apps/codecs/libwmapro/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
  41 void fdct_ifast (DCTELEM *data);
  42 void fdct_ifast248 (DCTELEM *data);
  43 void ff_jpeg_fdct_islow (DCTELEM *data);
  44 void ff_fdct248_islow (DCTELEM *data);
  45
  46 void j_rev_dct (DCTELEM *data);
  47 void j_rev_dct4 (DCTELEM *data);
  48 void j_rev_dct2 (DCTELEM *data);
  49 void j_rev_dct1 (DCTELEM *data);
  50 void ff_wmv2_idct_c(DCTELEM *data);
  51
  52 void ff_fdct_mmx(DCTELEM *block);
  53 void ff_fdct_mmx2(DCTELEM *block);
  54 void ff_fdct_sse2(DCTELEM *block);
  55
  56 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
  57 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
  58 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
  59 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
  60 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
  61 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
  62 void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  63 void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  64 void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  65 void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  66
  67 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
  68                              const float *win, float add_bias, int len);
  69 void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
  70 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
  71
  72 /* encoding scans */
  73 extern const uint8_t ff_alternate_horizontal_scan[64];
  74 extern const uint8_t ff_alternate_vertical_scan[64];
  75 extern const uint8_t ff_zigzag_direct[64];
  76 extern const uint8_t ff_zigzag248_direct[64];
  77
  78 /* pixel operations */
  79 #define MAX_NEG_CROP 1024
  80
  81 /* temporary */
  82 extern uint32_t ff_squareTbl[512];
  83 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
  84
  85 /* VP3 DSP functions */
  86 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
  87 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  88 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  89
  90 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
  91 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
  92
  93 /* VP6 DSP functions */
  94 void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride,
  95                            const int16_t *h_weights, const int16_t *v_weights);
  96
  97 /* Bink functions */
  98 void ff_bink_idct_c    (DCTELEM *block);
  99 void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
 100 void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
 101
 102 /* CAVS functions */
 103 void ff_put_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride);
 104 void ff_avg_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride);
 105 void ff_put_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride);
 106 void ff_avg_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride);
 107
 108 /* VC1 functions */
 109 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
 110 void ff_avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
 111
 112 /* EA functions */
 113 void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
 114
 115 /* 1/2^n downscaling functions from imgconvert.c */
 116 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 117 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 118 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 119 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 120
 121 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
 122               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 123
 124 /* minimum alignment rules ;)
 125 If you notice errors in the align stuff, need more alignment for some ASM code
 126 for some CPU or need to use a function with less aligned data then send a mail
 127 to the ffmpeg-devel mailing list, ...
 128
 129 !warning These alignments might not match reality, (missing attribute((align))
 130 stuff somewhere possible).
 131 I (Michael) did not check them, these are just the alignments which I think
 132 could be reached easily ...
 133
 134 !future video codecs might need functions with less strict alignment
 135 */
 136
 137 /*
 138 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
 139 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
 140 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 141 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 142 void clear_blocks_c(DCTELEM *blocks);
 143 */
 144
 145 /* add and put pixel (decoding) */
 146 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 147 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
 148 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
 149 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);
 150 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 151 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 152
 153 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
 154
 155 #define DEF_OLD_QPEL(name)\
 156 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 157 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 158 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 159
 160 DEF_OLD_QPEL(qpel16_mc11_old_c)
 161 DEF_OLD_QPEL(qpel16_mc31_old_c)
 162 DEF_OLD_QPEL(qpel16_mc12_old_c)
 163 DEF_OLD_QPEL(qpel16_mc32_old_c)
 164 DEF_OLD_QPEL(qpel16_mc13_old_c)
 165 DEF_OLD_QPEL(qpel16_mc33_old_c)
 166 DEF_OLD_QPEL(qpel8_mc11_old_c)
 167 DEF_OLD_QPEL(qpel8_mc31_old_c)
 168 DEF_OLD_QPEL(qpel8_mc12_old_c)
 169 DEF_OLD_QPEL(qpel8_mc32_old_c)
 170 DEF_OLD_QPEL(qpel8_mc13_old_c)
 171 DEF_OLD_QPEL(qpel8_mc33_old_c)
 172
 173 #define CALL_2X_PIXELS(a, b, n)\
 174 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 175     b(block  , pixels  , line_size, h);\
 176     b(block+n, pixels+n, line_size, h);\
 177 }
 178
 179 /* motion estimation */
 180 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 181 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
 182 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))*/;
 183
 184 /**
 185  * Scantable.
