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_add_add_c(float *dst, const float *src0, const float *src1,
  70                               const float *src2, int src3, int blocksize, int step);
  71 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
  72                              const float *win, float add_bias, int len);
  73 void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
  74 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
  75
  76 /* encoding scans */
  77 extern const uint8_t ff_alternate_horizontal_scan[64];
  78 extern const uint8_t ff_alternate_vertical_scan[64];
  79 extern const uint8_t ff_zigzag_direct[64];
  80 extern const uint8_t ff_zigzag248_direct[64];
  81
  82 /* pixel operations */
  83 #define MAX_NEG_CROP 1024
  84
  85 /* temporary */
  86 extern uint32_t ff_squareTbl[512];
  87 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
  88
  89 /* VP3 DSP functions */
  90 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
  91 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  92 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  93
  94 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
  95 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
  96
  97 /* 1/2^n downscaling functions from imgconvert.c */
  98 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  99 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 100 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 101 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 102
 103 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
 104               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 105
 106 /* minimum alignment rules ;)
 107 If you notice errors in the align stuff, need more alignment for some ASM code
 108 for some CPU or need to use a function with less aligned data then send a mail
 109 to the ffmpeg-devel mailing list, ...
 110
 111 !warning These alignments might not match reality, (missing attribute((align))
 112 stuff somewhere possible).
 113 I (Michael) did not check them, these are just the alignments which I think
 114 could be reached easily ...
 115
 116 !future video codecs might need functions with less strict alignment
 117 */
 118
 119 /*
 120 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
 121 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
 122 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 123 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 124 void clear_blocks_c(DCTELEM *blocks);
 125 */
 126
 127 /* add and put pixel (decoding) */
 128 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 129 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
 130 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
 131 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);
 132 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 133 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 134 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
 135 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
 136
 137 #define DEF_OLD_QPEL(name)\
 138 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 139 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 140 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 141
 142 DEF_OLD_QPEL(qpel16_mc11_old_c)
 143 DEF_OLD_QPEL(qpel16_mc31_old_c)
 144 DEF_OLD_QPEL(qpel16_mc12_old_c)
 145 DEF_OLD_QPEL(qpel16_mc32_old_c)
 146 DEF_OLD_QPEL(qpel16_mc13_old_c)
 147 DEF_OLD_QPEL(qpel16_mc33_old_c)
 148 DEF_OLD_QPEL(qpel8_mc11_old_c)
 149 DEF_OLD_QPEL(qpel8_mc31_old_c)
 150 DEF_OLD_QPEL(qpel8_mc12_old_c)
 151 DEF_OLD_QPEL(qpel8_mc32_old_c)
 152 DEF_OLD_QPEL(qpel8_mc13_old_c)
 153 DEF_OLD_QPEL(qpel8_mc33_old_c)
 154
 155 #define CALL_2X_PIXELS(a, b, n)\
 156 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 157     b(block  , pixels  , line_size, h);\
 158     b(block+n, pixels+n, line_size, h);\
 159 }
 160
 161 /* motion estimation */
 162 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 163 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
 164 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))*/;
 165
 166
 167 // for snow slices
 168 typedef struct slice_buffer_s slice_buffer;
 169
 170 /**
 171  * Scantable.
 172  */
 173 typedef struct ScanTable{
 174     const uint8_t *scantable;
 175     uint8_t permutated[64];
 176     uint8_t raster_end[64];
 177 #if ARCH_PPC
 178                 /** Used by dct_quantize_altivec to find last-non-zero */
 179     DECLARE_ALIGNED(16, uint8_t, inverse[64]);
 180 #endif
 181 } ScanTable;
 182
 183 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
 184
 185 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
 186                          int block_w, int block_h,
 187                          int src_x, int src_y, int w, int h);
 188
 189 /**
 190  * DSPContext.
 191  */
 192 typedef struct DSPContext {
 193     /* pixel ops : interface with DCT */
 194     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 195     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 196     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 197     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 198     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 199     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
 200     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
 201     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
 202     /**
 203      * translational global motion compensation.
 204      */
 205     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 206     /**
 207      * global motion compensation.
