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., 59 Temple Place, Suite 330, Boston, MA  02111-1307  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
  39 void fdct_ifast (DCTELEM *data);
  40 void fdct_ifast248 (DCTELEM *data);
  41 void ff_jpeg_fdct_islow (DCTELEM *data);
  42 void ff_fdct248_islow (DCTELEM *data);
  43
  44 void j_rev_dct (DCTELEM *data);
  45 void j_rev_dct4 (DCTELEM *data);
  46 void j_rev_dct2 (DCTELEM *data);
  47 void j_rev_dct1 (DCTELEM *data);
  48
  49 void ff_fdct_mmx(DCTELEM *block);
  50 void ff_fdct_mmx2(DCTELEM *block);
  51 void ff_fdct_sse2(DCTELEM *block);
  52
  53 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
  54 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
  55 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
  56 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
  57
  58 /* encoding scans */
  59 extern const uint8_t ff_alternate_horizontal_scan[64];
  60 extern const uint8_t ff_alternate_vertical_scan[64];
  61 extern const uint8_t ff_zigzag_direct[64];
  62 extern const uint8_t ff_zigzag248_direct[64];
  63
  64 /* pixel operations */
  65 #define MAX_NEG_CROP 1024
  66
  67 /* temporary */
  68 extern uint32_t squareTbl[512];
  69 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
  70
  71 /* VP3 DSP functions */
  72 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
  73 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  74 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
  75
  76 /* minimum alignment rules ;)
  77 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
  78 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
  79
  80 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
  81 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
  82
  83 !future video codecs might need functions with less strict alignment
  84 */
  85
  86 /*
  87 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
  88 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
  89 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
  90 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
  91 void clear_blocks_c(DCTELEM *blocks);
  92 */
  93
  94 /* add and put pixel (decoding) */
  95 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
  96 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
  97 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
  98 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);
  99 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 100 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 101 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
 102 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offsetd, int offsets);
 103
 104 #define DEF_OLD_QPEL(name)\
 105 void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 106 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
 107 void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 108
 109 DEF_OLD_QPEL(qpel16_mc11_old_c)
 110 DEF_OLD_QPEL(qpel16_mc31_old_c)
 111 DEF_OLD_QPEL(qpel16_mc12_old_c)
 112 DEF_OLD_QPEL(qpel16_mc32_old_c)
 113 DEF_OLD_QPEL(qpel16_mc13_old_c)
 114 DEF_OLD_QPEL(qpel16_mc33_old_c)
 115 DEF_OLD_QPEL(qpel8_mc11_old_c)
 116 DEF_OLD_QPEL(qpel8_mc31_old_c)
 117 DEF_OLD_QPEL(qpel8_mc12_old_c)
 118 DEF_OLD_QPEL(qpel8_mc32_old_c)
 119 DEF_OLD_QPEL(qpel8_mc13_old_c)
 120 DEF_OLD_QPEL(qpel8_mc33_old_c)
 121
 122 #define CALL_2X_PIXELS(a, b, n)\
 123 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 124     b(block  , pixels  , line_size, h);\
 125     b(block+n, pixels+n, line_size, h);\
 126 }
 127
 128 /* motion estimation */
 129 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
 130 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
 131 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))*/;
 132
 133
 134 /**
 135  * DSPContext.
 136  */
 137 typedef struct DSPContext {
 138     /* pixel ops : interface with DCT */
 139     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
 140     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
 141     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 142     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 143     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
 144     /**
 145      * translational global motion compensation.
 146      */
 147     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
 148     /**
 149      * global motion compensation.
 150      */
 151     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
 152                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
 153     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
 154     int (*pix_sum)(uint8_t * pix, int line_size);
 155     int (*pix_norm1)(uint8_t * pix, int line_size);
 156 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
 157
 158     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
 159     me_cmp_func sse[5];
 160     me_cmp_func hadamard8_diff[5];
 161     me_cmp_func dct_sad[5];
 162     me_cmp_func quant_psnr[5];
 163     me_cmp_func bit[5];
 164     me_cmp_func rd[5];
 165     me_cmp_func vsad[5];
 166     me_cmp_func vsse[5];
 167     me_cmp_func nsse[5];
 168     me_cmp_func w53[5];
 169     me_cmp_func w97[5];
 170     me_cmp_func dct_max[5];
 171
 172     me_cmp_func me_pre_cmp[5];
 173     me_cmp_func me_cmp[5];
 174     me_cmp_func me_sub_cmp[5];
 175     me_cmp_func mb_cmp[5];
 176     me_cmp_func ildct_cmp[5]; //only width 16 used
 177     me_cmp_func frame_skip_cmp[5]; //only width 8 used
 178
 179     /**
 180      * Halfpel motion compensation with rounding (a+b+1)>>1.
