1 /*****************************************************************************
2 * postprocessing_mmx.c: Post Processing library in MMX
3 *****************************************************************************
4 * Copyright (C) 2001 VideoLAN
5 * $Id: postprocessing_mmx.c,v 1.2 2002/12/06 16:34:05 sam Exp $
7 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program 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
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
22 *****************************************************************************/
24 #include <vlc/vlc.h> /* only use u8, u32 .... */
26 #include "postprocessing.h"
27 #include "postprocessing_common.h"
29 /*****************************************************************************
31 * Internals functions common to pp_Deblock_V and pp_Deblock_H
33 *****************************************************************************/
35 /*****************************************************************************
37 *****************************************************************************/
40 /* XXX PP_THR1 need to be defined as ULL */
42 /* Use same things as in idct but how it work ? */
43 #define UNUSED_LONGLONG( foo ) \
44 static const unsigned long long foo __asm__ (#foo) __attribute__((unused))
46 /* to calculate isDC_mode for mmx */
47 UNUSED_LONGLONG( mmx_127_thr1 ) = ( ( 127ULL - PP_THR1 ) << 56 )|
48 ( ( 127ULL - PP_THR1 ) << 48 )|
49 ( ( 127ULL - PP_THR1 ) << 40 )|
50 ( ( 127ULL - PP_THR1 ) << 32 )|
51 ( ( 127ULL - PP_THR1 ) << 24 )|
52 ( ( 127ULL - PP_THR1 ) << 16 )|
53 ( ( 127ULL - PP_THR1 ) << 8 )|
54 ( ( 127ULL - PP_THR1 ) );
56 UNUSED_LONGLONG( mmx_127_2xthr1_1 ) = ( ( 127ULL - PP_2xTHR1 -1) << 56 )|
57 ( ( 127ULL - PP_2xTHR1 -1 ) << 48 )|
58 ( ( 127ULL - PP_2xTHR1 -1 ) << 40 )|
59 ( ( 127ULL - PP_2xTHR1 -1 ) << 32 )|
60 ( ( 127ULL - PP_2xTHR1 -1 ) << 24 )|
61 ( ( 127ULL - PP_2xTHR1 -1 ) << 16 )|
62 ( ( 127ULL - PP_2xTHR1 -1 ) << 8 )|
63 ( ( 127ULL - PP_2xTHR1 -1 ) );
65 UNUSED_LONGLONG( mmx_m2_5_m5_2 ) = 0xfffe0005fffb0002ULL;
68 /* find min bytes from r ans set it in r, t is destroyed */
69 #define MMXEXT_GET_PMIN( r, t ) \
70 "movq " #r ", " #t " \n\
72 pminub " #t ", " #r " \n\
73 pshufw $0xf5, " #r ", " #t " #instead of shift with tmp reg \n\
74 pminub " #t ", " #r " \n\
75 pshufw $0xfe, " #r ", " #t " \n\
76 pminub " #t ", " #r " \n"
78 /* find mzx bytes from r ans set it in r, t is destroyed */
79 #define MMXEXT_GET_PMAX( r, t ) \
80 "movq " #r ", " #t " \n\
82 pmaxub " #t ", " #r " \n\
83 pshufw $0xf5, " #r ", " #t " \n\
84 pmaxub " #t ", " #r " \n\
85 pshufw $0xfe, " #r ", " #t " \n\
86 pmaxub " #t ", " #r " \n"
90 #define MMXEXT_GET_LMINMAX( s, m, M, t ) \
91 "movq " #s ", " #t " \n\
92 pminub " #t ", " #m " \n\
93 pmaxub " #t ", " #M " \n"
99 d1 = a - b with unsigned saturate
106 /****************************************************************************
107 * pp_deblock_isDC_mode : Check if we will use DC mode or Default mode
108 ****************************************************************************
109 * Use constant PP_THR1 and PP_THR2 ( PP_2xTHR1 )
111 * Called for for each pixel on a boundary block when doing deblocking
112 * so need to be fast ...