 208      */
 209     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 210                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 211     void (*clear_block)(DCTELEM *block/*align 16*/);
 212     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 213     int (*pix_sum)(uint8_t * pix, int line_size);
 214     int (*pix_norm1)(uint8_t * pix, int line_size);
 215 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 216
 217     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
 218     me_cmp_func sse[5];
 219     me_cmp_func hadamard8_diff[5];
 220     me_cmp_func dct_sad[5];
 221     me_cmp_func quant_psnr[5];
 222     me_cmp_func bit[5];
 223     me_cmp_func rd[5];
 224     me_cmp_func vsad[5];
 225     me_cmp_func vsse[5];
 226     me_cmp_func nsse[5];
 227     me_cmp_func w53[5];
 228     me_cmp_func w97[5];
 229     me_cmp_func dct_max[5];
 230     me_cmp_func dct264_sad[5];
 231
 232     me_cmp_func me_pre_cmp[5];
 233     me_cmp_func me_cmp[5];
 234     me_cmp_func me_sub_cmp[5];
 235     me_cmp_func mb_cmp[5];
 236     me_cmp_func ildct_cmp[5]; //only width 16 used
 237     me_cmp_func frame_skip_cmp[5]; //only width 8 used
 238
 239     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
 240                              int size);
 241
 242     /**
 243      * Halfpel motion compensation with rounding (a+b+1)>>1.
 244      * this is an array[4][4] of motion compensation functions for 4
 245      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 246      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 247      * @param block destination where the result is stored
 248      * @param pixels source
 249      * @param line_size number of bytes in a horizontal line of block
 250      * @param h height
 251      */
 252     op_pixels_func put_pixels_tab[4][4];
 253
 254     /**
 255      * Halfpel motion compensation with rounding (a+b+1)>>1.
 256      * This is an array[4][4] of motion compensation functions for 4
 257      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 258      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 259      * @param block destination into which the result is averaged (a+b+1)>>1
 260      * @param pixels source
 261      * @param line_size number of bytes in a horizontal line of block
 262      * @param h height
 263      */
 264     op_pixels_func avg_pixels_tab[4][4];
 265
 266     /**
 267      * Halfpel motion compensation with no rounding (a+b)>>1.
 268      * this is an array[2][4] of motion compensation functions for 2
 269      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 270      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 271      * @param block destination where the result is stored
 272      * @param pixels source
 273      * @param line_size number of bytes in a horizontal line of block
 274      * @param h height
 275      */
 276     op_pixels_func put_no_rnd_pixels_tab[4][4];
 277
 278     /**
 279      * Halfpel motion compensation with no rounding (a+b)>>1.
 280      * this is an array[2][4] of motion compensation functions for 2
 281      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 282      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 283      * @param block destination into which the result is averaged (a+b)>>1
 284      * @param pixels source
 285      * @param line_size number of bytes in a horizontal line of block
 286      * @param h height
 287      */
 288     op_pixels_func avg_no_rnd_pixels_tab[4][4];
 289
 290     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);
 291
 292     /**
 293      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 294      * this is an array[12] of motion compensation functions for the 9 thirdpe
 295      * positions<br>
 296      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 297      * @param block destination where the result is stored
 298      * @param pixels source
 299      * @param line_size number of bytes in a horizontal line of block
 300      * @param h height
 301      */
 302     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 303     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 304
 305     qpel_mc_func put_qpel_pixels_tab[2][16];
 306     qpel_mc_func avg_qpel_pixels_tab[2][16];
 307     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 308     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 309     qpel_mc_func put_mspel_pixels_tab[8];
 310
 311     /**
 312      * h264 Chroma MC
 313      */
 314     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 315     /* This is really one func used in VC-1 decoding */
 316     h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
 317     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 318
 319     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
 320     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
 321
 322     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
 323     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
 324
 325     h264_weight_func weight_h264_pixels_tab[10];
 326     h264_biweight_func biweight_h264_pixels_tab[10];
 327
 328     /* AVS specific */
 329     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
 330     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
 331     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 332     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 333     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 334     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 335     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 336
 337     me_cmp_func pix_abs[2][4];
 338
 339     /* huffyuv specific */
 340     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 341     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
 342     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 343     /**
 344      * subtract huffyuv's variant of median prediction
 345      * note, this might read from src1[-1], src2[-1]
 346      */
 347     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
 348     /* this might write to dst[w] */
 349     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
 350     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
 351
 352     void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
 353     void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
 354     /* v/h_loop_filter_luma_intra: align 16 */
 355     void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 356     void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 357     void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
 358     void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
 359     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
 360     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
 361     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
 362     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],
 363                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
 364
 365     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 366     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 367
 368     void (*h261_loop_filter)(uint8_t *src, int stride);
 369
 370     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
 371     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
 372
 373     void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
 374     void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
 375
 376     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 377     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
 378     void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
 379     /* no alignment needed */
 380     void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
 381     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 382     void (*vector_fmul)(float *dst, const float *src, int len);
 383     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
 384     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
 385     void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
 386     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 387     void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
 388     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 389     void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
 390
 391     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
 392      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
 393     void (*float_to_int16)(int16_t *dst, const float *src, long len);
 394     void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
 395
 396     /* (I)DCT */
 397     void (*fdct)(DCTELEM *block/* align 16*/);
 398     void (*fdct248)(DCTELEM *block/* align 16*/);
 399
 400     /* IDCT really*/
 401     void (*idct)(DCTELEM *block/* align 16*/);
 402
 403     /**
 404      * block -> idct -> clip to unsigned 8 bit -> dest.
