falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
12811 Commits
32 Branches
283MB
Tree: 015d0bb95c
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '015d0bb95c'
${ noResults }
FFmpeg/libavcodec/dsputil.h

752 lines
28KB
Raw Blame History 

  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 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 FFMPEG_DSPUTIL_H

  31. #define FFMPEG_DSPUTIL_H

  32. 

  33. #include "avcodec.h"

  34. 

  35. 

  36. //#define DEBUG

  37. /* dct code */

  38. typedef short DCTELEM;

  39. typedef int DWTELEM;

  40. typedef short IDWTELEM;

  41. 

  42. void fdct_ifast (DCTELEM *data);

  43. void fdct_ifast248 (DCTELEM *data);

  44. void ff_jpeg_fdct_islow (DCTELEM *data);

  45. void ff_fdct248_islow (DCTELEM *data);

  46. 

  47. void j_rev_dct (DCTELEM *data);

  48. void j_rev_dct4 (DCTELEM *data);

  49. void j_rev_dct2 (DCTELEM *data);

  50. void j_rev_dct1 (DCTELEM *data);

  51. void ff_wmv2_idct_c(DCTELEM *data);

  52. 

  53. void ff_fdct_mmx(DCTELEM *block);

  54. void ff_fdct_mmx2(DCTELEM *block);

  55. void ff_fdct_sse2(DCTELEM *block);

  56. 

  57. void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);

  58. void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);

  59. void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);

  60. void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);

  61. void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);

  62. void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);

  63. 

  64. void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,

  65.                               const float *src2, int src3, int blocksize, int step);

  66. void ff_float_to_int16_c(int16_t *dst, const float *src, int len);

  67. 

  68. /* encoding scans */

  69. extern const uint8_t ff_alternate_horizontal_scan[64];

  70. extern const uint8_t ff_alternate_vertical_scan[64];

  71. extern const uint8_t ff_zigzag_direct[64];

  72. extern const uint8_t ff_zigzag248_direct[64];

  73. 

  74. /* pixel operations */

  75. #define MAX_NEG_CROP 1024

  76. 

  77. /* temporary */

  78. extern uint32_t ff_squareTbl[512];

  79. extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];

  80. 

  81. /* VP3 DSP functions */

  82. void ff_vp3_idct_c(DCTELEM *block/* align 16*/);

  83. void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);

  84. void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);

  85. 

  86. /* 1/2^n downscaling functions from imgconvert.c */

  87. void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);

  88. void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);

  89. void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);

  90. void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);

  91. 

  92. void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,

  93.               int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);

  94. 

  95. /* minimum alignment rules ;)

  96. if u notice errors in the align stuff, need more alignment for some asm code for some cpu

  97. or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...

  98. 

  99. !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)

  100. i (michael) didnt check them, these are just the alignents which i think could be reached easily ...

  101. 

  102. !future video codecs might need functions with less strict alignment

  103. */

  104. 

  105. /*

  106. void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);

  107. void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);

  108. void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);

  109. void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);

  110. void clear_blocks_c(DCTELEM *blocks);

  111. */

  112. 

  113. /* add and put pixel (decoding) */

  114. // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16

  115. //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4

  116. typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);

  117. 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);

  118. typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);

  119. typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);

  120. typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);

  121. typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);

  122. 

  123. #define DEF_OLD_QPEL(name)\

  124. void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\

  125. void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\

  126. void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);

  127. 

  128. DEF_OLD_QPEL(qpel16_mc11_old_c)

  129. DEF_OLD_QPEL(qpel16_mc31_old_c)

  130. DEF_OLD_QPEL(qpel16_mc12_old_c)

  131. DEF_OLD_QPEL(qpel16_mc32_old_c)

  132. DEF_OLD_QPEL(qpel16_mc13_old_c)

  133. DEF_OLD_QPEL(qpel16_mc33_old_c)

  134. DEF_OLD_QPEL(qpel8_mc11_old_c)

  135. DEF_OLD_QPEL(qpel8_mc31_old_c)

  136. DEF_OLD_QPEL(qpel8_mc12_old_c)

  137. DEF_OLD_QPEL(qpel8_mc32_old_c)

  138. DEF_OLD_QPEL(qpel8_mc13_old_c)

  139. DEF_OLD_QPEL(qpel8_mc33_old_c)

  140. 