 186  */
 187 typedef struct ScanTable{
 188     const uint8_t *scantable;
 189     uint8_t permutated[64];
 190     uint8_t raster_end[64];
 191 #if ARCH_PPC
 192                 /** Used by dct_quantize_altivec to find last-non-zero */
 193     DECLARE_ALIGNED(16, uint8_t, inverse)[64];
 194 #endif
 195 } ScanTable;
 196
 197 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
 198
 199 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
 200                          int block_w, int block_h,
 201                          int src_x, int src_y, int w, int h);
 202
 203 /**
 204  * DSPContext.
 205  */
 206 typedef struct DSPContext {
 207     /* pixel ops : interface with DCT */
 208     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 209     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 210     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 211     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 212     void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 213     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 214     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
 215     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
 216     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
 217     /**
 218      * translational global motion compensation.
 219      */
 220     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 221     /**
 222      * global motion compensation.
 223      */
 224     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 225                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 226     void (*clear_block)(DCTELEM *block/*align 16*/);
 227     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 228     int (*pix_sum)(uint8_t * pix, int line_size);
 229     int (*pix_norm1)(uint8_t * pix, int line_size);
 230 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 231
 232     me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
 233     me_cmp_func sse[6];
 234     me_cmp_func hadamard8_diff[6];
 235     me_cmp_func dct_sad[6];
 236     me_cmp_func quant_psnr[6];
 237     me_cmp_func bit[6];
 238     me_cmp_func rd[6];
 239     me_cmp_func vsad[6];
 240     me_cmp_func vsse[6];
 241     me_cmp_func nsse[6];
 242     me_cmp_func w53[6];
 243     me_cmp_func w97[6];
 244     me_cmp_func dct_max[6];
 245     me_cmp_func dct264_sad[6];
 246
 247     me_cmp_func me_pre_cmp[6];
 248     me_cmp_func me_cmp[6];
 249     me_cmp_func me_sub_cmp[6];
 250     me_cmp_func mb_cmp[6];
 251     me_cmp_func ildct_cmp[6]; //only width 16 used
 252     me_cmp_func frame_skip_cmp[6]; //only width 8 used
 253
 254     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
 255                              int size);
 256
 257     /**
 258      * Halfpel motion compensation with rounding (a+b+1)>>1.
 259      * this is an array[4][4] of motion compensation functions for 4
 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_pixels_tab[4][4];
 268
 269     /**
 270      * Halfpel motion compensation with rounding (a+b+1)>>1.
 271      * This is an array[4][4] of motion compensation functions for 4
 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)>>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_pixels_tab[4][4];
 280
 281     /**
 282      * Halfpel motion compensation with no rounding (a+b)>>1.
 283      * this is an array[2][4] of motion compensation functions for 2
 284      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 285      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 286      * @param block destination where the result is stored
 287      * @param pixels source
 288      * @param line_size number of bytes in a horizontal line of block
 289      * @param h height
 290      */
 291     op_pixels_func put_no_rnd_pixels_tab[4][4];
 292
 293     /**
 294      * Halfpel motion compensation with no rounding (a+b)>>1.
 295      * this is an array[2][4] of motion compensation functions for 2
 296      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 297      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 298      * @param block destination into which the result is averaged (a+b)>>1
 299      * @param pixels source
 300      * @param line_size number of bytes in a horizontal line of block
 301      * @param h height
 302      */
 303     op_pixels_func avg_no_rnd_pixels_tab[4][4];
 304
 305     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);
 306
 307     /**
 308      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 309      * this is an array[12] of motion compensation functions for the 9 thirdpe
 310      * positions<br>
 311      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 312      * @param block destination where the result is stored
 313      * @param pixels source
 314      * @param line_size number of bytes in a horizontal line of block
 315      * @param h height
 316      */
 317     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 318     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 319
 320     qpel_mc_func put_qpel_pixels_tab[2][16];
 321     qpel_mc_func avg_qpel_pixels_tab[2][16];
 322     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 323     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 324     qpel_mc_func put_mspel_pixels_tab[8];
 325
 326     /**
 327      * h264 Chroma MC
 328      */
 329     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 330     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 331     /* This is really one func used in VC-1 decoding */
 332     h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
 333     h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
 334
 335     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
 336     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
 337
 338     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
 339     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
 340
 341     /* AVS specific */
 342     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
 343     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
 344     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 345     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 346     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 347     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 348     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 349
 350     me_cmp_func pix_abs[2][4];
 351
 352     /* huffyuv specific */
 353     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 354     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
 355     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 356     /**
 357      * subtract huffyuv's variant of median prediction
 358      * note, this might read from src1[-1], src2[-1]
 359      */
 360     void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
 361     void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
 362     int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
 363     void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
 364     /* this might write to dst[w] */
 365     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
 366     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
 367
 368     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 369     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 370
 371     void (*h261_loop_filter)(uint8_t *src, int stride);
 372
 373     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
 374     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
 375
 376     void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
 377     void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
 378
 379     void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride,
 380                              const int16_t *h_weights,const int16_t *v_weights);
 381
 382     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 383     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
 384     void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
 385     /* no alignment needed */
 386     void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
 387     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 388     void (*vector_fmul)(float *dst, const float *src, int len);
 389     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
 390     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
 391     void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
 392     /* assume len is a multiple of 4, and arrays are 16-byte aligned */
 393     void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
 394     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
 395     void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
 396     void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
 397     /**
 398      * Multiply a vector of floats by a scalar float.  Source and
 399      * destination vectors must overlap exactly or not at all.
