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 /* Bink functions */
 100 void ff_bink_idct_c    (DCTELEM *block);
 101 void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
 102 void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
 103
 104 /* CAVS functions */
 105 void ff_put_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride);
 106 void ff_avg_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride);
 107 void ff_put_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride);
 108 void ff_avg_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride);
 109
 110 /* VC1 functions */
 111 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
 112 void ff_avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
 113
 114 /* EA functions */
 115 void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
 116
 117 /* 1/2^n downscaling functions from imgconvert.c */
 118 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 119 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 120 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 121 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 122
 123 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
 124               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 125
 126 /* minimum alignment rules ;)
 127 If you notice errors in the align stuff, need more alignment for some ASM code
 128 for some CPU or need to use a function with less aligned data then send a mail
 129 to the ffmpeg-devel mailing list, ...
 130
 131 !warning These alignments might not match reality, (missing attribute((align))
 132 stuff somewhere possible).
 133 I (Michael) did not check them, these are just the alignments which I think
 134 could be reached easily ...
 135
 136 !future video codecs might need functions with less strict alignment
 137 */
 138
 139 /*
 140 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
 141 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
 142 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 143 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 144 void clear_blocks_c(DCTELEM *blocks);
 145 */
 146
 147 /* add and put pixel (decoding) */
 148 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 149 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
 150 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
 151 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);
 152 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 153 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 154 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
 155 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
 156
 157 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
 158
 159 #define DEF_OLD_QPEL(name)\
 160 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 161 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 162 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 163
 164 DEF_OLD_QPEL(qpel16_mc11_old_c)
 165 DEF_OLD_QPEL(qpel16_mc31_old_c)
 166 DEF_OLD_QPEL(qpel16_mc12_old_c)
 167 DEF_OLD_QPEL(qpel16_mc32_old_c)
 168 DEF_OLD_QPEL(qpel16_mc13_old_c)
 169 DEF_OLD_QPEL(qpel16_mc33_old_c)
 170 DEF_OLD_QPEL(qpel8_mc11_old_c)
 171 DEF_OLD_QPEL(qpel8_mc31_old_c)
 172 DEF_OLD_QPEL(qpel8_mc12_old_c)
 173 DEF_OLD_QPEL(qpel8_mc32_old_c)
 174 DEF_OLD_QPEL(qpel8_mc13_old_c)
 175 DEF_OLD_QPEL(qpel8_mc33_old_c)
 176
 177 #define CALL_2X_PIXELS(a, b, n)\
 178 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 179     b(block  , pixels  , line_size, h);\
 180     b(block+n, pixels+n, line_size, h);\
 181 }
 182
 183 /* motion estimation */
 184 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 185 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
 186 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
 187
 188
 189 // for snow slices
 190 typedef struct slice_buffer_s slice_buffer;
 191
 192 /**
 193  * Scantable.
 194  */
 195 typedef struct ScanTable{
 196     const uint8_t *scantable;
 197     uint8_t permutated[64];
 198     uint8_t raster_end[64];
 199 #if ARCH_PPC
 200                 /** Used by dct_quantize_altivec to find last-non-zero */
 201     DECLARE_ALIGNED(16, uint8_t, inverse)[64];
 202 #endif
 203 } ScanTable;
 204
 205 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
 206
 207 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
 208                          int block_w, int block_h,
 209                          int src_x, int src_y, int w, int h);
 210
 211 /**
 212  * DSPContext.
 213  */
 214 typedef struct DSPContext {
 215     /* pixel ops : interface with DCT */
 216     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 217     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 218     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 219     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 220     void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 221     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 222     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
 223     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
 224     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
 225     /**
 226      * translational global motion compensation.
 227      */
 228     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 229     /**
 230      * global motion compensation.
