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.
22319 Commits
32 Branches
449MB
Tree: ddb901b74d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ddb901b74d'
${ noResults }
FFmpeg/libavcodec/bfin/dsputil_bfin.c

313 lines
12KB
Raw Blame History 

  1. /*

  2.  * BlackFin DSPUTILS

  3.  *

  4.  * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>

  5.  * Copyright (c) 2006 Michael Benjamin <michael.benjamin@analog.com>

  6.  *

  7.  * This file is part of FFmpeg.

  8.  *

  9.  * FFmpeg is free software; you can redistribute it and/or

  10.  * modify it under the terms of the GNU Lesser General Public

  11.  * License as published by the Free Software Foundation; either

  12.  * version 2.1 of the License, or (at your option) any later version.

  13.  *

  14.  * FFmpeg is distributed in the hope that it will be useful,

  15.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  16.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  17.  * Lesser General Public License for more details.

  18.  *

  19.  * You should have received a copy of the GNU Lesser General Public

  20.  * License along with FFmpeg; if not, write to the Free Software

  21.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  22.  */

  23. 

  24. #include "libavcodec/avcodec.h"

  25. #include "libavcodec/dsputil.h"

  26. #include "dsputil_bfin.h"

  27. 

  28. int off;

  29. 

  30. 

  31. void ff_bfin_idct (DCTELEM *block) attribute_l1_text;

  32. void ff_bfin_fdct (DCTELEM *block) attribute_l1_text;

  33. void ff_bfin_vp3_idct (DCTELEM *block);

  34. void ff_bfin_vp3_idct_put (uint8_t *dest, int line_size, DCTELEM *block);

  35. void ff_bfin_vp3_idct_add (uint8_t *dest, int line_size, DCTELEM *block);

  36. void ff_bfin_add_pixels_clamped (const DCTELEM *block, uint8_t *dest, int line_size) attribute_l1_text;

  37. void ff_bfin_put_pixels_clamped (const DCTELEM *block, uint8_t *dest, int line_size) attribute_l1_text;

  38. void ff_bfin_diff_pixels (DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)  attribute_l1_text;

  39. void ff_bfin_get_pixels  (DCTELEM *restrict block, const uint8_t *pixels, int line_size) attribute_l1_text;

  40. int  ff_bfin_pix_norm1  (uint8_t * pix, int line_size) attribute_l1_text;

  41. int  ff_bfin_z_sad8x8   (uint8_t *blk1, uint8_t *blk2, int dsz, int line_size, int h) attribute_l1_text;

  42. int  ff_bfin_z_sad16x16 (uint8_t *blk1, uint8_t *blk2, int dsz, int line_size, int h) attribute_l1_text;

  43. 

  44. void ff_bfin_z_put_pixels16_xy2     (uint8_t *block, const uint8_t *s0, int dest_size, int line_size, int h) attribute_l1_text;

  45. void ff_bfin_z_put_pixels8_xy2      (uint8_t *block, const uint8_t *s0, int dest_size, int line_size, int h) attribute_l1_text;

  46. void ff_bfin_put_pixels16_xy2_nornd (uint8_t *block, const uint8_t *s0, int line_size, int h) attribute_l1_text;

  47. void ff_bfin_put_pixels8_xy2_nornd  (uint8_t *block, const uint8_t *s0, int line_size, int h) attribute_l1_text;

  48. 

  49. 

  50. int  ff_bfin_pix_sum (uint8_t *p, int stride) attribute_l1_text;

  51. 

  52. void ff_bfin_put_pixels8uc        (uint8_t *block, const uint8_t *s0, const uint8_t *s1, int dest_size, int line_size, int h) attribute_l1_text;

  53. void ff_bfin_put_pixels16uc       (uint8_t *block, const uint8_t *s0, const uint8_t *s1, int dest_size, int line_size, int h) attribute_l1_text;

  54. void ff_bfin_put_pixels8uc_nornd  (uint8_t *block, const uint8_t *s0, const uint8_t *s1, int line_size, int h) attribute_l1_text;

  55. void ff_bfin_put_pixels16uc_nornd (uint8_t *block, const uint8_t *s0, const uint8_t *s1, int line_size, int h) attribute_l1_text;

  56. 

