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 library is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This library is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with this library; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 /**
  22  * @file dsputil.h
  23  * DSP utils.
  24  * note, many functions in here may use MMX which trashes the FPU state, it is
  25  * absolutely necessary to call emms_c() between dsp & float/double code
  26  */
  27
  28 #ifndef DSPUTIL_H
  29 #define DSPUTIL_H
  30
  31 #include "common.h"
  32 #include "avcodec.h"
  33
  34
  35 //#define DEBUG
  36 /* dct code */
  37 typedef short DCTELEM;
  38 typedef int DWTELEM;
  39
  40 void fdct_ifast (DCTELEM *data);
  41 void fdct_ifast248 (DCTELEM *data);
  42 void ff_jpeg_fdct_islow (DCTELEM *data);
  43 void ff_fdct248_islow (DCTELEM *data);
  44
  45 void j_rev_dct (DCTELEM *data);
  46 void j_rev_dct4 (DCTELEM *data);
  47 void j_rev_dct2 (DCTELEM *data);
  48 void j_rev_dct1 (DCTELEM *data);
  49
  50 void ff_fdct_mmx(DCTELEM *block);
  51 void ff_fdct_mmx2(DCTELEM *block);
  52 void ff_fdct_sse2(DCTELEM *block);
  53
  54 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
  55 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
  56 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
  57 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
  58 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
  59 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
  60
  61 /* encoding scans */
  62 extern const uint8_t ff_alternate_horizontal_scan[64];
  63 extern const uint8_t ff_alternate_vertical_scan[64];
  64 extern const uint8_t ff_zigzag_direct[64];
  65 extern const uint8_t ff_zigzag248_direct[64];
  66
  67 /* pixel operations */
  68 #define MAX_NEG_CROP 1024
  69
  70 /* temporary */
  71 extern uint32_t squareTbl[512];
  72 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
  73
  74 /* VP3 DSP functions */
  75 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
  76 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  77 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  78
  79 /* 1/2^n downscaling functions from imgconvert.c */
  80 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  81 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  82 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  83 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
  84
  85 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
  86               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
  87
  88 /* minimum alignment rules ;)
  89 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
  90 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
  91
  92 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
  93 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
  94
  95 !future video codecs might need functions with less strict alignment
  96 */
  97
  98 /*
  99 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
 100 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
 101 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 102 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
 103 void clear_blocks_c(DCTELEM *blocks);
 104 */
 105
 106 /* add and put pixel (decoding) */
 107 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 108 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
 109 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
 110 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);
 111 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 112 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 113 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
 114 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
 115
 116 #define DEF_OLD_QPEL(name)\
 117 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 118 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 119 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 120
 121 DEF_OLD_QPEL(qpel16_mc11_old_c)
 122 DEF_OLD_QPEL(qpel16_mc31_old_c)
 123 DEF_OLD_QPEL(qpel16_mc12_old_c)
 124 DEF_OLD_QPEL(qpel16_mc32_old_c)
 125 DEF_OLD_QPEL(qpel16_mc13_old_c)
 126 DEF_OLD_QPEL(qpel16_mc33_old_c)
 127 DEF_OLD_QPEL(qpel8_mc11_old_c)
 128 DEF_OLD_QPEL(qpel8_mc31_old_c)
 129 DEF_OLD_QPEL(qpel8_mc12_old_c)
 130 DEF_OLD_QPEL(qpel8_mc32_old_c)
 131 DEF_OLD_QPEL(qpel8_mc13_old_c)
 132 DEF_OLD_QPEL(qpel8_mc33_old_c)
 133
 134 #define CALL_2X_PIXELS(a, b, n)\
 135 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 136     b(block  , pixels  , line_size, h);\
 137     b(block+n, pixels+n, line_size, h);\
 138 }
 139
 140 /* motion estimation */
 141 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 142 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
 143 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))*/;
 144
 145
 146 // for snow slices
 147 typedef struct slice_buffer_s slice_buffer;
 148
 149 /**
 150  * DSPContext.
 151  */
 152 typedef struct DSPContext {
 153     /* pixel ops : interface with DCT */
 154     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 155     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 156     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 157     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 158     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 159     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
 160     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
 161     /**
 162      * translational global motion compensation.
 163      */
 164     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 165     /**
 166      * global motion compensation.