 231      */
 232     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 233                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 234     void (*clear_block)(DCTELEM *block/*align 16*/);
 235     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 236     int (*pix_sum)(uint8_t * pix, int line_size);
 237     int (*pix_norm1)(uint8_t * pix, int line_size);
 238 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 239
 240     me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
 241     me_cmp_func sse[6];
 242     me_cmp_func hadamard8_diff[6];
 243     me_cmp_func dct_sad[6];
 244     me_cmp_func quant_psnr[6];
 245     me_cmp_func bit[6];
 246     me_cmp_func rd[6];
 247     me_cmp_func vsad[6];
 248     me_cmp_func vsse[6];
 249     me_cmp_func nsse[6];
 250     me_cmp_func w53[6];
 251     me_cmp_func w97[6];
 252     me_cmp_func dct_max[6];
 253     me_cmp_func dct264_sad[6];
 254
 255     me_cmp_func me_pre_cmp[6];
 256     me_cmp_func me_cmp[6];
 257     me_cmp_func me_sub_cmp[6];
 258     me_cmp_func mb_cmp[6];
 259     me_cmp_func ildct_cmp[6]; //only width 16 used
 260     me_cmp_func frame_skip_cmp[6]; //only width 8 used
 261
 262     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
 263                              int size);
 264
 265     /**
 266      * Halfpel motion compensation with rounding (a+b+1)>>1.
 267      * this is an array[4][4] of motion compensation functions for 4
 268      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 269      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 270      * @param block destination where the result is stored
 271      * @param pixels source
 272      * @param line_size number of bytes in a horizontal line of block
 273      * @param h height
 274      */
 275     op_pixels_func put_pixels_tab[4][4];
 276
 277     /**
 278      * Halfpel motion compensation with rounding (a+b+1)>>1.
 279      * This is an array[4][4] of motion compensation functions for 4
 280      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 281      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 282      * @param block destination into which the result is averaged (a+b+1)>>1
 283      * @param pixels source
 284      * @param line_size number of bytes in a horizontal line of block
 285      * @param h height
 286      */
 287     op_pixels_func avg_pixels_tab[4][4];
 288
 289     /**
 290      * Halfpel motion compensation with no rounding (a+b)>>1.
 291      * this is an array[2][4] of motion compensation functions for 2
 292      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 293      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 294      * @param block destination where the result is stored
 295      * @param pixels source
 296      * @param line_size number of bytes in a horizontal line of block
 297      * @param h height
 298      */
 299     op_pixels_func put_no_rnd_pixels_tab[4][4];
 300
 301     /**
 302      * Halfpel motion compensation with no rounding (a+b)>>1.
 303      * this is an array[2][4] of motion compensation functions for 2
 304      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 305      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 306      * @param block destination into which the result is averaged (a+b)>>1
 307      * @param pixels source
 308      * @param line_size number of bytes in a horizontal line of block
 309      * @param h height
 310      */
 311     op_pixels_func avg_no_rnd_pixels_tab[4][4];
 312
 313     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);
 314
 315     /**
 316      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 317      * this is an array[12] of motion compensation functions for the 9 thirdpe
 318      * positions<br>
 319      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 320      * @param block destination where the result is stored
 321      * @param pixels source
 322      * @param line_size number of bytes in a horizontal line of block
 323      * @param h height
 324      */
 325     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 326     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 327
 328     qpel_mc_func put_qpel_pixels_tab[2][16];
 329     qpel_mc_func avg_qpel_pixels_tab[2][16];
 330     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 331     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 332     qpel_mc_func put_mspel_pixels_tab[8];
 333
 334     /**
 335      * h264 Chroma MC
 336      */
 337     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 338     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 339     /* This is really one func used in VC-1 decoding */
 340     h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
 341     h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
 342
 343     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
 344     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
 345
 346     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
 347     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
 348
 349     h264_weight_func weight_h264_pixels_tab[10];
 350     h264_biweight_func biweight_h264_pixels_tab[10];
 351
 352     /* AVS specific */
 353     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
 354     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
 355     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 356     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 357     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 358     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 359     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 360
 361     me_cmp_func pix_abs[2][4];
 362
 363     /* huffyuv specific */
 364     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 365     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
 366     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 367     /**
 368      * subtract huffyuv's variant of median prediction
 369      * note, this might read from src1[-1], src2[-1]
 370      */
 371     void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
 372     void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
 373     int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
 374     void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
 375     /* this might write to dst[w] */
 376     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
 377     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
 378
 379     void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
 380     void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
 381     /* v/h_loop_filter_luma_intra: align 16 */
 382     void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 383     void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 384     void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
 385     void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
 386     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
 387     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
 388     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
 389     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],
 390                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
 391
 392     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 393     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 394
 395     void (*h261_loop_filter)(uint8_t *src, int stride);
 396
 397     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
 398     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
 399
 400     void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
 401     void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
 402
 403     void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride,
 404                              const int16_t *h_weights,const int16_t *v_weights);
 405
 406     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 407     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
 408     void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
 409     /* no alignment needed */
 410     void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
 411     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 412     void (*vector_fmul)(float *dst, const float *src, int len);
 413     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
 414     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
 415     void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
 416     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 417     void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
 418     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 419     void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
 420     void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
 421     /**
 422      * Multiply a vector of floats by a scalar float.  Source and
 423      * destination vectors must overlap exactly or not at all.
