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 AVCODEC_DSPUTIL_H
  31 #define AVCODEC_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 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
  70
  71 /* encoding scans */
  72 extern const uint8_t ff_alternate_horizontal_scan[64];
  73 extern const uint8_t ff_alternate_vertical_scan[64];
  74 extern const uint8_t ff_zigzag_direct[64];
  75 extern const uint8_t ff_zigzag248_direct[64];
  76
  77 /* pixel operations */
  78 #define MAX_NEG_CROP 1024
  79
  80 /* temporary */
  81 extern uint32_t ff_squareTbl[512];
  82 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
  83
  84 /* VP3 DSP functions */
  85 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
  86 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  87 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  88
  89 /* 1/2^n downscaling functions from imgconvert.c */
  90 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  91 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  92 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  93 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  94
  95 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
  96               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
  97
  98 /* minimum alignment rules ;)
  99 If you notice errors in the align stuff, need more alignment for some ASM code
 100 for some CPU or need to use a function with less aligned data then send a mail
 101 to the ffmpeg-devel mailing list, ...
 102
 103 !warning These alignments might not match reality, (missing attribute((align))
 104 stuff somewhere possible).
 105 I (Michael) did not check them, these are just the alignments which I think
 106 could be reached easily ...
 107
 108 !future video codecs might need functions with less strict alignment
 109 */
 110
 111 /*
 112 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
 113 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
 114 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 115 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 116 void clear_blocks_c(DCTELEM *blocks);
 117 */
 118
 119 /* add and put pixel (decoding) */
 120 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 121 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
 122 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
 123 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);
 124 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 125 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 126 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
 127 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
 128
 129 #define DEF_OLD_QPEL(name)\
 130 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 131 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 132 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 133
 134 DEF_OLD_QPEL(qpel16_mc11_old_c)
 135 DEF_OLD_QPEL(qpel16_mc31_old_c)
 136 DEF_OLD_QPEL(qpel16_mc12_old_c)
 137 DEF_OLD_QPEL(qpel16_mc32_old_c)
 138 DEF_OLD_QPEL(qpel16_mc13_old_c)
 139 DEF_OLD_QPEL(qpel16_mc33_old_c)
 140 DEF_OLD_QPEL(qpel8_mc11_old_c)
 141 DEF_OLD_QPEL(qpel8_mc31_old_c)
 142 DEF_OLD_QPEL(qpel8_mc12_old_c)
 143 DEF_OLD_QPEL(qpel8_mc32_old_c)
 144 DEF_OLD_QPEL(qpel8_mc13_old_c)
 145 DEF_OLD_QPEL(qpel8_mc33_old_c)
 146
 147 #define CALL_2X_PIXELS(a, b, n)\
 148 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 149     b(block  , pixels  , line_size, h);\
 150     b(block+n, pixels+n, line_size, h);\
 151 }
 152
 153 /* motion estimation */
 154 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 155 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
 156 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))*/;
 157
 158
 159 // for snow slices
 160 typedef struct slice_buffer_s slice_buffer;
 161
 162 /**
 163  * Scantable.
 164  */
 165 typedef struct ScanTable{
 166     const uint8_t *scantable;
 167     uint8_t permutated[64];
 168     uint8_t raster_end[64];
 169 #ifdef ARCH_POWERPC
 170                 /** Used by dct_quantize_altivec to find last-non-zero */
 171     DECLARE_ALIGNED(16, uint8_t, inverse[64]);
 172 #endif
 173 } ScanTable;
 174
 175 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
 176
 177 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
 178                          int block_w, int block_h,
 179                          int src_x, int src_y, int w, int h);
 180
 181 /**
 182  * DSPContext.
 183  */
 184 typedef struct DSPContext {
 185     /* pixel ops : interface with DCT */
 186     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 187     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 188     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 189     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 190     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 191     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
 192     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
 193     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
 194     /**
 195      * translational global motion compensation.
 196      */
 197     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 198     /**
 199      * global motion compensation.
