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