  57. int ff_bfin_sse4  (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) attribute_l1_text;

  58. int ff_bfin_sse8  (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) attribute_l1_text;

  59. int ff_bfin_sse16 (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) attribute_l1_text;

  60. 

  61. 

  62. static void bfin_idct_add (uint8_t *dest, int line_size, DCTELEM *block)

  63. {

  64.     ff_bfin_idct (block);

  65.     ff_bfin_add_pixels_clamped (block, dest, line_size);

  66. }

  67. 

  68. static void bfin_idct_put (uint8_t *dest, int line_size, DCTELEM *block)

  69. {

  70.     ff_bfin_idct (block);

  71.     ff_bfin_put_pixels_clamped (block, dest, line_size);

  72. }

  73. 

  74. 

  75. static void bfin_clear_blocks (DCTELEM *blocks)

  76. {

  77.     // This is just a simple memset.

  78.     //

  79.     __asm__("P0=192; "

  80.         "I0=%0;  "

  81.         "R0=0;   "

  82.         "LSETUP(clear_blocks_blkfn_lab,clear_blocks_blkfn_lab)LC0=P0;"

  83.         "clear_blocks_blkfn_lab:"

  84.         "[I0++]=R0;"

  85.         ::"a" (blocks):"P0","I0","R0");

  86. }

  87. 

  88. 

  89. 

  90. static void bfin_put_pixels8 (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  91. {

  92.     ff_bfin_put_pixels8uc (block, pixels, pixels, line_size, line_size, h);

  93. }

  94. 

  95. static void bfin_put_pixels8_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  96. {

  97.     ff_bfin_put_pixels8uc (block, pixels, pixels+1, line_size, line_size, h);

  98. }

  99. 

  100. static void bfin_put_pixels8_y2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  101. {

  102.     ff_bfin_put_pixels8uc (block, pixels, pixels+line_size, line_size, line_size, h);

  103. }

  104. 

  105. static void bfin_put_pixels8_xy2 (uint8_t *block, const uint8_t *s0, int line_size, int h)

  106. {

  107.     ff_bfin_z_put_pixels8_xy2 (block,s0,line_size, line_size, h);

  108. }

  109. 

  110. static void bfin_put_pixels16 (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  111. {

  112.     ff_bfin_put_pixels16uc (block, pixels, pixels, line_size, line_size, h);

  113. }

  114. 

  115. static void bfin_put_pixels16_x2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  116. {

  117.     ff_bfin_put_pixels16uc (block, pixels, pixels+1, line_size, line_size, h);

  118. }

  119. 

  120. static void bfin_put_pixels16_y2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  121. {

  122.     ff_bfin_put_pixels16uc (block, pixels, pixels+line_size, line_size, line_size, h);

  123. }

  124. 

  125. static void bfin_put_pixels16_xy2 (uint8_t *block, const uint8_t *s0, int line_size, int h)

  126. {

  127.     ff_bfin_z_put_pixels16_xy2 (block,s0,line_size, line_size, h);

  128. }

  129. 

  130. void bfin_put_pixels8_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  131. {

  132.     ff_bfin_put_pixels8uc_nornd (block, pixels, pixels, line_size, h);

  133. }

  134. 

  135. static void bfin_put_pixels8_x2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  136. {

  137.     ff_bfin_put_pixels8uc_nornd (block, pixels, pixels+1, line_size, h);

  138. }

  139. 

  140. static void bfin_put_pixels8_y2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  141. {

  142.     ff_bfin_put_pixels8uc_nornd (block, pixels, pixels+line_size, line_size, h);

  143. }

  144. 

  145. 

  146. void bfin_put_pixels16_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  147. {

  148.     ff_bfin_put_pixels16uc_nornd (block, pixels, pixels, line_size, h);

  149. }

  150. 