 181      * this is an array[4][4] of motion compensation funcions for 4
 182      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 183      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 184      * @param block destination where the result is stored
 185      * @param pixels source
 186      * @param line_size number of bytes in a horizontal line of block
 187      * @param h height
 188      */
 189     op_pixels_func put_pixels_tab[4][4];
 190
 191     /**
 192      * Halfpel motion compensation with rounding (a+b+1)>>1.
 193      * This is an array[4][4] of motion compensation functions for 4
 194      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 195      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 196      * @param block destination into which the result is averaged (a+b+1)>>1
 197      * @param pixels source
 198      * @param line_size number of bytes in a horizontal line of block
 199      * @param h height
 200      */
 201     op_pixels_func avg_pixels_tab[4][4];
 202
 203     /**
 204      * Halfpel motion compensation with no rounding (a+b)>>1.
 205      * this is an array[2][4] of motion compensation funcions for 2
 206      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 207      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 208      * @param block destination where the result is stored
 209      * @param pixels source
 210      * @param line_size number of bytes in a horizontal line of block
 211      * @param h height
 212      */
 213     op_pixels_func put_no_rnd_pixels_tab[4][4];
 214
 215     /**
 216      * Halfpel motion compensation with no rounding (a+b)>>1.
 217      * this is an array[2][4] of motion compensation funcions for 2
 218      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
 219      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
 220      * @param block destination into which the result is averaged (a+b)>>1
 221      * @param pixels source
 222      * @param line_size number of bytes in a horizontal line of block
 223      * @param h height
 224      */
 225     op_pixels_func avg_no_rnd_pixels_tab[4][4];
 226
 227     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);
 228
 229     /**
 230      * Thirdpel motion compensation with rounding (a+b+1)>>1.
 231      * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
 232      * *pixels_tab[ xthirdpel + 4*ythirdpel ]
 233      * @param block destination where the result is stored
 234      * @param pixels source
 235      * @param line_size number of bytes in a horizontal line of block
 236      * @param h height
 237      */
 238     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 239     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
 240
 241     qpel_mc_func put_qpel_pixels_tab[2][16];
 242     qpel_mc_func avg_qpel_pixels_tab[2][16];
 243     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 244     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 245     qpel_mc_func put_mspel_pixels_tab[8];
 246
 247     /**
 248      * h264 Chram MC
 249      */
 250     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
 251     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
 252
 253     qpel_mc_func put_h264_qpel_pixels_tab[3][16];
 254     qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
 255
 256     h264_weight_func weight_h264_pixels_tab[10];
 257     h264_biweight_func biweight_h264_pixels_tab[10];
 258
 259     me_cmp_func pix_abs[2][4];
 260
 261     /* huffyuv specific */
 262     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 263     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 264     /**
 265      * subtract huffyuv's variant of median prediction
 266      * note, this might read from src1[-1], src2[-1]
 267      */
 268     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
 269     void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
 270
 271     void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 272     void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 273     void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 274     void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
 275     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 276     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
 277
 278     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 279     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
 280
 281     void (*h261_loop_filter)(uint8_t *src, int stride);
 282
 283     /* (I)DCT */
 284     void (*fdct)(DCTELEM *block/* align 16*/);
 285     void (*fdct248)(DCTELEM *block/* align 16*/);
 286
 287     /* IDCT really*/
 288     void (*idct)(DCTELEM *block/* align 16*/);
 289
 290     /**
 291      * block -> idct -> clip to unsigned 8 bit -> dest.
 292      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
 293      * @param line_size size in bytes of a horizotal line of dest
 294      */
 295     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 296
 297     /**
 298      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
 299      * @param line_size size in bytes of a horizotal line of dest
 300      */
 301     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
 302
 303     /**
 304      * idct input permutation.