 200      */
 201     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 202                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 203     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 204     int (*pix_sum)(uint8_t * pix, int line_size);
 205     int (*pix_norm1)(uint8_t * pix, int line_size);
 206 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 207
 208     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
 209     me_cmp_func sse[5];
 210     me_cmp_func hadamard8_diff[5];
 211     me_cmp_func dct_sad[5];
 212     me_cmp_func quant_psnr[5];
 213     me_cmp_func bit[5];
 214     me_cmp_func rd[5];
 215     me_cmp_func vsad[5];
 216     me_cmp_func vsse[5];
 217     me_cmp_func nsse[5];
 218     me_cmp_func w53[5];
 219     me_cmp_func w97[5];
 220     me_cmp_func dct_max[5];
 221     me_cmp_func dct264_sad[5];
 222
 223     me_cmp_func me_pre_cmp[5];
 224     me_cmp_func me_cmp[5];
 225     me_cmp_func me_sub_cmp[5];
 226     me_cmp_func mb_cmp[5];
 227     me_cmp_func ildct_cmp[5]; //only width 16 used
 228     me_cmp_func frame_skip_cmp[5]; //only width 8 used
 229
 230     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
 231                              int size);
 232
 233     /**
 234      * Halfpel motion compensation with rounding (a+b+1)>>1.
 235      * this is an array[4][4] of motion compensation functions for 4
 236      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 237      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 238      * @param block destination where the result is stored
 239      * @param pixels source
 240      * @param line_size number of bytes in a horizontal line of block
 241      * @param h height
 242      */
 243     op_pixels_func put_pixels_tab[4][4];
 244
 245     /**
 246      * Halfpel motion compensation with rounding (a+b+1)>>1.
 247      * This is an array[4][4] of motion compensation functions for 4
 248      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 249      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 250      * @param block destination into which the result is averaged (a+b+1)>>1
 251      * @param pixels source
 252      * @param line_size number of bytes in a horizontal line of block
 253      * @param h height
 254      */
 255     op_pixels_func avg_pixels_tab[4][4];
 256
 257     /**
 258      * Halfpel motion compensation with no rounding (a+b)>>1.
 259      * this is an array[2][4] of motion compensation functions for 2
 260      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 261      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 262      * @param block destination where the result is stored
 263      * @param pixels source
 264      * @param line_size number of bytes in a horizontal line of block
 265      * @param h height
 266      */
 267     op_pixels_func put_no_rnd_pixels_tab[4][4];
 268
 269     /**
 270      * Halfpel motion compensation with no rounding (a+b)>>1.
 271      * this is an array[2][4] of motion compensation functions for 2
 272      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 273      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 274      * @param block destination into which the result is averaged (a+b)>>1
 275      * @param pixels source
 276      * @param line_size number of bytes in a horizontal line of block
 277      * @param h height
 278      */
 279     op_pixels_func avg_no_rnd_pixels_tab[4][4];
 280
 281     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);
 282
 283     /**
 284      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 285      * this is an array[12] of motion compensation functions for the 9 thirdpe
 286      * positions<br>
 287      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 288      * @param block destination where the result is stored
 289      * @param pixels source
 290      * @param line_size number of bytes in a horizontal line of block
 291      * @param h height
 292      */
 293     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 294     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 295
 296     qpel_mc_func put_qpel_pixels_tab[2][16];
 297     qpel_mc_func avg_qpel_pixels_tab[2][16];
 298     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 299     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 300     qpel_mc_func put_mspel_pixels_tab[8];
 301
 302     /**
 303      * h264 Chroma MC
 304      */
 305     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 306     /* This is really one func used in VC-1 decoding */
 307     h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
 308     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 309
 310     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
 311     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
 312
 313     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
 314     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
 315
 316     h264_weight_func weight_h264_pixels_tab[10];
 317     h264_biweight_func biweight_h264_pixels_tab[10];
 318
 319     /* AVS specific */
 320     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
 321     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
 322     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 323     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 324     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 325     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 326     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 327
 328     me_cmp_func pix_abs[2][4];
 329
 330     /* huffyuv specific */
 331     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 332     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
 333     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 334     /**
 335      * subtract huffyuv's variant of median prediction
 336      * note, this might read from src1[-1], src2[-1]
 337      */
 338     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
 339     /* this might write to dst[w] */
 340     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
 341     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
 342
 343     void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
 344     void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
 345     /* v/h_loop_filter_luma_intra: align 16 */
 346     void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
 347     void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
 348     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
 349     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
 350     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
 351     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],
 352                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
 353
 354     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 355     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 356
 357     void (*h261_loop_filter)(uint8_t *src, int stride);
 358
 359     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
 360     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
 361
 362     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 363     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
 364     void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
 365     /* no alignment needed */
 366     void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
 367     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 368     void (*vector_fmul)(float *dst, const float *src, int len);
 369     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
 370     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
 371     void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
 372     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 373     void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
 374     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 375     void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
 376
 377     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
 378      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
 379     void (*float_to_int16)(int16_t *dst, const float *src, long len);
 380     void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
 381
 382     /* (I)DCT */
 383     void (*fdct)(DCTELEM *block/* align 16*/);
 384     void (*fdct248)(DCTELEM *block/* align 16*/);
 385
 386     /* IDCT really*/
 387     void (*idct)(DCTELEM *block/* align 16*/);
 388
 389     /**
 390      * block -> idct -> clip to unsigned 8 bit -> dest.