 167      */
 168     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 169                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 170     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 171     int (*pix_sum)(uint8_t * pix, int line_size);
 172     int (*pix_norm1)(uint8_t * pix, int line_size);
 173 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 174
 175     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
 176     me_cmp_func sse[5];
 177     me_cmp_func hadamard8_diff[5];
 178     me_cmp_func dct_sad[5];
 179     me_cmp_func quant_psnr[5];
 180     me_cmp_func bit[5];
 181     me_cmp_func rd[5];
 182     me_cmp_func vsad[5];
 183     me_cmp_func vsse[5];
 184     me_cmp_func nsse[5];
 185     me_cmp_func w53[5];
 186     me_cmp_func w97[5];
 187     me_cmp_func dct_max[5];
 188     me_cmp_func dct264_sad[5];
 189
 190     me_cmp_func me_pre_cmp[5];
 191     me_cmp_func me_cmp[5];
 192     me_cmp_func me_sub_cmp[5];
 193     me_cmp_func mb_cmp[5];
 194     me_cmp_func ildct_cmp[5]; //only width 16 used
 195     me_cmp_func frame_skip_cmp[5]; //only width 8 used
 196
 197     /**
 198      * Halfpel motion compensation with rounding (a+b+1)>>1.
 199      * this is an array[4][4] of motion compensation funcions for 4
 200      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 201      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 202      * @param block destination where the result is stored
 203      * @param pixels source
 204      * @param line_size number of bytes in a horizontal line of block
 205      * @param h height
 206      */
 207     op_pixels_func put_pixels_tab[4][4];
 208
 209     /**
 210      * Halfpel motion compensation with rounding (a+b+1)>>1.
 211      * This is an array[4][4] of motion compensation functions for 4
 212      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 213      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 214      * @param block destination into which the result is averaged (a+b+1)>>1
 215      * @param pixels source
 216      * @param line_size number of bytes in a horizontal line of block
 217      * @param h height
 218      */
 219     op_pixels_func avg_pixels_tab[4][4];
 220
 221     /**
 222      * Halfpel motion compensation with no rounding (a+b)>>1.
 223      * this is an array[2][4] of motion compensation funcions for 2
 224      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 225      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 226      * @param block destination where the result is stored
 227      * @param pixels source
 228      * @param line_size number of bytes in a horizontal line of block
 229      * @param h height
 230      */
 231     op_pixels_func put_no_rnd_pixels_tab[4][4];
 232
 233     /**
 234      * Halfpel motion compensation with no rounding (a+b)>>1.
 235      * this is an array[2][4] of motion compensation funcions for 2
 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 into which the result is averaged (a+b)>>1
 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 avg_no_rnd_pixels_tab[4][4];
 244
 245     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);
 246
 247     /**
 248      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 249      * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
 250      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 251      * @param block destination where the result is stored
 252      * @param pixels source
 253      * @param line_size number of bytes in a horizontal line of block
 254      * @param h height
 255      */
 256     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 257     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 258
 259     qpel_mc_func put_qpel_pixels_tab[2][16];
 260     qpel_mc_func avg_qpel_pixels_tab[2][16];
 261     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 262     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 263     qpel_mc_func put_mspel_pixels_tab[8];
 264
 265     /**
 266      * h264 Chram MC
 267      */
 268     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 269     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 270
 271     qpel_mc_func put_h264_qpel_pixels_tab[4][16];
 272     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
 273
 274     h264_weight_func weight_h264_pixels_tab[10];
 275     h264_biweight_func biweight_h264_pixels_tab[10];
 276
 277     /* AVS specific */
 278     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
 279     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
 280     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 281     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 282     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 283     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
 284     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 285
 286     me_cmp_func pix_abs[2][4];
 287
 288     /* huffyuv specific */
 289     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 290     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 291     /**
 292      * subtract huffyuv's variant of median prediction
 293      * note, this might read from src1[-1], src2[-1]
 294      */
 295     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
 296     void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
 297
 298     void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 299     void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 300     void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 301     void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 302     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 303     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 304
 305     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 306     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 307
 308     void (*h261_loop_filter)(uint8_t *src, int stride);
 309
 310     /* (I)DCT */
 311     void (*fdct)(DCTELEM *block/* align 16*/);
 312     void (*fdct248)(DCTELEM *block/* align 16*/);
 313
 314     /* IDCT really*/
 315     void (*idct)(DCTELEM *block/* align 16*/);
 316
 317     /**
 318      * block -> idct -> clip to unsigned 8 bit -> dest.
 319      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 320      * @param line_size size in bytes of a horizotal line of dest
 321      */
 322     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 323
 324     /**
 325      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 326      * @param line_size size in bytes of a horizotal line of dest
 327      */
 328     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 329
 330     /**
 331      * idct input permutation.