 405      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 406      * @param line_size size in bytes of a horizontal line of dest
 407      */
 408     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 409
 410     /**
 411      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 412      * @param line_size size in bytes of a horizontal line of dest
 413      */
 414     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 415
 416     /**
 417      * idct input permutation.
 418      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 419      * IDCT)
 420      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 421      * with the zigzag/alternate scan<br>
 422      * an example to avoid confusion:
 423      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 424      * - (x -> referece dct -> reference idct -> x)
 425      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 426      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 427      */
 428     uint8_t idct_permutation[64];
 429     int idct_permutation_type;
 430 #define FF_NO_IDCT_PERM 1
 431 #define FF_LIBMPEG2_IDCT_PERM 2
 432 #define FF_SIMPLE_IDCT_PERM 3
 433 #define FF_TRANSPOSE_IDCT_PERM 4
 434 #define FF_PARTTRANS_IDCT_PERM 5
 435 #define FF_SSE2_IDCT_PERM 6
 436
 437     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 438     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 439 #define BASIS_SHIFT 16
 440 #define RECON_SHIFT 6
 441
 442     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
 443 #define EDGE_WIDTH 16
 444
 445     /* h264 functions */
 446     /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them
 447        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
 448         The reason for above, is that no 2 out of one list may use a different permutation.
 449     */
 450     void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
 451     void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
 452     void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
 453     void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
 454     void (*h264_dct)(DCTELEM block[4][4]);
 455     void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 456     void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 457     void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 458     void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 459
 460     /* snow wavelet */
 461     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
 462     void (*horizontal_compose97i)(IDWTELEM *b, int width);
 463     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);
 464
 465     void (*prefetch)(void *mem, int stride, int h);
 466
 467     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 468
 469     /* vc1 functions */
 470     void (*vc1_inv_trans_8x8)(DCTELEM *b);
 471     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
 472     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
 473     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
 474     void (*vc1_v_overlap)(uint8_t* src, int stride);
 475     void (*vc1_h_overlap)(uint8_t* src, int stride);
 476     /* put 8x8 block with bicubic interpolation and quarterpel precision
 477      * last argument is actually round value instead of height
 478      */
 479     op_pixels_func put_vc1_mspel_pixels_tab[16];
 480
 481     /* intrax8 functions */
 482     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
 483     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
 484            int * range, int * sum,  int edges);
 485
 486     /* ape functions */
 487     /**
 488      * Add contents of the second vector to the first one.
 489      * @param len length of vectors, should be multiple of 16
 490      */
 491     void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
 492     /**
 493      * Add contents of the second vector to the first one.
 494      * @param len length of vectors, should be multiple of 16
 495      */
 496     void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
 497     /**
 498      * Calculate scalar product of two vectors.
 499      * @param len length of vectors, should be multiple of 16
 500      * @param shift number of bits to discard from product
 501      */
 502     int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
 503
 504     /* rv30 functions */
 505     qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
 506     qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
 507
 508     /* rv40 functions */
 509     qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
 510     qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
 511     h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
 512     h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
 513 } DSPContext;
 514
 515 void dsputil_static_init(void);
 516 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 517
 518 int ff_check_alignment(void);
 519
 520 /**
 521  * permute block according to permuatation.
 522  * @param last last non zero element in scantable order
 523  */
 524 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 525
 526 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 527
 528 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 529
 530 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 531 {
 532     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 533 }
 534
 535 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 536 {
 537     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 538 }
 539
 540 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 541     switch(type&0xFF){
 542     default:
 543     case FF_CMP_SAD:
 544         return lambda>>FF_LAMBDA_SHIFT;
 545     case FF_CMP_DCT:
 546         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 547     case FF_CMP_W53:
 548         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 549     case FF_CMP_W97:
 550         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 551     case FF_CMP_SATD:
 552     case FF_CMP_DCT264:
 553         return (2*lambda)>>FF_LAMBDA_SHIFT;
 554     case FF_CMP_RD:
 555     case FF_CMP_PSNR:
 556     case FF_CMP_SSE:
 557     case FF_CMP_NSSE:
 558         return lambda2>>FF_LAMBDA_SHIFT;
 559     case FF_CMP_BIT:
 560         return 1;
 561     }
 562 }
 563
 564 /**
 565  * Empty mmx state.