  141. #define CALL_2X_PIXELS(a, b, n)\

  142. static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\

  143.     b(block  , pixels  , line_size, h);\

  144.     b(block+n, pixels+n, line_size, h);\

  145. }

  146. 

  147. /* motion estimation */

  148. // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2

  149. // although currently h<4 is not used as functions with width <8 are neither used nor implemented

  150. 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))*/;

  151. 

  152. 

  153. // for snow slices

  154. typedef struct slice_buffer_s slice_buffer;

  155. 

  156. /**

  157.  * Scantable.

  158.  */

  159. typedef struct ScanTable{

  160.     const uint8_t *scantable;

  161.     uint8_t permutated[64];

  162.     uint8_t raster_end[64];

  163. #ifdef ARCH_POWERPC

  164.                 /** Used by dct_quantize_altivec to find last-non-zero */

  165.     DECLARE_ALIGNED_8(uint8_t, inverse[64]);

  166. #endif

  167. } ScanTable;

  168. 

  169. void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);

  170. 

  171. /**

  172.  * DSPContext.

  173.  */

  174. typedef struct DSPContext {

  175.     /* pixel ops : interface with DCT */

  176.     void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);

  177.     void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);

  178.     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);

  179.     void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);

  180.     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);

  181.     void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);

  182.     void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);

  183.     int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);

  184.     /**

  185.      * translational global motion compensation.

  186.      */

  187.     void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);

  188.     /**

  189.      * global motion compensation.

  190.      */

  191.     void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,

  192.                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);

  193.     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);

  194.     int (*pix_sum)(uint8_t * pix, int line_size);

  195.     int (*pix_norm1)(uint8_t * pix, int line_size);

  196. // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4

  197. 

  198.     me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */

  199.     me_cmp_func sse[5];

  200.     me_cmp_func hadamard8_diff[5];

  201.     me_cmp_func dct_sad[5];

  202.     me_cmp_func quant_psnr[5];

  203.     me_cmp_func bit[5];

  204.     me_cmp_func rd[5];

  205.     me_cmp_func vsad[5];

  206.     me_cmp_func vsse[5];

  207.     me_cmp_func nsse[5];

  208.     me_cmp_func w53[5];

  209.     me_cmp_func w97[5];

  210.     me_cmp_func dct_max[5];

  211.     me_cmp_func dct264_sad[5];

  212. 

  213.     me_cmp_func me_pre_cmp[5];

  214.     me_cmp_func me_cmp[5];

  215.     me_cmp_func me_sub_cmp[5];

  216.     me_cmp_func mb_cmp[5];

  217.     me_cmp_func ildct_cmp[5]; //only width 16 used

  218.     me_cmp_func frame_skip_cmp[5]; //only width 8 used

  219. 

  220.     int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,

  221.                              int size);

  222. 

  223.     /**

  224.      * Halfpel motion compensation with rounding (a+b+1)>>1.

  225.      * this is an array[4][4] of motion compensation functions for 4

  226.      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>

  227.      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]

  228.      * @param block destination where the result is stored

  229.      * @param pixels source

  230.      * @param line_size number of bytes in a horizontal line of block

  231.      * @param h height

  232.      */

  233.     op_pixels_func put_pixels_tab[4][4];

  234. 

  235.     /**

  236.      * Halfpel motion compensation with rounding (a+b+1)>>1.

  237.      * This is an array[4][4] of motion compensation functions for 4

  238.      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>

  239.      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]

  240.      * @param block destination into which the result is averaged (a+b+1)>>1

  241.      * @param pixels source

  242.      * @param line_size number of bytes in a horizontal line of block

  243.      * @param h height

  244.      */

  245.     op_pixels_func avg_pixels_tab[4][4];

  246. 

  247.     /**

  248.      * Halfpel motion compensation with no rounding (a+b)>>1.

  249.      * this is an array[2][4] of motion compensation functions for 2

  250.      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>

  251.      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]

  252.      * @param block destination where the result is stored

  253.      * @param pixels source

  254.      * @param line_size number of bytes in a horizontal line of block

  255.      * @param h height

  256.      */

  257.     op_pixels_func put_no_rnd_pixels_tab[4][4];

  258. 