 391      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 392      * @param line_size size in bytes of a horizontal line of dest
 393      */
 394     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 395
 396     /**
 397      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 398      * @param line_size size in bytes of a horizontal line of dest
 399      */
 400     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 401
 402     /**
 403      * idct input permutation.
 404      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 405      * IDCT)
 406      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 407      * with the zigzag/alternate scan<br>
 408      * an example to avoid confusion:
 409      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 410      * - (x -> referece dct -> reference idct -> x)
 411      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 412      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 413      */
 414     uint8_t idct_permutation[64];
 415     int idct_permutation_type;
 416 #define FF_NO_IDCT_PERM 1
 417 #define FF_LIBMPEG2_IDCT_PERM 2
 418 #define FF_SIMPLE_IDCT_PERM 3
 419 #define FF_TRANSPOSE_IDCT_PERM 4
 420 #define FF_PARTTRANS_IDCT_PERM 5
 421 #define FF_SSE2_IDCT_PERM 6
 422
 423     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 424     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 425 #define BASIS_SHIFT 16
 426 #define RECON_SHIFT 6
 427
 428     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
 429 #define EDGE_WIDTH 16
 430
 431     /* h264 functions */
 432     void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
 433     void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
 434     void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
 435     void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
 436     void (*h264_dct)(DCTELEM block[4][4]);
 437
 438     /* snow wavelet */
 439     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
 440     void (*horizontal_compose97i)(IDWTELEM *b, int width);
 441     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);
 442
 443     void (*prefetch)(void *mem, int stride, int h);
 444
 445     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 446
 447     /* vc1 functions */
 448     void (*vc1_inv_trans_8x8)(DCTELEM *b);
 449     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
 450     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
 451     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
 452     void (*vc1_v_overlap)(uint8_t* src, int stride);
 453     void (*vc1_h_overlap)(uint8_t* src, int stride);
 454     /* put 8x8 block with bicubic interpolation and quarterpel precision
 455      * last argument is actually round value instead of height
 456      */
 457     op_pixels_func put_vc1_mspel_pixels_tab[16];
 458
 459     /* intrax8 functions */
 460     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
 461     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
 462            int * range, int * sum,  int edges);
 463
 464     /* ape functions */
 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 (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
 470     /**
 471      * Add contents of the second vector to the first one.
 472      * @param len length of vectors, should be multiple of 16
 473      */
 474     void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
 475     /**
 476      * Calculate scalar product of two vectors.
 477      * @param len length of vectors, should be multiple of 16
 478      * @param shift number of bits to discard from product
 479      */
 480     int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
 481 } DSPContext;
 482
 483 void dsputil_static_init(void);
 484 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 485
 486 int ff_check_alignment(void);
 487
 488 /**
 489  * permute block according to permuatation.