 566  * this must be called between any dsp function and float/double code.
 567  * for example sin(); dsp->idct_put(); emms_c(); cos()
 568  */
 569 #define emms_c()
 570
 571 /* should be defined by architectures supporting
 572    one or more MultiMedia extension */
 573 int mm_support(void);
 574
 575 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 576 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
 577 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
 578 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 579 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 580 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 581 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 582 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 583 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 584
 585 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
 586
 587 #if HAVE_MMX
 588
 589 #undef emms_c
 590
 591 extern int mm_flags;
 592
 593 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 594 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 595 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 596
 597 static inline void emms(void)
 598 {
 599     __asm__ volatile ("emms;":::"memory");
 600 }
 601
 602
 603 #define emms_c() \
 604 {\
 605     if (mm_flags & FF_MM_MMX)\
 606         emms();\
 607 }
 608
 609 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
 610
 611 #elif ARCH_ARM
 612
 613 extern int mm_flags;
 614
 615 #if HAVE_NEON
 616 #   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 617 #   define STRIDE_ALIGN 16
 618 #endif
 619
 620 #elif ARCH_PPC
 621
 622 extern int mm_flags;
 623
 624 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 625 #define STRIDE_ALIGN 16
 626
 627 #elif HAVE_MMI
 628
 629 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
 630 #define STRIDE_ALIGN 16
 631
 632 #else
 633
 634 #define mm_flags 0
 635 #define mm_support() 0
 636
 637 #endif
 638
 639 #ifndef DECLARE_ALIGNED_8
 640 #   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
 641 #endif
 642
 643 #ifndef STRIDE_ALIGN
 644 #   define STRIDE_ALIGN 8
 645 #endif
 646
 647 /* PSNR */
 648 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 649               int orig_linesize[3], int coded_linesize,
 650               AVCodecContext *avctx);
 651
 652 /* FFT computation */
 653
 654 /* NOTE: soon integer code will be added, so you must use the
 655    FFTSample type */
 656 typedef float FFTSample;
 657
 658 struct MDCTContext;
 659
 660 typedef struct FFTComplex {
 661     FFTSample re, im;
 662 } FFTComplex;
 663
 664 typedef struct FFTContext {
 665     int nbits;
 666     int inverse;
 667     uint16_t *revtab;
 668     FFTComplex *exptab;
 669     FFTComplex *exptab1; /* only used by SSE code */
 670     FFTComplex *tmp_buf;
 671     void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
 672     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 673     void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
 674     void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
 675 } FFTContext;
 676
 677 extern FFTSample* ff_cos_tabs[13];
 678
 679 /**
 680  * Sets up a complex FFT.
 681  * @param nbits           log2 of the length of the input array
 682  * @param inverse         if 0 perform the forward transform, if 1 perform the inverse
 683  */
 684 int ff_fft_init(FFTContext *s, int nbits, int inverse);
 685 void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
 686 void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
 687 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 688 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
 689 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
 690 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
 691 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
 692
 693 /**
 694  * Do the permutation needed BEFORE calling ff_fft_calc().
 695  */
 696 static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
 697 {
 698     s->fft_permute(s, z);
 699 }
 700 /**
 701  * Do a complex FFT with the parameters defined in ff_fft_init(). The
 702  * input data must be permuted before. No 1.0/sqrt(n) normalization is done.
 703  */
 704 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
 705 {
 706     s->fft_calc(s, z);
 707 }
 708 void ff_fft_end(FFTContext *s);
 709
 710 /* MDCT computation */
 711
 712 typedef struct MDCTContext {
 713     int n;  /* size of MDCT (i.e. number of input data * 2) */
 714     int nbits; /* n = 2^nbits */
 715     /* pre/post rotation tables */
 716     FFTSample *tcos;
 717     FFTSample *tsin;
 718     FFTContext fft;
 719 } MDCTContext;
 720
 721 static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
 722 {
 723     s->fft.imdct_calc(s, output, input);
 724 }
 725 static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
 726 {
 727     s->fft.imdct_half(s, output, input);
 728 }
 729
 730 /**
 731  * Generate a Kaiser-Bessel Derived Window.