  259.     /**

  260.      * Halfpel motion compensation with no rounding (a+b)>>1.

  261.      * this is an array[2][4] of motion compensation functions for 2

  262.      * horizontal blocksizes (8,16) and the 4 halfpel positions<br>

  263.      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]

  264.      * @param block destination into which the result is averaged (a+b)>>1

  265.      * @param pixels source

  266.      * @param line_size number of bytes in a horizontal line of block

  267.      * @param h height

  268.      */

  269.     op_pixels_func avg_no_rnd_pixels_tab[4][4];

  270. 

  271.     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);

  272. 

  273.     /**

  274.      * Thirdpel motion compensation with rounding (a+b+1)>>1.

  275.      * this is an array[12] of motion compensation functions for the 9 thirdpe

  276.      * positions<br>

  277.      * *pixels_tab[ xthirdpel + 4*ythirdpel ]

  278.      * @param block destination where the result is stored

  279.      * @param pixels source

  280.      * @param line_size number of bytes in a horizontal line of block

  281.      * @param h height

  282.      */

  283.     tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?

  284.     tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?

  285. 

  286.     qpel_mc_func put_qpel_pixels_tab[2][16];

  287.     qpel_mc_func avg_qpel_pixels_tab[2][16];

  288.     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];

  289.     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];

  290.     qpel_mc_func put_mspel_pixels_tab[8];

  291. 

  292.     /**

  293.      * h264 Chroma MC

  294.      */

  295.     h264_chroma_mc_func put_h264_chroma_pixels_tab[3];

  296.     /* This is really one func used in VC-1 decoding */

  297.     h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];

  298.     h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];

  299. 

  300.     qpel_mc_func put_h264_qpel_pixels_tab[4][16];

  301.     qpel_mc_func avg_h264_qpel_pixels_tab[4][16];

  302. 

  303.     qpel_mc_func put_2tap_qpel_pixels_tab[4][16];

  304.     qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];

  305. 

  306.     h264_weight_func weight_h264_pixels_tab[10];

  307.     h264_biweight_func biweight_h264_pixels_tab[10];

  308. 

  309.     /* AVS specific */

  310.     qpel_mc_func put_cavs_qpel_pixels_tab[2][16];

  311.     qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];

  312.     void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);

  313.     void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);

  314.     void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);

  315.     void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);

  316.     void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);

  317. 

  318.     me_cmp_func pix_abs[2][4];

  319. 

  320.     /* huffyuv specific */

  321.     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);

  322.     void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);

  323.     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);

  324.     /**

  325.      * subtract huffyuv's variant of median prediction

  326.      * note, this might read from src1[-1], src2[-1]

  327.      */

  328.     void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);

  329.     /* this might write to dst[w] */

  330.     void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);

  331.     void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);

  332. 

  333.     void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);

  334.     void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);

  335.     void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);

  336.     void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);

  337.     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);

  338.     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);

  339.     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c

  340.     void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],

  341.                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1);

  342. 

  343.     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);

  344.     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);

  345. 

  346.     void (*h261_loop_filter)(uint8_t *src, int stride);

  347. 

  348.     void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);

  349.     void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);

  350. 

  351.     /* assume len is a multiple of 4, and arrays are 16-byte aligned */

  352.     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);

  353.     /* no alignment needed */

  354.     void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);

  355.     /* assume len is a multiple of 8, and arrays are 16-byte aligned */

  356.     void (*vector_fmul)(float *dst, const float *src, int len);

  357.     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);

  358.     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */

  359.     void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);

  360. 

  361.     /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]

  362.      * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */

  363.     void (*float_to_int16)(int16_t *dst, const float *src, int len);

  364. 

  365.     /* (I)DCT */

  366.     void (*fdct)(DCTELEM *block/* align 16*/);

  367.     void (*fdct248)(DCTELEM *block/* align 16*/);

  368. 

  369.     /* IDCT really*/

  370.     void (*idct)(DCTELEM *block/* align 16*/);

  371. 

  372.     /**

  373.      * block -> idct -> clip to unsigned 8 bit -> dest.