 490  * @param last last non zero element in scantable order
 491  */
 492 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 493
 494 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 495
 496 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 497
 498 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 499 {
 500     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 501 }
 502
 503 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 504 {
 505     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 506 }
 507
 508 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 509     switch(type&0xFF){
 510     default:
 511     case FF_CMP_SAD:
 512         return lambda>>FF_LAMBDA_SHIFT;
 513     case FF_CMP_DCT:
 514         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 515     case FF_CMP_W53:
 516         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 517     case FF_CMP_W97:
 518         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 519     case FF_CMP_SATD:
 520     case FF_CMP_DCT264:
 521         return (2*lambda)>>FF_LAMBDA_SHIFT;
 522     case FF_CMP_RD:
 523     case FF_CMP_PSNR:
 524     case FF_CMP_SSE:
 525     case FF_CMP_NSSE:
 526         return lambda2>>FF_LAMBDA_SHIFT;
 527     case FF_CMP_BIT:
 528         return 1;
 529     }
 530 }
 531
 532 /**
 533  * Empty mmx state.
 534  * this must be called between any dsp function and float/double code.
 535  * for example sin(); dsp->idct_put(); emms_c(); cos()
 536  */
 537 #define emms_c()
 538
 539 /* should be defined by architectures supporting
 540    one or more MultiMedia extension */
 541 int mm_support(void);
 542
 543 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 544 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
 545 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
 546 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 547 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 548 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 549 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 550 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 551 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 552
 553 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
 554
 555 #if defined(HAVE_MMX)
 556
 557 #undef emms_c
 558
 559 #define MM_MMX    0x0001 /* standard MMX */
 560 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
 561 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
 562 #define MM_SSE    0x0008 /* SSE functions */
 563 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
 564 #define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
 565 #define MM_SSE3   0x0040 /* Prescott SSE3 functions */
 566 #define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
 567
 568 extern int mm_flags;
 569
 570 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 571 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 572 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 573
 574 static inline void emms(void)
 575 {
 576     asm volatile ("emms;":::"memory");
 577 }
 578
 579
 580 #define emms_c() \
 581 {\
 582     if (mm_flags & MM_MMX)\
 583         emms();\
 584 }
 585
 586 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
 587
 588 #elif defined(ARCH_ARMV4L)
 589
 590 #define MM_IWMMXT    0x0100 /* XScale IWMMXT */
 591
 592 extern int mm_flags;
 593
 594 #ifdef HAVE_NEON
 595 #   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 596 #   define STRIDE_ALIGN 16
 597 #endif
 598
 599 #elif defined(ARCH_POWERPC)
 600
 601 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
 602
 603 extern int mm_flags;
 604
 605 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 606 #define STRIDE_ALIGN 16
 607
 608 #elif defined(HAVE_MMI)
 609
 610 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 611 #define STRIDE_ALIGN 16
 612
 613 #else
 614
 615 #define mm_flags 0
 616 #define mm_support() 0
 617
 618 #endif
 619
 620 #ifndef DECLARE_ALIGNED_8
 621 #   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 622 #endif
 623
 624 #ifndef STRIDE_ALIGN
 625 #   define STRIDE_ALIGN 8
 626 #endif
 627
 628 /* PSNR */
 629 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 630               int orig_linesize[3], int coded_linesize,
 631               AVCodecContext *avctx);
 632
 633 /* FFT computation */
 634
 635 /* NOTE: soon integer code will be added, so you must use the
 636    FFTSample type */
 637 typedef float FFTSample;
 638
 639 struct MDCTContext;
 640
 641 typedef struct FFTComplex {
 642     FFTSample re, im;
 643 } FFTComplex;
 644
 645 typedef struct FFTContext {
 646     int nbits;
 647     int inverse;
 648     uint16_t *revtab;
 649     FFTComplex *exptab;
 650     FFTComplex *exptab1; /* only used by SSE code */
 651     FFTComplex *tmp_buf;
 652     void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
 653     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 654     void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
 655     void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
 656 } FFTContext;
 657
 658 int ff_fft_init(FFTContext *s, int nbits, int inverse);
 659 void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
 660 void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
 661 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 662 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
 663 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
 664 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
 665 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
 666
 667 static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
 668 {
 669     s->fft_permute(s, z);
 670 }
 671 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
 672 {
 673     s->fft_calc(s, z);
 674 }
 675 void ff_fft_end(FFTContext *s);
 676
 677 /* MDCT computation */
 678
 679 typedef struct MDCTContext {
 680     int n;  /* size of MDCT (i.e. number of input data * 2) */
 681     int nbits; /* n = 2^nbits */
 682     /* pre/post rotation tables */
 683     FFTSample *tcos;
 684     FFTSample *tsin;
 685     FFTContext fft;
 686 } MDCTContext;
 687
 688 static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
 689 {
 690     s->fft.imdct_calc(s, output, input);
 691 }
 692 static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
 693 {
 694     s->fft.imdct_half(s, output, input);
 695 }
 696
 697 /**
 698  * Generate a Kaiser-Bessel Derived Window.