 305      * several optimized IDCTs need a permutated input (relative to the normal order of the reference
 306      * IDCT)
 307      * this permutation must be performed before the idct_put/add, note, normally this can be merged
 308      * with the zigzag/alternate scan<br>
 309      * an example to avoid confusion:
 310      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
 311      * - (x -> referece dct -> reference idct -> x)
 312      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
 313      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
 314      */
 315     uint8_t idct_permutation[64];
 316     int idct_permutation_type;
 317 #define FF_NO_IDCT_PERM 1
 318 #define FF_LIBMPEG2_IDCT_PERM 2
 319 #define FF_SIMPLE_IDCT_PERM 3
 320 #define FF_TRANSPOSE_IDCT_PERM 4
 321 #define FF_PARTTRANS_IDCT_PERM 5
 322
 323     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
 324     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 325 #define BASIS_SHIFT 16
 326 #define RECON_SHIFT 6
 327
 328     void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
 329     void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
 330 } DSPContext;
 331
 332 void dsputil_static_init(void);
 333 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
 334
 335 /**
 336  * permute block according to permuatation.
 337  * @param last last non zero element in scantable order
 338  */
 339 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
 340
 341 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
 342
 343 #define BYTE_VEC32(c)   ((c)*0x01010101UL)
 344
 345 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
 346 {
 347     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 348 }
 349
 350 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
 351 {
 352     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
 353 }
 354
 355 static inline int get_penalty_factor(int lambda, int lambda2, int type){
 356     switch(type&0xFF){
 357     default:
 358     case FF_CMP_SAD:
 359         return lambda>>FF_LAMBDA_SHIFT;
 360     case FF_CMP_DCT:
 361         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 362     case FF_CMP_W53:
 363         return (4*lambda)>>(FF_LAMBDA_SHIFT);
 364     case FF_CMP_W97:
 365         return (2*lambda)>>(FF_LAMBDA_SHIFT);
 366     case FF_CMP_SATD:
 367         return (2*lambda)>>FF_LAMBDA_SHIFT;
 368     case FF_CMP_RD:
 369     case FF_CMP_PSNR:
 370     case FF_CMP_SSE:
 371     case FF_CMP_NSSE:
 372         return lambda2>>FF_LAMBDA_SHIFT;
 373     case FF_CMP_BIT:
 374         return 1;
 375     }
 376 }
 377
 378 /**
 379  * Empty mmx state.
 380  * this must be called between any dsp function and float/double code.
 381  * for example sin(); dsp->idct_put(); emms_c(); cos()
 382  */
 383 #define emms_c()
 384
 385 /* should be defined by architectures supporting
 386    one or more MultiMedia extension */
 387 int mm_support(void);
 388
 389 #define __align16 __attribute__ ((aligned (16)))
 390
 391 #if defined(HAVE_MMX)
 392
 393 #undef emms_c
 394
 395 #define MM_MMX    0x0001 /* standard MMX */
 396 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
 397 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
 398 #define MM_SSE    0x0008 /* SSE functions */
 399 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
 400 #define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
 401
 402 extern int mm_flags;
 403
 404 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 405 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 406 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
 407
 408 static inline void emms(void)
 409 {
 410     __asm __volatile ("emms;":::"memory");
 411 }
 412
 413
 414 #define emms_c() \
 415 {\
 416     if (mm_flags & MM_MMX)\
 417         emms();\
 418 }
 419
 420 #define __align8 __attribute__ ((aligned (8)))
 421 #define STRIDE_ALIGN 8
 422
 423 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
 424 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
 425
 426 #elif defined(ARCH_ARMV4L)
 427
 428 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
 429    line optimizations */
 430 #define __align8 __attribute__ ((aligned (4)))
 431 #define STRIDE_ALIGN 4
 432
 433 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
 434
 435 #elif defined(HAVE_MLIB)
 436
 437 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 438 #define __align8 __attribute__ ((aligned (8)))
 439 #define STRIDE_ALIGN 8
 440
 441 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
 442
 443 #elif defined(ARCH_SPARC)
 444
 445 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 446 #define __align8 __attribute__ ((aligned (8)))
 447 #define STRIDE_ALIGN 8
 448 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 449
 450 #elif defined(ARCH_ALPHA)
 451
 452 #define __align8 __attribute__ ((aligned (8)))
 453 #define STRIDE_ALIGN 8
 454
 455 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
 456
 457 #elif defined(ARCH_POWERPC)