 332      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 333      * IDCT)
 334      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 335      * with the zigzag/alternate scan<br>
 336      * an example to avoid confusion:
 337      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 338      * - (x -> referece dct -> reference idct -> x)
 339      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 340      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 341      */
 342     uint8_t idct_permutation[64];
 343     int idct_permutation_type;
 344 #define FF_NO_IDCT_PERM 1
 345 #define FF_LIBMPEG2_IDCT_PERM 2
 346 #define FF_SIMPLE_IDCT_PERM 3
 347 #define FF_TRANSPOSE_IDCT_PERM 4
 348 #define FF_PARTTRANS_IDCT_PERM 5
 349
 350     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 351     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 352 #define BASIS_SHIFT 16
 353 #define RECON_SHIFT 6
 354
 355     void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
 356     void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 357     void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
 358     void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
 359
 360     /* snow wavelet */
 361     void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
 362     void (*horizontal_compose97i)(DWTELEM *b, int width);
 363     void (*inner_add_yblock)(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);
 364
 365     void (*prefetch)(void *mem, int stride, int h);
 366
 367     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
 368 } DSPContext;
 369
 370 void dsputil_static_init(void);
 371 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 372
 373 /**
 374  * permute block according to permuatation.
 375  * @param last last non zero element in scantable order
 376  */
 377 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 378
 379 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 380
 381 #define         BYTE_VEC32(c)   ((c)*0x01010101UL)
 382
 383 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 384 {
 385     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 386 }
 387
 388 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 389 {
 390     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 391 }
 392
 393 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 394     switch(type&0xFF){
 395     default:
 396     case FF_CMP_SAD:
 397         return lambda>>FF_LAMBDA_SHIFT;
 398     case FF_CMP_DCT:
 399         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 400     case FF_CMP_W53:
 401         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 402     case FF_CMP_W97:
 403         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 404     case FF_CMP_SATD:
 405     case FF_CMP_DCT264:
 406         return (2*lambda)>>FF_LAMBDA_SHIFT;
 407     case FF_CMP_RD:
 408     case FF_CMP_PSNR:
 409     case FF_CMP_SSE:
 410     case FF_CMP_NSSE:
 411         return lambda2>>FF_LAMBDA_SHIFT;
 412     case FF_CMP_BIT:
 413         return 1;
 414     }
 415 }
 416
 417 /**
 418  * Empty mmx state.
 419  * this must be called between any dsp function and float/double code.
 420  * for example sin(); dsp->idct_put(); emms_c(); cos()
 421  */
 422 #define emms_c()
 423
 424 /* should be defined by architectures supporting
 425    one or more MultiMedia extension */
 426 int mm_support(void);
 427
 428 #ifdef __GNUC__
 429   #define DECLARE_ALIGNED_16(t,v)       t v __attribute__ ((aligned (16)))
 430 #else
 431   #define DECLARE_ALIGNED_16(t,v)      __declspec(align(16)) t v
 432 #endif
 433
 434 #if defined(HAVE_MMX)
 435
 436 #undef emms_c
 437
 438 #define MM_MMX    0x0001 /* standard MMX */
 439 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
 440 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
 441 #define MM_SSE    0x0008 /* SSE functions */
 442 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
 443 #define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
 444 #define MM_SSE3   0x0040 /* Prescott SSE3 functions */
 445
 446 extern int mm_flags;
 447
 448 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 449 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 450 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 451
 452 static inline void emms(void)
 453 {
 454     __asm __volatile ("emms;":::"memory");
 455 }
 456
 457
 458 #define emms_c() \
 459 {\
 460     if (mm_flags & MM_MMX)\
 461         emms();\
 462 }
 463
 464 #ifdef __GNUC__
 465   #define DECLARE_ALIGNED_8(t,v)       t v __attribute__ ((aligned (8)))
 466 #else
 467   #define DECLARE_ALIGNED_8(t,v)      __declspec(align(8)) t v
 468 #endif
 469
 470 #define STRIDE_ALIGN 8
 471
 472 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 473 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
 474
 475 #elif defined(ARCH_ARMV4L)
 476
 477 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
 478    line optimizations */
 479 #define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (4)))
 480 #define STRIDE_ALIGN 4
 481
 482 #define MM_IWMMXT    0x0100 /* XScale IWMMXT */
 483
 484 extern int mm_flags;
 485
 486 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
 487