  151. static void bfin_put_pixels16_x2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  152. {

  153.     ff_bfin_put_pixels16uc_nornd (block, pixels, pixels+1, line_size, h);

  154. }

  155. 

  156. static void bfin_put_pixels16_y2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)

  157. {

  158.     ff_bfin_put_pixels16uc_nornd (block, pixels, pixels+line_size, line_size, h);

  159. }

  160. 

  161. static int bfin_pix_abs16 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)

  162. {

  163.     return ff_bfin_z_sad16x16 (blk1,blk2,line_size,line_size,h);

  164. }

  165. 

  166. static int bfin_vsad_intra16 (void *c, uint8_t *blk1, uint8_t *dummy, int stride, int h) {

  167.     return ff_bfin_z_sad16x16 (blk1,blk1+stride,stride<<1,stride<<1,h);

  168. }

  169. 

  170. static int bfin_vsad (void *c, uint8_t *blk1, uint8_t *blk2, int stride, int h) {

  171.     return ff_bfin_z_sad16x16 (blk1,blk1+stride,stride<<1,stride<<1,h)

  172.         + ff_bfin_z_sad16x16 (blk2,blk2+stride,stride<<1,stride<<1,h);

  173. }

  174. 

  175. static uint8_t vtmp_blk[256] attribute_l1_data_b;

  176. 

  177. static int bfin_pix_abs16_x2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)

  178. {

  179.     ff_bfin_put_pixels16uc (vtmp_blk, blk2, blk2+1, 16, line_size, h);

  180.     return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);

  181. }

  182. 

  183. static int bfin_pix_abs16_y2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)

  184. {

  185.     ff_bfin_put_pixels16uc (vtmp_blk, blk2, blk2+line_size, 16, line_size, h);

  186.     return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);

  187. }

  188. 

  189. static int bfin_pix_abs16_xy2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)

  190. {

  191.     ff_bfin_z_put_pixels16_xy2 (vtmp_blk, blk2, 16, line_size, h);

  192.     return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);

  193. }

  194. 

  195. static int bfin_pix_abs8 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)

  196. {

  197.     return ff_bfin_z_sad8x8 (blk1,blk2,line_size,line_size, h);

  198. }

  199. 

  200. static int bfin_pix_abs8_x2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)

  201. {

  202.     ff_bfin_put_pixels8uc (vtmp_blk, blk2, blk2+1, 8, line_size, h);

  203.     return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);

  204. }

  205. 

  206. static int bfin_pix_abs8_y2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)

  207. {

  208.     ff_bfin_put_pixels8uc (vtmp_blk, blk2, blk2+line_size, 8, line_size, h);

  209.     return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);

  210. }

  211. 

  212. static int bfin_pix_abs8_xy2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)

  213. {

  214.     ff_bfin_z_put_pixels8_xy2 (vtmp_blk, blk2, 8, line_size, h);

  215.     return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);

  216. }

  217. 

  218. 

  219. /*

  220.   decoder optimization

  221.   start on 2/11 100 frames of 352x240@25 compiled with no optimization -g debugging

  222.   9.824s ~ 2.44x off

  223.   6.360s ~ 1.58x off with -O2

  224.   5.740s ~ 1.43x off with idcts

  225. 

  226.   2.64s    2/20 same sman.mp4 decode only

  227. 

  228. */

  229. 

  230. void dsputil_init_bfin( DSPContext* c, AVCodecContext *avctx )

  231. {

  232.     c->get_pixels         = ff_bfin_get_pixels;

  233.     c->diff_pixels        = ff_bfin_diff_pixels;

  234.     c->put_pixels_clamped = ff_bfin_put_pixels_clamped;

  235.     c->add_pixels_clamped = ff_bfin_add_pixels_clamped;

  236. 

  237.     c->clear_blocks       = bfin_clear_blocks;

  238.     c->pix_sum            = ff_bfin_pix_sum;

  239.     c->pix_norm1          = ff_bfin_pix_norm1;

  240. 