 732  * @param   window  pointer to half window
 733  * @param   alpha   determines window shape
 734  * @param   n       size of half window
 735  */
 736 void ff_kbd_window_init(float *window, float alpha, int n);
 737
 738 /**
 739  * Generate a sine window.
 740  * @param   window  pointer to half window
 741  * @param   n       size of half window
 742  */
 743 void ff_sine_window_init(float *window, int n);
 744 extern float ff_sine_128 [ 128];
 745 extern float ff_sine_256 [ 256];
 746 extern float ff_sine_512 [ 512];
 747 extern float ff_sine_1024[1024];
 748 extern float ff_sine_2048[2048];
 749 extern float ff_sine_4096[4096];
 750 extern float *ff_sine_windows[6];
 751
 752 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
 753 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
 754 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
 755 void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
 756 void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
 757 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
 758 void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
 759 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
 760 void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
 761 void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input);
 762 void ff_mdct_end(MDCTContext *s);
 763
 764 /* Real Discrete Fourier Transform */
 765
 766 enum RDFTransformType {
 767     RDFT,
 768     IRDFT,
 769     RIDFT,
 770     IRIDFT,
 771 };
 772
 773 typedef struct {
 774     int nbits;
 775     int inverse;
 776     int sign_convention;
 777
 778     /* pre/post rotation tables */
 779     FFTSample *tcos;
 780     FFTSample *tsin;
 781     FFTContext fft;
 782 } RDFTContext;
 783
 784 /**
 785  * Sets up a real FFT.
 786  * @param nbits           log2 of the length of the input array
 787  * @param trans           the type of transform
 788  */
 789 int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans);
 790 void ff_rdft_calc(RDFTContext *s, FFTSample *data);
 791 void ff_rdft_end(RDFTContext *s);
 792
 793 #define WRAPPER8_16(name8, name16)\
 794 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 795     return name8(s, dst           , src           , stride, h)\
 796           +name8(s, dst+8         , src+8         , stride, h);\
 797 }
 798
 799 #define WRAPPER8_16_SQ(name8, name16)\
 800 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 801     int score=0;\
 802     score +=name8(s, dst           , src           , stride, 8);\
 803     score +=name8(s, dst+8         , src+8         , stride, 8);\
 804     if(h==16){\
 805         dst += 8*stride;\
 806         src += 8*stride;\
 807         score +=name8(s, dst           , src           , stride, 8);\
 808         score +=name8(s, dst+8         , src+8         , stride, 8);\
 809     }\
 810     return score;\
 811 }
 812
 813
 814 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 815 {
 816     int i;
 817     for(i=0; i<h; i++)
 818     {
 819         AV_WN16(dst   , AV_RN16(src   ));
 820         dst+=dstStride;
 821         src+=srcStride;
 822     }
 823 }
 824
 825 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 826 {
 827     int i;
 828     for(i=0; i<h; i++)
 829     {
 830         AV_WN32(dst   , AV_RN32(src   ));
 831         dst+=dstStride;
 832         src+=srcStride;
 833     }
 834 }
 835
 836 static inline void copy_block8(uint8_t *dst, 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         dst+=dstStride;
 844         src+=srcStride;
 845     }
 846 }
 847
 848 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 849 {
 850     int i;
 851     for(i=0; i<h; i++)
 852     {
 853         AV_WN32(dst   , AV_RN32(src   ));
 854         AV_WN32(dst+4 , AV_RN32(src+4 ));
 855         dst[8]= src[8];
 856         dst+=dstStride;
 857         src+=srcStride;
 858     }
 859 }
 860
 861 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 862 {
 863     int i;
 864     for(i=0; i<h; i++)
 865     {
 866         AV_WN32(dst   , AV_RN32(src   ));
 867         AV_WN32(dst+4 , AV_RN32(src+4 ));
 868         AV_WN32(dst+8 , AV_RN32(src+8 ));
 869         AV_WN32(dst+12, AV_RN32(src+12));
 870         dst+=dstStride;
 871         src+=srcStride;
 872     }
 873 }
 874
 875 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
 876 {
 877     int i;
 878     for(i=0; i<h; i++)
 879     {
 880         AV_WN32(dst   , AV_RN32(src   ));
 881         AV_WN32(dst+4 , AV_RN32(src+4 ));
 882         AV_WN32(dst+8 , AV_RN32(src+8 ));
 883         AV_WN32(dst+12, AV_RN32(src+12));
 884         dst[16]= src[16];
 885         dst+=dstStride;
 886         src+=srcStride;
 887     }
 888 }
 889
 890 #endif /* AVCODEC_DSPUTIL_H */