 488 #elif defined(HAVE_MLIB)
 489
 490 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 491 #define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
 492 #define STRIDE_ALIGN 8
 493
 494 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 495
 496 #elif defined(ARCH_SPARC)
 497
 498 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 499 #define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
 500 #define STRIDE_ALIGN 8
 501 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 502
 503 #elif defined(ARCH_ALPHA)
 504
 505 #define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
 506 #define STRIDE_ALIGN 8
 507
 508 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 509
 510 #elif defined(ARCH_POWERPC)
 511
 512 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
 513
 514 extern int mm_flags;
 515
 516 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
 517 #define pixel altivec_pixel
 518 #include <altivec.h>
 519 #undef pixel
 520 #endif
 521
 522 #define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (16)))
 523 #define STRIDE_ALIGN 16
 524
 525 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 526
 527 #elif defined(HAVE_MMI)
 528
 529 #define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (16)))
 530 #define STRIDE_ALIGN 16
 531
 532 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 533
 534 #elif defined(ARCH_SH4)
 535
 536 #define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
 537 #define STRIDE_ALIGN 8
 538
 539 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 540
 541 #else
 542
 543 #define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
 544 #define STRIDE_ALIGN 8
 545
 546 #endif
 547
 548 #ifdef __GNUC__
 549
 550 struct unaligned_64 { uint64_t l; } __attribute__((packed));
 551 struct unaligned_32 { uint32_t l; } __attribute__((packed));
 552 struct unaligned_16 { uint16_t l; } __attribute__((packed));
 553
 554 #define LD16(a) (((const struct unaligned_16 *) (a))->l)
 555 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
 556 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
 557
 558 #define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
 559 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
 560
 561 #else /* __GNUC__ */
 562
 563 #define LD16(a) (*((uint16_t*)(a)))
 564 #define LD32(a) (*((uint32_t*)(a)))
 565 #define LD64(a) (*((uint64_t*)(a)))
 566
 567 #define ST16(a, b) *((uint16_t*)(a)) = (b)
 568 #define ST32(a, b) *((uint32_t*)(a)) = (b)
 569
 570 #endif /* !__GNUC__ */
 571
 572 /* PSNR */
 573 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 574               int orig_linesize[3], int coded_linesize,
 575               AVCodecContext *avctx);
 576
 577 /* FFT computation */
 578
 579 /* NOTE: soon integer code will be added, so you must use the
 580    FFTSample type */
 581 typedef float FFTSample;
 582
 583 typedef struct FFTComplex {
 584     FFTSample re, im;
 585 } FFTComplex;
 586
 587 typedef struct FFTContext {
 588     int nbits;
 589     int inverse;
 590     uint16_t *revtab;
 591     FFTComplex *exptab;
 592     FFTComplex *exptab1; /* only used by SSE code */
 593     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 594 } FFTContext;
 595
 596 int ff_fft_init(FFTContext *s, int nbits, int inverse);
 597 void ff_fft_permute(FFTContext *s, FFTComplex *z);
 598 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 599 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
 600 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
 601 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
 602 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
 603
 604 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
 605 {
 606     s->fft_calc(s, z);
 607 }
 608 void ff_fft_end(FFTContext *s);
 609
 610 /* MDCT computation */
 611
 612 typedef struct MDCTContext {
 613     int n;  /* size of MDCT (i.e. number of input data * 2) */
 614     int nbits; /* n = 2^nbits */
 615     /* pre/post rotation tables */
 616     FFTSample *tcos;
 617     FFTSample *tsin;
 618     FFTContext fft;
 619 } MDCTContext;
 620
 621 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
 622 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
 623                 const FFTSample *input, FFTSample *tmp);
 624 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
 625                const FFTSample *input, FFTSample *tmp);
 626 void ff_mdct_end(MDCTContext *s);
 627
 628 #define WARPER8_16(name8, name16)\
 629 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 630     return name8(s, dst           , src           , stride, h)\
 631           +name8(s, dst+8         , src+8         , stride, h);\
 632 }
 633
 634 #define WARPER8_16_SQ(name8, name16)\
 635 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 636     int score=0;\
 637     score +=name8(s, dst           , src           , stride, 8);\
 638     score +=name8(s, dst+8         , src+8         , stride, 8);\
 639     if(h==16){\
 640         dst += 8*stride;\
 641         src += 8*stride;\
 642         score +=name8(s, dst           , src           , stride, 8);\
 643         score +=name8(s, dst+8         , src+8         , stride, 8);\
 644     }\
 645     return score;\
 646 }
 647
 648 #endif