 400      * @param dst result vector, 16-byte aligned
 401      * @param src input vector, 16-byte aligned
 402      * @param mul scalar value
 403      * @param len length of vector, multiple of 4
 404      */
 405     void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
 406                                int len);
 407     /**
 408      * Multiply a vector of floats by concatenated short vectors of
 409      * floats and by a scalar float.  Source and destination vectors
 410      * must overlap exactly or not at all.
 411      * [0]: short vectors of length 2, 8-byte aligned
 412      * [1]: short vectors of length 4, 16-byte aligned
 413      * @param dst output vector, 16-byte aligned
 414      * @param src input vector, 16-byte aligned
 415      * @param sv  array of pointers to short vectors
 416      * @param mul scalar value
 417      * @param len number of elements in src and dst, multiple of 4
 418      */
 419     void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
 420                                      const float **sv, float mul, int len);
 421     /**
 422      * Multiply short vectors of floats by a scalar float, store
 423      * concatenated result.
 424      * [0]: short vectors of length 2, 8-byte aligned
 425      * [1]: short vectors of length 4, 16-byte aligned
 426      * @param dst output vector, 16-byte aligned
 427      * @param sv  array of pointers to short vectors
 428      * @param mul scalar value
 429      * @param len number of output elements, multiple of 4
 430      */
 431     void (*sv_fmul_scalar[2])(float *dst, const float **sv,
 432                               float mul, int len);
 433     /**
 434      * Calculate the scalar product of two vectors of floats.
 435      * @param v1  first vector, 16-byte aligned
 436      * @param v2  second vector, 16-byte aligned
 437      * @param len length of vectors, multiple of 4
 438      */
 439     float (*scalarproduct_float)(const float *v1, const float *v2, int len);
 440     /**
 441      * Calculate the sum and difference of two vectors of floats.
 442      * @param v1  first input vector, sum output, 16-byte aligned
 443      * @param v2  second input vector, difference output, 16-byte aligned
 444      * @param len length of vectors, multiple of 4
 445      */
 446     void (*butterflies_float)(float * v1, float * v2, int len);
 447
 448     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
 449      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
 450     void (*float_to_int16)(int16_t *dst, const float *src, long len);
 451     void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
 452
 453     /* (I)DCT */
 454     void (*fdct)(DCTELEM *block/* align 16*/);
 455     void (*fdct248)(DCTELEM *block/* align 16*/);
 456
 457     /* IDCT really*/
 458     void (*idct)(DCTELEM *block/* align 16*/);
 459
 460     /**
 461      * block -> idct -> clip to unsigned 8 bit -> dest.
 462      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 463      * @param line_size size in bytes of a horizontal line of dest
 464      */
 465     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 466
 467     /**
 468      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 469      * @param line_size size in bytes of a horizontal line of dest
 470      */
 471     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 472
 473     /**
 474      * idct input permutation.