 699  * @param   window  pointer to half window
 700  * @param   alpha   determines window shape
 701  * @param   n       size of half window
 702  */
 703 void ff_kbd_window_init(float *window, float alpha, int n);
 704
 705 /**
 706  * Generate a sine window.
 707  * @param   window  pointer to half window
 708  * @param   n       size of half window
 709  */
 710 void ff_sine_window_init(float *window, int n);
 711 extern float ff_sine_128 [ 128];
 712 extern float ff_sine_256 [ 256];
 713 extern float ff_sine_512 [ 512];
 714 extern float ff_sine_1024[1024];
 715 extern float ff_sine_2048[2048];
 716 extern float *ff_sine_windows[5];
 717
 718 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
 719 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
 720 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
 721 void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
 722 void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
 723 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
 724 void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
 725 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
 726 void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
 727 void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input);
 728 void ff_mdct_end(MDCTContext *s);
 729
 730 #define WRAPPER8_16(name8, name16)\
 731 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 732     return name8(s, dst           , src           , stride, h)\
 733           +name8(s, dst+8         , src+8         , stride, h);\
 734 }
 735
 736 #define WRAPPER8_16_SQ(name8, name16)\
 737 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 738     int score=0;\
 739     score +=name8(s, dst           , src           , stride, 8);\
 740     score +=name8(s, dst+8         , src+8         , stride, 8);\
 741     if(h==16){\
 742         dst += 8*stride;\
 743         src += 8*stride;\
 744         score +=name8(s, dst           , src           , stride, 8);\
 745         score +=name8(s, dst+8         , src+8         , stride, 8);\
 746     }\
 747     return score;\
 748 }
 749
 750
 751 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 752 {
 753     int i;
 754     for(i=0; i<h; i++)
 755     {
 756         AV_WN16(dst   , AV_RN16(src   ));
 757         dst+=dstStride;
 758         src+=srcStride;
 759     }
 760 }
 761
 762 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 763 {
 764     int i;
 765     for(i=0; i<h; i++)
 766     {
 767         AV_WN32(dst   , AV_RN32(src   ));
 768         dst+=dstStride;
 769         src+=srcStride;
 770     }
 771 }
 772
 773 static inline void copy_block8(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         dst+=dstStride;
 781         src+=srcStride;
 782     }
 783 }
 784
 785 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 786 {
 787     int i;
 788     for(i=0; i<h; i++)
 789     {
 790         AV_WN32(dst   , AV_RN32(src   ));
 791         AV_WN32(dst+4 , AV_RN32(src+4 ));
 792         dst[8]= src[8];
 793         dst+=dstStride;
 794         src+=srcStride;
 795     }
 796 }
 797
 798 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 799 {
 800     int i;
 801     for(i=0; i<h; i++)
 802     {
 803         AV_WN32(dst   , AV_RN32(src   ));
 804         AV_WN32(dst+4 , AV_RN32(src+4 ));
 805         AV_WN32(dst+8 , AV_RN32(src+8 ));
 806         AV_WN32(dst+12, AV_RN32(src+12));
 807         dst+=dstStride;
 808         src+=srcStride;
 809     }
 810 }
 811
 812 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 813 {
 814     int i;
 815     for(i=0; i<h; i++)
 816     {
 817         AV_WN32(dst   , AV_RN32(src   ));
 818         AV_WN32(dst+4 , AV_RN32(src+4 ));
 819         AV_WN32(dst+8 , AV_RN32(src+8 ));
 820         AV_WN32(dst+12, AV_RN32(src+12));
 821         dst[16]= src[16];
 822         dst+=dstStride;
 823         src+=srcStride;
 824     }
 825 }
 826
 827 #endif /* AVCODEC_DSPUTIL_H */