  374.      * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)

  375.      * @param line_size size in bytes of a horizontal line of dest

  376.      */

  377.     void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);

  378. 

  379.     /**

  380.      * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.

  381.      * @param line_size size in bytes of a horizontal line of dest

  382.      */

  383.     void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);

  384. 

  385.     /**

  386.      * idct input permutation.

  387.      * several optimized IDCTs need a permutated input (relative to the normal order of the reference

  388.      * IDCT)

  389.      * this permutation must be performed before the idct_put/add, note, normally this can be merged

  390.      * with the zigzag/alternate scan<br>

  391.      * an example to avoid confusion:

  392.      * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)

  393.      * - (x -> referece dct -> reference idct -> x)

  394.      * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)

  395.      * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)

  396.      */

  397.     uint8_t idct_permutation[64];

  398.     int idct_permutation_type;

  399. #define FF_NO_IDCT_PERM 1

  400. #define FF_LIBMPEG2_IDCT_PERM 2

  401. #define FF_SIMPLE_IDCT_PERM 3

  402. #define FF_TRANSPOSE_IDCT_PERM 4

  403. #define FF_PARTTRANS_IDCT_PERM 5

  404. 

  405.     int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);

  406.     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);

  407. #define BASIS_SHIFT 16

  408. #define RECON_SHIFT 6

  409. 

  410.     void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);

  411. 

  412.     /* h264 functions */

  413.     void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);

  414.     void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);

  415.     void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);

  416.     void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);

  417.     void (*h264_dct)(DCTELEM block[4][4]);

  418. 

  419.     /* snow wavelet */

  420.     void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);

  421.     void (*horizontal_compose97i)(IDWTELEM *b, int width);

  422.     void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);

  423. 

  424.     void (*prefetch)(void *mem, int stride, int h);

  425. 

  426.     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);

  427. 

  428.     /* vc1 functions */

  429.     void (*vc1_inv_trans_8x8)(DCTELEM *b);

  430.     void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);

  431.     void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);

  432.     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);

  433.     void (*vc1_v_overlap)(uint8_t* src, int stride);

  434.     void (*vc1_h_overlap)(uint8_t* src, int stride);

  435.     /* put 8x8 block with bicubic interpolation and quarterpel precision

  436.      * last argument is actually round value instead of height

  437.      */

  438.     op_pixels_func put_vc1_mspel_pixels_tab[16];

  439. 

  440.     /* intrax8 functions */

  441.     void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);

  442.     void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,

  443.            int * range, int * sum,  int edges);

  444. 

  445. } DSPContext;

  446. 

  447. void dsputil_static_init(void);

  448. void dsputil_init(DSPContext* p, AVCodecContext *avctx);

  449. 

  450. int ff_check_alignment(void);

  451. 

  452. /**

  453.  * permute block according to permuatation.

  454.  * @param last last non zero element in scantable order

  455.  */

  456. void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);

  457. 

  458. void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);

  459. 

  460. #define         BYTE_VEC32(c)   ((c)*0x01010101UL)

  461. 

  462. static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)

  463. {

  464.     return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);

  465. }

  466. 

  467. static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)

  468. {

  469.     return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);

  470. }

  471. 

  472. static inline int get_penalty_factor(int lambda, int lambda2, int type){

  473.     switch(type&0xFF){

  474.     default:

  475.     case FF_CMP_SAD:

  476.         return lambda>>FF_LAMBDA_SHIFT;

  477.     case FF_CMP_DCT:

  478.         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);

  479.     case FF_CMP_W53:

  480.         return (4*lambda)>>(FF_LAMBDA_SHIFT);

  481.     case FF_CMP_W97:

  482.         return (2*lambda)>>(FF_LAMBDA_SHIFT);

  483.     case FF_CMP_SATD:

  484.     case FF_CMP_DCT264:

  485.         return (2*lambda)>>FF_LAMBDA_SHIFT;

  486.     case FF_CMP_RD:

  487.     case FF_CMP_PSNR:

  488.     case FF_CMP_SSE:

  489.     case FF_CMP_NSSE:

  490.         return lambda2>>FF_LAMBDA_SHIFT;

  491.     case FF_CMP_BIT:

  492.         return 1;

  493.     }

  494. }

  495. 