 424      * @param dst result vector, 16-byte aligned
 425      * @param src input vector, 16-byte aligned
 426      * @param mul scalar value
 427      * @param len length of vector, multiple of 4
 428      */
 429     void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
 430                                int len);
 431     /**
 432      * Multiply a vector of floats by concatenated short vectors of
 433      * floats and by a scalar float.  Source and destination vectors
 434      * must overlap exactly or not at all.
 435      * [0]: short vectors of length 2, 8-byte aligned
 436      * [1]: short vectors of length 4, 16-byte aligned
 437      * @param dst output vector, 16-byte aligned
 438      * @param src input vector, 16-byte aligned
 439      * @param sv  array of pointers to short vectors
 440      * @param mul scalar value
 441      * @param len number of elements in src and dst, multiple of 4
 442      */
 443     void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
 444                                      const float **sv, float mul, int len);
 445     /**
 446      * Multiply short vectors of floats by a scalar float, store
 447      * concatenated result.
 448      * [0]: short vectors of length 2, 8-byte aligned
 449      * [1]: short vectors of length 4, 16-byte aligned
 450      * @param dst output vector, 16-byte aligned
 451      * @param sv  array of pointers to short vectors
 452      * @param mul scalar value
 453      * @param len number of output elements, multiple of 4
 454      */
 455     void (*sv_fmul_scalar[2])(float *dst, const float **sv,
 456                               float mul, int len);
 457     /**
 458      * Calculate the scalar product of two vectors of floats.
 459      * @param v1  first vector, 16-byte aligned
 460      * @param v2  second vector, 16-byte aligned
 461      * @param len length of vectors, multiple of 4
 462      */
 463     float (*scalarproduct_float)(const float *v1, const float *v2, int len);
 464     /**
 465      * Calculate the sum and difference of two vectors of floats.
 466      * @param v1  first input vector, sum output, 16-byte aligned
 467      * @param v2  second input vector, difference output, 16-byte aligned
 468      * @param len length of vectors, multiple of 4
 469      */
 470     void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
 471
 472     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
 473      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
 474     void (*float_to_int16)(int16_t *dst, const float *src, long len);
 475     void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
 476
 477     /* (I)DCT */
 478     void (*fdct)(DCTELEM *block/* align 16*/);
 479     void (*fdct248)(DCTELEM *block/* align 16*/);
 480
 481     /* IDCT really*/
 482     void (*idct)(DCTELEM *block/* align 16*/);
 483
 484     /**
 485      * block -> idct -> clip to unsigned 8 bit -> dest.
 486      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 487      * @param line_size size in bytes of a horizontal line of dest
 488      */
 489     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 490
 491     /**
 492      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 493      * @param line_size size in bytes of a horizontal line of dest
 494      */
 495     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 496
 497     /**
 498      * idct input permutation.
 499      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 500      * IDCT)
 501      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 502      * with the zigzag/alternate scan<br>
 503      * an example to avoid confusion:
 504      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 505      * - (x -> referece dct -> reference idct -> x)
 506      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 507      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 508      */
 509     uint8_t idct_permutation[64];
 510     int idct_permutation_type;
 511 #define FF_NO_IDCT_PERM 1
 512 #define FF_LIBMPEG2_IDCT_PERM 2
 513 #define FF_SIMPLE_IDCT_PERM 3
 514 #define FF_TRANSPOSE_IDCT_PERM 4
 515 #define FF_PARTTRANS_IDCT_PERM 5
 516 #define FF_SSE2_IDCT_PERM 6
 517
 518     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 519     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 520 #define BASIS_SHIFT 16