 475      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 476      * IDCT)
 477      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 478      * with the zigzag/alternate scan<br>
 479      * an example to avoid confusion:
 480      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 481      * - (x -> referece dct -> reference idct -> x)
 482      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 483      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 484      */
 485     uint8_t idct_permutation[64];
 486     int idct_permutation_type;
 487 #define FF_NO_IDCT_PERM 1
 488 #define FF_LIBMPEG2_IDCT_PERM 2
 489 #define FF_SIMPLE_IDCT_PERM 3
 490 #define FF_TRANSPOSE_IDCT_PERM 4
 491 #define FF_PARTTRANS_IDCT_PERM 5
 492 #define FF_SSE2_IDCT_PERM 6
 493
 494     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 495     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 496 #define BASIS_SHIFT 16
 497 #define RECON_SHIFT 6
 498
 499     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
 500 #define EDGE_WIDTH 16
 501
 502     void (*prefetch)(void *mem, int stride, int h);
 503
 504     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 505
 506     /* mlp/truehd functions */
 507     void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
 508                                int firorder, int iirorder,
 509                                unsigned int filter_shift, int32_t mask, int blocksize,
 510                                int32_t *sample_buffer);
 511
 512     /* vc1 functions */
 513     void (*vc1_inv_trans_8x8)(DCTELEM *b);
 514     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
 515     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
 516     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
 517     void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 518     void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 519     void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 520     void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
 521     void (*vc1_v_overlap)(uint8_t* src, int stride);
 522     void (*vc1_h_overlap)(uint8_t* src, int stride);
 523     void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
 524     void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
 525     void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
 526     void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
 527     void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
 528     void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
 529     /* put 8x8 block with bicubic interpolation and quarterpel precision
 530      * last argument is actually round value instead of height
 531      */
 532     op_pixels_func put_vc1_mspel_pixels_tab[16];
 533     op_pixels_func avg_vc1_mspel_pixels_tab[16];
 534
 535     /* intrax8 functions */
 536     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
 537     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
 538            int * range, int * sum,  int edges);
 539
 540     /**
 541      * Calculate scalar product of two vectors.
 542      * @param len length of vectors, should be multiple of 16
 543      * @param shift number of bits to discard from product
 544      */
 545     int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
 546     /* ape functions */
 547     /**
 548      * Calculate scalar product of v1 and v2,
 549      * and v1[i] += v3[i] * mul
 550      * @param len length of vectors, should be multiple of 16
 551      */
 552     int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, int16_t *v2, int16_t *v3, int len, int mul);
 553
 554     /* rv30 functions */
 555     qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
 556     qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
 557
 558     /* rv40 functions */
 559     qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
 560     qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
 561     h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
 562     h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
 563
 564     /* bink functions */
 565     op_fill_func fill_block_tab[2];
 566     void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
 567 } DSPContext;
 568
 569 void dsputil_static_init(void);
 570 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 571
 572 int ff_check_alignment(void);
 573
 574 /**
 575  * permute block according to permuatation.
 576  * @param last last non zero element in scantable order
 577  */
 578 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 579
 580 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 581
 582 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 583
 584 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 585 {
 586     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 587 }
 588
 589 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 590 {
 591     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 592 }
 593
 594 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 595     switch(type&0xFF){
 596     default:
 597     case FF_CMP_SAD:
 598         return lambda>>FF_LAMBDA_SHIFT;
 599     case FF_CMP_DCT:
 600         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 601     case FF_CMP_W53:
 602         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 603     case FF_CMP_W97:
 604         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 605     case FF_CMP_SATD:
 606     case FF_CMP_DCT264:
 607         return (2*lambda)>>FF_LAMBDA_SHIFT;
 608     case FF_CMP_RD:
 609     case FF_CMP_PSNR:
 610     case FF_CMP_SSE:
 611     case FF_CMP_NSSE:
 612         return lambda2>>FF_LAMBDA_SHIFT;
 613     case FF_CMP_BIT:
 614         return 1;
 615     }
 616 }
 617
 618 /**
 619  * Empty mmx state.
 620  * this must be called between any dsp function and float/double code.