  496. /**

  497.  * Empty mmx state.

  498.  * this must be called between any dsp function and float/double code.

  499.  * for example sin(); dsp->idct_put(); emms_c(); cos()

  500.  */

  501. #define emms_c()

  502. 

  503. /* should be defined by architectures supporting

  504.    one or more MultiMedia extension */

  505. int mm_support(void);

  506. 

  507. void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);

  508. void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);

  509. void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);

  510. void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);

  511. void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);

  512. void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);

  513. void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);

  514. void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);

  515. void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);

  516. 

  517. #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)

  518. 

  519. #if defined(HAVE_MMX)

  520. 

  521. #undef emms_c

  522. 

  523. #define MM_MMX    0x0001 /* standard MMX */

  524. #define MM_3DNOW  0x0004 /* AMD 3DNOW */

  525. #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */

  526. #define MM_SSE    0x0008 /* SSE functions */

  527. #define MM_SSE2   0x0010 /* PIV SSE2 functions */

  528. #define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */

  529. #define MM_SSE3   0x0040 /* Prescott SSE3 functions */

  530. #define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */

  531. 

  532. extern int mm_flags;

  533. 

  534. void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);

  535. void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);

  536. void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);

  537. 

  538. static inline void emms(void)

  539. {

  540.     asm volatile ("emms;":::"memory");

  541. }

  542. 

  543. 

  544. #define emms_c() \

  545. {\

  546.     if (mm_flags & MM_MMX)\

  547.         emms();\

  548. }

  549. 

  550. void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);

  551. 

  552. #elif defined(ARCH_ARMV4L)

  553. 

  554. #define MM_IWMMXT    0x0100 /* XScale IWMMXT */

  555. 

  556. extern int mm_flags;

  557. 

  558. #elif defined(ARCH_POWERPC)

  559. 

  560. #define MM_ALTIVEC    0x0001 /* standard AltiVec */

  561. 

  562. extern int mm_flags;

  563. 

  564. #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)

  565. #define STRIDE_ALIGN 16

  566. 

  567. #elif defined(HAVE_MMI)

  568. 

  569. #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)

  570. #define STRIDE_ALIGN 16

  571. 

  572. #endif

  573. 

  574. #ifndef DECLARE_ALIGNED_8

  575. #   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)

  576. #endif

  577. 

  578. #ifndef STRIDE_ALIGN

  579. #   define STRIDE_ALIGN 8

  580. #endif

  581. 

  582. /* PSNR */

  583. void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],

  584.               int orig_linesize[3], int coded_linesize,

  585.               AVCodecContext *avctx);

  586. 

  587. /* FFT computation */

  588. 

  589. /* NOTE: soon integer code will be added, so you must use the

  590.    FFTSample type */

  591. typedef float FFTSample;

  592. 

  593. struct MDCTContext;

  594. 

  595. typedef struct FFTComplex {

  596.     FFTSample re, im;

  597. } FFTComplex;

  598. 

  599. typedef struct FFTContext {

  600.     int nbits;

  601.     int inverse;

  602.     uint16_t *revtab;

  603.     FFTComplex *exptab;

  604.     FFTComplex *exptab1; /* only used by SSE code */

  605.     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);

  606.     void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,

  607.                        const FFTSample *input, FFTSample *tmp);

  608. } FFTContext;

  609. 

  610. int ff_fft_init(FFTContext *s, int nbits, int inverse);

  611. void ff_fft_permute(FFTContext *s, FFTComplex *z);

  612. void ff_fft_calc_c(FFTContext *s, FFTComplex *z);

  613. void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);

  614. void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);

  615. void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);

  616. void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);

  617. 

  618. static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)

  619. {

  620.     s->fft_calc(s, z);

  621. }

  622. void ff_fft_end(FFTContext *s);

  623. 

  624. /* MDCT computation */

  625. 

  626. typedef struct MDCTContext {

  627.     int n;  /* size of MDCT (i.e. number of input data * 2) */

  628.     int nbits; /* n = 2^nbits */

  629.     /* pre/post rotation tables */

  630.     FFTSample *tcos;

  631.     FFTSample *tsin;

  632.     FFTContext fft;

  633. } MDCTContext;

  634. 