 521 #define RECON_SHIFT 6
 522
 523     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
 524 #define EDGE_WIDTH 16
 525
 526     /* h264 functions */
 527     /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them
 528        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
 529         The reason for above, is that no 2 out of one list may use a different permutation.
 530     */
 531     void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
 532     void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
 533     void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
 534     void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
 535     void (*h264_dct)(DCTELEM block[4][4]);
 536     void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 537     void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 538     void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 539     void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 540
 541     /* snow wavelet */
 542     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
 543     void (*horizontal_compose97i)(IDWTELEM *b, int width);
 544     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);
 545
 546     void (*prefetch)(void *mem, int stride, int h);
 547
 548     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 549
 550     /* mlp/truehd functions */
 551     void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
 552                                int firorder, int iirorder,
 553                                unsigned int filter_shift, int32_t mask, int blocksize,
 554                                int32_t *sample_buffer);
 555
 556     /* vc1 functions */
 557     void (*vc1_inv_trans_8x8)(DCTELEM *b);
 558     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
 559     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
 560     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
 561     void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 562     void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 563     void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 564     void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 565     void (*vc1_v_overlap)(uint8_t* src, int stride);
 566     void (*vc1_h_overlap)(uint8_t* src, int stride);
 567     void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
 568     void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
 569     void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
 570     void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
 571     void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
 572     void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
 573     /* put 8x8 block with bicubic interpolation and quarterpel precision
 574      * last argument is actually round value instead of height
 575      */
 576     op_pixels_func put_vc1_mspel_pixels_tab[16];
 577     op_pixels_func avg_vc1_mspel_pixels_tab[16];
 578
 579     /* intrax8 functions */
 580     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
 581     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
 582            int * range, int * sum,  int edges);
 583
 584     /**
 585      * Calculate scalar product of two vectors.
 586      * @param len length of vectors, should be multiple of 16
 587      * @param shift number of bits to discard from product
 588      */
 589     int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
 590     /* ape functions */
 591     /**
 592      * Calculate scalar product of v1 and v2,
 593      * and v1[i] += v3[i] * mul
 594      * @param len length of vectors, should be multiple of 16
 595      */
 596     int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, int16_t *v2, int16_t *v3, int len, int mul);
 597
 598     /* rv30 functions */
 599     qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
 600     qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
 601
 602     /* rv40 functions */
 603     qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
 604     qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
 605     h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
 606     h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
 607
 608     /* bink functions */
 609     op_fill_func fill_block_tab[2];
 610     void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
 611 } DSPContext;
 612
 613 void dsputil_static_init(void);
 614 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 615
 616 int ff_check_alignment(void);
 617
 618 /**
 619  * permute block according to permuatation.
 620  * @param last last non zero element in scantable order
 621  */
 622 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 623
 624 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 625
 626 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 627
 628 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 629 {
 630     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 631 }
 632
 633 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 634 {
 635     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 636 }
 637
 638 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 639     switch(type&0xFF){
 640     default:
 641     case FF_CMP_SAD:
 642         return lambda>>FF_LAMBDA_SHIFT;
 643     case FF_CMP_DCT:
 644         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 645     case FF_CMP_W53:
 646         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 647     case FF_CMP_W97:
 648         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 649     case FF_CMP_SATD:
 650     case FF_CMP_DCT264:
 651         return (2*lambda)>>FF_LAMBDA_SHIFT;
 652     case FF_CMP_RD:
 653     case FF_CMP_PSNR:
 654     case FF_CMP_SSE:
 655     case FF_CMP_NSSE:
 656         return lambda2>>FF_LAMBDA_SHIFT;
 657     case FF_CMP_BIT:
 658         return 1;
 659     }
 660 }
 661
 662 /**
 663  * Empty mmx state.
 664  * this must be called between any dsp function and float/double code.