  241.     c->sad[0]             = bfin_pix_abs16;

  242.     c->sad[1]             = bfin_pix_abs8;

  243. 

  244. /*     c->vsad[0]            = bfin_vsad; */

  245. /*     c->vsad[4]            = bfin_vsad_intra16; */

  246. 

  247.     /* TODO [0] 16  [1] 8 */

  248.     c->pix_abs[0][0] = bfin_pix_abs16;

  249.     c->pix_abs[0][1] = bfin_pix_abs16_x2;

  250.     c->pix_abs[0][2] = bfin_pix_abs16_y2;

  251.     c->pix_abs[0][3] = bfin_pix_abs16_xy2;

  252. 

  253.     c->pix_abs[1][0] = bfin_pix_abs8;

  254.     c->pix_abs[1][1] = bfin_pix_abs8_x2;

  255.     c->pix_abs[1][2] = bfin_pix_abs8_y2;

  256.     c->pix_abs[1][3] = bfin_pix_abs8_xy2;

  257. 

  258. 

  259.     c->sse[0] = ff_bfin_sse16;

  260.     c->sse[1] = ff_bfin_sse8;

  261.     c->sse[2] = ff_bfin_sse4;

  262. 

  263. 

  264.     /**

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

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

  267.      * horizontal blocksizes (8,16) and the 4 halfpel positions

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

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

  270.      * @param pixels source

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

  272.      * @param h height

  273.      */

  274. 

  275.     c->put_pixels_tab[0][0] = bfin_put_pixels16;

  276.     c->put_pixels_tab[0][1] = bfin_put_pixels16_x2;

  277.     c->put_pixels_tab[0][2] = bfin_put_pixels16_y2;

  278.     c->put_pixels_tab[0][3] = bfin_put_pixels16_xy2;

  279. 

  280.     c->put_pixels_tab[1][0] = bfin_put_pixels8;

  281.     c->put_pixels_tab[1][1] = bfin_put_pixels8_x2;

  282.     c->put_pixels_tab[1][2] = bfin_put_pixels8_y2;

  283.     c->put_pixels_tab[1][3] = bfin_put_pixels8_xy2;

  284. 

  285.     c->put_no_rnd_pixels_tab[1][0] = bfin_put_pixels8_nornd;

  286.     c->put_no_rnd_pixels_tab[1][1] = bfin_put_pixels8_x2_nornd;

  287.     c->put_no_rnd_pixels_tab[1][2] = bfin_put_pixels8_y2_nornd;

  288. /*     c->put_no_rnd_pixels_tab[1][3] = ff_bfin_put_pixels8_xy2_nornd; */

  289. 

  290.     c->put_no_rnd_pixels_tab[0][0] = bfin_put_pixels16_nornd;

  291.     c->put_no_rnd_pixels_tab[0][1] = bfin_put_pixels16_x2_nornd;

  292.     c->put_no_rnd_pixels_tab[0][2] = bfin_put_pixels16_y2_nornd;

  293. /*     c->put_no_rnd_pixels_tab[0][3] = ff_bfin_put_pixels16_xy2_nornd; */

  294. 

  295.     if (avctx->dct_algo == FF_DCT_AUTO)

  296.         c->fdct               = ff_bfin_fdct;

  297. 

  298.     if (avctx->idct_algo==FF_IDCT_VP3) {

  299.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  300.         c->idct               = ff_bfin_vp3_idct;

  301.         c->idct_add           = ff_bfin_vp3_idct_add;

  302.         c->idct_put           = ff_bfin_vp3_idct_put;

  303.     } else if (avctx->idct_algo == FF_IDCT_AUTO) {

  304.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  305.         c->idct               = ff_bfin_idct;

  306.         c->idct_add           = bfin_idct_add;

  307.         c->idct_put           = bfin_idct_put;

  308.     }

  309. }

  310. 

  311.