  635. /**

  636.  * Generate a Kaiser-Bessel Derived Window.

  637.  * @param   window  pointer to half window

  638.  * @param   alpha   determines window shape

  639.  * @param   n       size of half window

  640.  */

  641. void ff_kbd_window_init(float *window, float alpha, int n);

  642. 

  643. int ff_mdct_init(MDCTContext *s, int nbits, int inverse);

  644. void ff_imdct_calc(MDCTContext *s, FFTSample *output,

  645.                 const FFTSample *input, FFTSample *tmp);

  646. void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,

  647.                         const FFTSample *input, FFTSample *tmp);

  648. void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,

  649.                        const FFTSample *input, FFTSample *tmp);

  650. void ff_mdct_calc(MDCTContext *s, FFTSample *out,

  651.                const FFTSample *input, FFTSample *tmp);

  652. void ff_mdct_end(MDCTContext *s);

  653. 

  654. #define WRAPPER8_16(name8, name16)\

  655. static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\

  656.     return name8(s, dst           , src           , stride, h)\

  657.           +name8(s, dst+8         , src+8         , stride, h);\

  658. }

  659. 

  660. #define WRAPPER8_16_SQ(name8, name16)\

  661. static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\

  662.     int score=0;\

  663.     score +=name8(s, dst           , src           , stride, 8);\

  664.     score +=name8(s, dst+8         , src+8         , stride, 8);\

  665.     if(h==16){\

  666.         dst += 8*stride;\

  667.         src += 8*stride;\

  668.         score +=name8(s, dst           , src           , stride, 8);\

  669.         score +=name8(s, dst+8         , src+8         , stride, 8);\

  670.     }\

  671.     return score;\

  672. }

  673. 

  674. 

  675. static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)

  676. {

  677.     int i;

  678.     for(i=0; i<h; i++)

  679.     {

  680.         AV_WN16(dst   , AV_RN16(src   ));

  681.         dst+=dstStride;

  682.         src+=srcStride;

  683.     }

  684. }

  685. 

  686. static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)

  687. {

  688.     int i;

  689.     for(i=0; i<h; i++)

  690.     {

  691.         AV_WN32(dst   , AV_RN32(src   ));

  692.         dst+=dstStride;

  693.         src+=srcStride;

  694.     }

  695. }

  696. 

  697. static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)

  698. {

  699.     int i;

  700.     for(i=0; i<h; i++)

  701.     {

  702.         AV_WN32(dst   , AV_RN32(src   ));

  703.         AV_WN32(dst+4 , AV_RN32(src+4 ));

  704.         dst+=dstStride;

  705.         src+=srcStride;

  706.     }

  707. }

  708. 

  709. static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)

  710. {

  711.     int i;

  712.     for(i=0; i<h; i++)

  713.     {

  714.         AV_WN32(dst   , AV_RN32(src   ));

  715.         AV_WN32(dst+4 , AV_RN32(src+4 ));

  716.         dst[8]= src[8];

  717.         dst+=dstStride;

  718.         src+=srcStride;

  719.     }

  720. }

  721. 

  722. static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)

  723. {

  724.     int i;

  725.     for(i=0; i<h; i++)

  726.     {

  727.         AV_WN32(dst   , AV_RN32(src   ));

  728.         AV_WN32(dst+4 , AV_RN32(src+4 ));

  729.         AV_WN32(dst+8 , AV_RN32(src+8 ));

  730.         AV_WN32(dst+12, AV_RN32(src+12));

  731.         dst+=dstStride;

  732.         src+=srcStride;

  733.     }

  734. }

  735. 

  736. static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)

  737. {

  738.     int i;

  739.     for(i=0; i<h; i++)

  740.     {

  741.         AV_WN32(dst   , AV_RN32(src   ));

  742.         AV_WN32(dst+4 , AV_RN32(src+4 ));

  743.         AV_WN32(dst+8 , AV_RN32(src+8 ));

  744.         AV_WN32(dst+12, AV_RN32(src+12));

  745.         dst[16]= src[16];

  746.         dst+=dstStride;

  747.         src+=srcStride;

  748.     }

  749. }

  750. 

  751. #endif /* FFMPEG_DSPUTIL_H */