 665  * for example sin(); dsp->idct_put(); emms_c(); cos()
 666  */
 667 #define emms_c()
 668
 669 /* should be defined by architectures supporting
 670    one or more MultiMedia extension */
 671 int mm_support(void);
 672 extern int mm_flags;
 673
 674 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 675 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
 676 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
 677 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 678 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 679 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 680 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 681 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 682 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 683
 684 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx);
 685 void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
 686 void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
 687 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
 688 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
 689 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
 690 void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
 691
 692 #if HAVE_MMX
 693
 694 #undef emms_c
 695
 696 static inline void emms(void)
 697 {
 698     __asm__ volatile ("emms;":::"memory");
 699 }
 700
 701
 702 #define emms_c() \
 703 {\
 704     if (mm_flags & FF_MM_MMX)\
 705         emms();\
 706 }
 707
 708 #elif ARCH_ARM
 709
 710 #if HAVE_NEON
 711 #   define STRIDE_ALIGN 16
 712 #endif
 713
 714 #elif ARCH_PPC
 715
 716 #define STRIDE_ALIGN 16
 717
 718 #elif HAVE_MMI
 719
 720 #define STRIDE_ALIGN 16
 721
 722 #else
 723
 724 #define mm_flags 0
 725 #define mm_support() 0
 726
 727 #endif
 728
 729 #ifndef STRIDE_ALIGN
 730 #   define STRIDE_ALIGN 8
 731 #endif
 732
 733 #define LOCAL_ALIGNED(a, t, v, s, ...)                          \
 734     uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)];              \
 735     t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a)
 736
 737 #if HAVE_LOCAL_ALIGNED_8
 738 #   define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__
 739 #else
 740 #   define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__)
 741 #endif
 742
 743 #if HAVE_LOCAL_ALIGNED_16
 744 #   define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__
 745 #else
 746 #   define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__)
 747 #endif
 748
 749 /* PSNR */
 750 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 751               int orig_linesize[3], int coded_linesize,
 752               AVCodecContext *avctx);
 753
 754 #define WRAPPER8_16(name8, name16)\
 755 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 756     return name8(s, dst           , src           , stride, h)\
 757           +name8(s, dst+8         , src+8         , stride, h);\
 758 }
 759
 760 #define WRAPPER8_16_SQ(name8, name16)\
 761 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 762     int score=0;\
 763     score +=name8(s, dst           , src           , stride, 8);\
 764     score +=name8(s, dst+8         , src+8         , stride, 8);\
 765     if(h==16){\
 766         dst += 8*stride;\
 767         src += 8*stride;\
 768         score +=name8(s, dst           , src           , stride, 8);\
 769         score +=name8(s, dst+8         , src+8         , stride, 8);\
 770     }\
 771     return score;\
 772 }
 773
 774
 775 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 776 {
 777     int i;
 778     for(i=0; i<h; i++)
 779     {
 780         AV_WN16(dst   , AV_RN16(src   ));
 781         dst+=dstStride;
 782         src+=srcStride;
 783     }
 784 }
 785
 786 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 787 {
 788     int i;
 789     for(i=0; i<h; i++)
 790     {
 791         AV_WN32(dst   , AV_RN32(src   ));
 792         dst+=dstStride;
 793         src+=srcStride;
 794     }
 795 }
 796
 797 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 798 {
 799     int i;
 800     for(i=0; i<h; i++)
 801     {
 802         AV_WN32(dst   , AV_RN32(src   ));
 803         AV_WN32(dst+4 , AV_RN32(src+4 ));
 804         dst+=dstStride;
 805         src+=srcStride;
 806     }
 807 }
 808
 809 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 810 {
 811     int i;
 812     for(i=0; i<h; i++)
 813     {
 814         AV_WN32(dst   , AV_RN32(src   ));
 815         AV_WN32(dst+4 , AV_RN32(src+4 ));
 816         dst[8]= src[8];
 817         dst+=dstStride;
 818         src+=srcStride;
 819     }
 820 }
 821
 822 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 823 {
 824     int i;
 825     for(i=0; i<h; i++)
 826     {
 827         AV_WN32(dst   , AV_RN32(src   ));
 828         AV_WN32(dst+4 , AV_RN32(src+4 ));
 829         AV_WN32(dst+8 , AV_RN32(src+8 ));
 830         AV_WN32(dst+12, AV_RN32(src+12));
 831         dst+=dstStride;
 832         src+=srcStride;
 833     }
 834 }
 835
 836 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 837 {
 838     int i;
 839     for(i=0; i<h; i++)
 840     {
 841         AV_WN32(dst   , AV_RN32(src   ));
 842         AV_WN32(dst+4 , AV_RN32(src+4 ));
 843         AV_WN32(dst+8 , AV_RN32(src+8 ));
 844         AV_WN32(dst+12, AV_RN32(src+12));
 845         dst[16]= src[16];
 846         dst+=dstStride;
 847         src+=srcStride;
 848     }
 849 }
 850
 851 #endif /* AVCODEC_DSPUTIL_H */