766 extern COSTABLE(8192);
 767 extern COSTABLE(16384);
 768 extern COSTABLE(32768);
 769 extern COSTABLE(65536);
 770 extern COSTABLE_CONST FFTSample* const ff_cos_tabs[17];
 771
 772 /**
 773  * Initializes the cosine table in ff_cos_tabs[index]
 774  * \param index index in ff_cos_tabs array of the table to initialize
 775  */
 776 void ff_init_ff_cos_tabs(int index);
 777
 778 extern SINTABLE(16);
 779 extern SINTABLE(32);
 780 extern SINTABLE(64);
 781 extern SINTABLE(128);
 782 extern SINTABLE(256);
 783 extern SINTABLE(512);
 784 extern SINTABLE(1024);
 785 extern SINTABLE(2048);
 786 extern SINTABLE(4096);
 787 extern SINTABLE(8192);
 788 extern SINTABLE(16384);
 789 extern SINTABLE(32768);
 790 extern SINTABLE(65536);
 791
 792 /**
 793  * Sets up a complex FFT.
 794  * @param nbits           log2 of the length of the input array
 795  * @param inverse         if 0 perform the forward transform, if 1 perform the inverse
 796  */
 797 int ff_fft_init(FFTContext *s, int nbits, int inverse);
 798 void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
 799 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 800
 801 void ff_fft_init_altivec(FFTContext *s);
 802 void ff_fft_init_mmx(FFTContext *s);
 803 void ff_fft_init_arm(FFTContext *s);
 804
 805 /**
 806  * Do the permutation needed BEFORE calling ff_fft_calc().
 807  */
 808 static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
 809 {
 810     s->fft_permute(s, z);
 811 }
 812 /**
 813  * Do a complex FFT with the parameters defined in ff_fft_init(). The
 814  * input data must be permuted before. No 1.0/sqrt(n) normalization is done.
 815  */
 816 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
 817 {
 818     s->fft_calc(s, z);
 819 }
 820 void ff_fft_end(FFTContext *s);
 821
 822 /* MDCT computation */
 823
 824 static inline void ff_imdct_calc(FFTContext *s, FFTSample *output, const FFTSample *input)
 825 {
 826     s->imdct_calc(s, output, input);
 827 }
 828 static inline void ff_imdct_half(FFTContext *s, FFTSample *output, const FFTSample *input)
 829 {
 830     s->imdct_half(s, output, input);
 831 }
 832
 833 static inline void ff_mdct_calc(FFTContext *s, FFTSample *output,
 834                                 const FFTSample *input)
 835 {
 836     s->mdct_calc(s, output, input);
 837 }
 838
 839 /**
 840  * Generate a Kaiser-Bessel Derived Window.
 841  * @param   window  pointer to half window
 842  * @param   alpha   determines window shape
 843  * @param   n       size of half window
 844  */
 845 void ff_kbd_window_init(float *window, float alpha, int n);
 846
 847 /**
 848  * Generate a sine window.
 849  * @param   window  pointer to half window
 850  * @param   n       size of half window
 851  */
 852 void ff_sine_window_init(float *window, int n);
 853 extern float ff_sine_32  [  32];
 854 extern float ff_sine_64  [  64];
 855 extern float ff_sine_128 [ 128];
 856 extern float ff_sine_256 [ 256];
 857 extern float ff_sine_512 [ 512];
 858 extern float ff_sine_1024[1024];
 859 extern float ff_sine_2048[2048];
 860 extern float ff_sine_4096[4096];
 861 extern float * const ff_sine_windows[13];
 862
 863 int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale);
 864 void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input);
 865 void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input);
 866 void ff_mdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input);
 867 void ff_mdct_end(FFTContext *s);
 868
 869 /* Real Discrete Fourier Transform */
 870
 871 enum RDFTransformType {
 872     RDFT,
 873     IRDFT,
 874     RIDFT,
 875     IRIDFT,
 876 };
 877
 878 typedef struct {
 879     int nbits;
 880     int inverse;
 881     int sign_convention;
 882
 883     /* pre/post rotation tables */
 884     const FFTSample *tcos;
 885     SINTABLE_CONST FFTSample *tsin;
 886     FFTContext fft;
 887 } RDFTContext;
 888
 889 /**
 890  * Sets up a real FFT.
 891  * @param nbits           log2 of the length of the input array
 892  * @param trans           the type of transform
 893  */
 894 int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans);
 895 void ff_rdft_calc(RDFTContext *s, FFTSample *data);
 896 void ff_rdft_end(RDFTContext *s);
 897
 898 #define WRAPPER8_16(name8, name16)\
 899 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 900     return name8(s, dst           , src           , stride, h)\
 901           +name8(s, dst+8         , src+8         , stride, h);\
 902 }
 903
 904 #define WRAPPER8_16_SQ(name8, name16)\
 905 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 906     int score=0;\
 907     score +=name8(s, dst           , src           , stride, 8);\
 908     score +=name8(s, dst+8         , src+8         , stride, 8);\
 909     if(h==16){\
 910         dst += 8*stride;\
 911         src += 8*stride;\
 912         score +=name8(s, dst           , src           , stride, 8);\
 913         score +=name8(s, dst+8         , src+8         , stride, 8);\
 914     }\
 915     return score;\
 916 }
 917
 918
 919 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 920 {
 921     int i;
 922     for(i=0; i<h; i++)
 923     {
 924         AV_WN16(dst   , AV_RN16(src   ));
 925         dst+=dstStride;
 926         src+=srcStride;
 927     }
 928 }
 929
 930 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 931 {
 932     int i;
 933     for(i=0; i<h; i++)
 934     {
 935         AV_WN32(dst   , AV_RN32(src   ));
 936         dst+=dstStride;
 937         src+=srcStride;
 938     }
 939 }
 940
 941 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 942 {
 943     int i;
 944     for(i=0; i<h; i++)
 945     {
 946         AV_WN32(dst   , AV_RN32(src   ));
 947         AV_WN32(dst+4 , AV_RN32(src+4 ));
 948         dst+=dstStride;
 949         src+=srcStride;
 950     }
 951 }
 952
 953 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 954 {
 955     int i;
 956     for(i=0; i<h; i++)
 957     {
 958         AV_WN32(dst   , AV_RN32(src   ));
 959         AV_WN32(dst+4 , AV_RN32(src+4 ));
 960         dst[8]= src[8];
 961         dst+=dstStride;
 962         src+=srcStride;
 963     }
 964 }
 965
 966 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 967 {
 968     int i;
 969     for(i=0; i<h; i++)
 970     {
 971         AV_WN32(dst   , AV_RN32(src   ));
 972         AV_WN32(dst+4 , AV_RN32(src+4 ));
 973         AV_WN32(dst+8 , AV_RN32(src+8 ));
 974         AV_WN32(dst+12, AV_RN32(src+12));
 975         dst+=dstStride;
 976         src+=srcStride;
 977     }
 978 }
 979
 980 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 981 {
 982     int i;
 983     for(i=0; i<h; i++)
 984     {
 985         AV_WN32(dst   , AV_RN32(src   ));
 986         AV_WN32(dst+4 , AV_RN32(src+4 ));
 987         AV_WN32(dst+8 , AV_RN32(src+8 ));
 988         AV_WN32(dst+12, AV_RN32(src+12));
 989         dst[16]= src[16];
 990         dst+=dstStride;
 991         src+=srcStride;
 992     }
 993 }
 994
 995 #endif /* AVCODEC_DSPUTIL_H */