 621  * for example sin(); dsp->idct_put(); emms_c(); cos()
 622  */
 623 #define emms_c()
 624
 625 /* should be defined by architectures supporting
 626    one or more MultiMedia extension */
 627 int mm_support(void);
 628 extern int mm_flags;
 629
 630 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 631 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
 632 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
 633 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 634 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 635 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 636 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 637 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 638 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 639
 640 void ff_dsputil_init_dwt(DSPContext *c);
 641 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx);
 642 void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
 643 void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
 644 void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
 645 void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
 646 void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
 647 void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
 648
 649 #if HAVE_MMX
 650
 651 #undef emms_c
 652
 653 static inline void emms(void)
 654 {
 655     __asm__ volatile ("emms;":::"memory");
 656 }
 657
 658
 659 #define emms_c() \
 660 {\
 661     if (mm_flags & FF_MM_MMX)\
 662         emms();\
 663 }
 664
 665 #elif ARCH_ARM
 666
 667 #if HAVE_NEON
 668 #   define STRIDE_ALIGN 16
 669 #endif
 670
 671 #elif ARCH_PPC
 672
 673 #define STRIDE_ALIGN 16
 674
 675 #elif HAVE_MMI
 676
 677 #define STRIDE_ALIGN 16
 678
 679 #else
 680
 681 #define mm_flags 0
 682 #define mm_support() 0
 683
 684 #endif
 685
 686 #ifndef STRIDE_ALIGN
 687 #   define STRIDE_ALIGN 8
 688 #endif
 689
 690 #define LOCAL_ALIGNED(a, t, v, s, ...)                          \
 691     uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)];              \
 692     t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a)
 693
 694 #if HAVE_LOCAL_ALIGNED_8
 695 #   define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__
 696 #else
 697 #   define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__)
 698 #endif
 699
 700 #if HAVE_LOCAL_ALIGNED_16
 701 #   define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__
 702 #else
 703 #   define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__)
 704 #endif
 705
 706 /* PSNR */
 707 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 708               int orig_linesize[3], int coded_linesize,
 709               AVCodecContext *avctx);
 710
 711 #define WRAPPER8_16(name8, name16)\
 712 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 713     return name8(s, dst           , src           , stride, h)\
 714           +name8(s, dst+8         , src+8         , stride, h);\
 715 }
 716
 717 #define WRAPPER8_16_SQ(name8, name16)\
 718 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 719     int score=0;\
 720     score +=name8(s, dst           , src           , stride, 8);\
 721     score +=name8(s, dst+8         , src+8         , stride, 8);\
 722     if(h==16){\
 723         dst += 8*stride;\
 724         src += 8*stride;\
 725         score +=name8(s, dst           , src           , stride, 8);\
 726         score +=name8(s, dst+8         , src+8         , stride, 8);\
 727     }\
 728     return score;\
 729 }
 730
 731
 732 static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 733 {
 734     int i;
 735     for(i=0; i<h; i++)
 736     {
 737         AV_WN16(dst   , AV_RN16(src   ));
 738         dst+=dstStride;
 739         src+=srcStride;
 740     }
 741 }
 742
 743 static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 744 {
 745     int i;
 746     for(i=0; i<h; i++)
 747     {
 748         AV_WN32(dst   , AV_RN32(src   ));
 749         dst+=dstStride;
 750         src+=srcStride;
 751     }
 752 }
 753
 754 static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 755 {
 756     int i;
 757     for(i=0; i<h; i++)
 758     {
 759         AV_WN32(dst   , AV_RN32(src   ));
 760         AV_WN32(dst+4 , AV_RN32(src+4 ));
 761         dst+=dstStride;
 762         src+=srcStride;
 763     }
 764 }
 765
 766 static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 767 {
 768     int i;
 769     for(i=0; i<h; i++)
 770     {
 771         AV_WN32(dst   , AV_RN32(src   ));
 772         AV_WN32(dst+4 , AV_RN32(src+4 ));
 773         dst[8]= src[8];
 774         dst+=dstStride;
 775         src+=srcStride;
 776     }
 777 }
 778
 779 static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 780 {
 781     int i;
 782     for(i=0; i<h; i++)
 783     {
 784         AV_WN32(dst   , AV_RN32(src   ));
 785         AV_WN32(dst+4 , AV_RN32(src+4 ));
 786         AV_WN32(dst+8 , AV_RN32(src+8 ));
 787         AV_WN32(dst+12, AV_RN32(src+12));
 788         dst+=dstStride;
 789         src+=srcStride;
 790     }
 791 }
 792
 793 static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
 794 {
 795     int i;
 796     for(i=0; i<h; i++)
 797     {
 798         AV_WN32(dst   , AV_RN32(src   ));
 799         AV_WN32(dst+4 , AV_RN32(src+4 ));
 800         AV_WN32(dst+8 , AV_RN32(src+8 ));
 801         AV_WN32(dst+12, AV_RN32(src+12));
 802         dst[16]= src[16];
 803         dst+=dstStride;
 804         src+=srcStride;
 805     }
 806 }
 807
 808 #endif /* AVCODEC_DSPUTIL_H */