 458
 459 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
 460
 461 extern int mm_flags;
 462
 463 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
 464 #define pixel altivec_pixel
 465 #include <altivec.h>
 466 #undef pixel
 467 #endif
 468
 469 #define __align8 __attribute__ ((aligned (16)))
 470 #define STRIDE_ALIGN 16
 471
 472 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 473
 474 #elif defined(HAVE_MMI)
 475
 476 #define __align8 __attribute__ ((aligned (16)))
 477 #define STRIDE_ALIGN 16
 478
 479 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
 480
 481 #elif defined(ARCH_SH4)
 482
 483 #define __align8 __attribute__ ((aligned (8)))
 484 #define STRIDE_ALIGN 8
 485
 486 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 487
 488 #else
 489
 490 #define __align8 __attribute__ ((aligned (8)))
 491 #define STRIDE_ALIGN 8
 492
 493 #endif
 494
 495 #ifdef __GNUC__
 496
 497 struct unaligned_64 { uint64_t l; } __attribute__((packed));
 498 struct unaligned_32 { uint32_t l; } __attribute__((packed));
 499 struct unaligned_16 { uint16_t l; } __attribute__((packed));
 500
 501 #define LD16(a) (((const struct unaligned_16 *) (a))->l)
 502 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
 503 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
 504
 505 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
 506
 507 #else /* __GNUC__ */
 508
 509 #define LD16(a) (*((uint16_t*)(a)))
 510 #define LD32(a) (*((uint32_t*)(a)))
 511 #define LD64(a) (*((uint64_t*)(a)))
 512
 513 #define ST32(a, b) *((uint32_t*)(a)) = (b)
 514
 515 #endif /* !__GNUC__ */
 516
 517 /* PSNR */
 518 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
 519               int orig_linesize[3], int coded_linesize,
 520               AVCodecContext *avctx);
 521
 522 /* FFT computation */
 523
 524 /* NOTE: soon integer code will be added, so you must use the
 525    FFTSample type */
 526 typedef float FFTSample;
 527
 528 typedef struct FFTComplex {
 529     FFTSample re, im;
 530 } FFTComplex;
 531
 532 typedef struct FFTContext {
 533     int nbits;
 534     int inverse;
 535     uint16_t *revtab;
 536     FFTComplex *exptab;
 537     FFTComplex *exptab1; /* only used by SSE code */
 538     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 539 } FFTContext;
 540
 541 int ff_fft_init(FFTContext *s, int nbits, int inverse);
 542 void ff_fft_permute(FFTContext *s, FFTComplex *z);
 543 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
 544 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
 545 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
 546
 547 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
 548 {
 549     s->fft_calc(s, z);
 550 }
 551 void ff_fft_end(FFTContext *s);
 552
 553 /* MDCT computation */
 554
 555 typedef struct MDCTContext {
 556     int n;  /* size of MDCT (i.e. number of input data * 2) */
 557     int nbits; /* n = 2^nbits */
 558     /* pre/post rotation tables */
 559     FFTSample *tcos;
 560     FFTSample *tsin;
 561     FFTContext fft;
 562 } MDCTContext;
 563
 564 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
 565 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
 566                 const FFTSample *input, FFTSample *tmp);
 567 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
 568                const FFTSample *input, FFTSample *tmp);
 569 void ff_mdct_end(MDCTContext *s);
 570
 571 #define WARPER8_16(name8, name16)\
 572 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 573     return name8(s, dst           , src           , stride, h)\
 574           +name8(s, dst+8         , src+8         , stride, h);\
 575 }
 576
 577 #define WARPER8_16_SQ(name8, name16)\
 578 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
 579     int score=0;\
 580     score +=name8(s, dst           , src           , stride, 8);\
 581     score +=name8(s, dst+8         , src+8         , stride, 8);\
 582     if(h==16){\
 583         dst += 8*stride;\
 584         src += 8*stride;\
 585         score +=name8(s, dst           , src           , stride, 8);\
 586         score +=name8(s, dst+8         , src+8         , stride, 8);\
 587     }\
 588     return score;\
 589 }
 590
 591 #ifndef HAVE_LRINTF
 592 /* XXX: add ISOC specific test to avoid specific BSD testing. */
 593 /* better than nothing implementation. */
 594 /* btw, rintf() is existing on fbsd too -- alex */
 595 static always_inline long int lrintf(float x)
 596 {
 597 #ifdef CONFIG_WIN32
 598 #  ifdef ARCH_X86
 599     int32_t i;
 600     asm volatile(
 601         "fistpl %0\n\t"
 602         : "=m" (i) : "t" (x) : "st"
 603     );
 604     return i;
 605 #  else
 606     /* XXX: incorrect, but make it compile */
 607     return (int)(x + (x < 0 ? -0.5 : 0.5));
 608 #  endif
 609 #else
 610     return (int)(rint(x));
 611 #endif
 612 }
 613 #else
 614 #ifndef _ISOC9X_SOURCE
 615 #define _ISOC9X_SOURCE
 616 #endif
 617 #include <math.h>
 618 #endif
 619
 620 #endif