114 ****************************************************************************/
115 static inline int pp_deblock_isDC_mode( u8 *p_v )
117 unsigned int i_eq_cnt;
120 x = v[i] - v[i+1] without signed saturation
121 ( XXX see if there is'nt problem, but can't be with signed
122 sat because pixel will be saturate :(
123 so x within [-128, 127] and we have to test if it fit in [-M, M]
124 we add 127-M with wrap around -> good value fit in [ 127-2*M, 127]
125 and if x >= 127 - 2 * M ie x > 127 -2*M - 1 value is good
127 __asm__ __volatile__ (" \n\
128 #* Do (v0-v1) to (v7-v8) \n\
129 movq (%1), %%mm1 # load v0->v7 \n\
130 movq 1(%1), %%mm2 # load v1->v8 \n\
131 psubb %%mm2, %%mm1 # v[i]-v[i+1] \n\
132 paddb mmx_127_thr1, %%mm1 # + 127-THR1 with wrap \n\
133 pcmpgtb mmx_127_2xthr1_1, %%mm1 # > 127 -2*thr1 - 1 \n"
134 "movq %%mm1, %%mm0 # \n\
135 psrlw $8, %%mm1 # \n\
136 paddb %%mm1, %%mm0 # \n\
138 movq %%mm0, %%mm1 # Now sum to create eq_cnt \n\
139 psrld $16, %%mm0 # \n\
140 paddb %%mm0, %%mm1 # \n\
142 movq %%mm1, %%mm0 # \n\
143 psrlq $32, %%mm1 # \n\
144 paddb %%mm1, %%mm0 \n"
145 "movd %%mm0, %0 # \n\
149 : "=r"(i_eq_cnt) : "r" (p_v) );
151 /* last test, hey, 9 don't fit in MMX */
152 if(( ( p_v[8] - p_v[9] + PP_THR1 )&0xffff )<= PP_2xTHR1 )
158 /* algo : if ( | v[i] -v[i+1] | <= PP_THR1 ) { i_eq_cnt++; } */
161 for( i =0; i < 9; i++ )
163 if(( ( p_v[i] - p_v[i+1] + PP_THR1 )&0xffff )<= PP_2xTHR1 )
170 return( (i_eq_cnt >= PP_THR2 ) ? 1 : 0 );
173 static inline int pp_deblock_isMinMaxOk( u8 *p_v, int i_QP )
178 __asm__ __volatile__ (
179 "movq 1(%1), %%mm0 # 8 bytes \n"
180 "movq %%mm0, %%mm1 \n"
181 MMXEXT_GET_PMIN( %%mm0, %%mm7 )
182 MMXEXT_GET_PMAX( %%mm1, %%mm7 )
183 "psubd %%mm0, %%mm1 # max - min \n\
185 andl $255, %0" : "=r"(i_range) : "r"(p_v) );
191 i_min = i_max = p_v[1];
192 for( i = 2; i < 9; i++ )
194 if( i_max < p_v[i] ) i_max = p_v[i];
195 if( i_min > p_v[i] ) i_min = p_v[i];
197 i_range = i_max - i_min;
199 return( i_range< 2*i_QP ? 1 : 0 );
203 static inline void pp_deblock_DefaultMode( u8 i_v[10], int i_stride,
207 int a3x0, a3x0_, a3x1, a3x2;
210 /* d = CLIP( 5(a3x0' - a3x0)//8, 0, (v4-v5)/2 ).d( abs(a3x0) < QP ) */
212 /* First calculate a3x0 */
213 __asm__ __volatile__ ( " \n\
214 pxor %%mm7, %%mm7 # mm7 = 0 \n\
215 movq mmx_m2_5_m5_2, %%mm6 # mm6 =(2,-5,5,-2) \n\
216 movd 3(%1), %%mm0 \n\
217 punpcklbw %%mm7,%%mm0 \n\
218 pmaddwd %%mm6, %%mm0 \n"
219 "movq %%mm0, %%mm1 \n\
221 "paddd %%mm1, %%mm0 \n\
222 movd %%mm0, %0" : "=r"(a3x0) :"r"(i_v) );
224 a3x0 = 2 * ( i_v[3] - i_v[6] ) + 5 *( i_v[5] - i_v[4] );
236 /* XXX Now a3x0 is abs( a3x0 ) */
237 if( ( a3x0 < 8 * i_QP )&&( a3x0 != 0 ) ) /* |a3x0| < 8*i_QP */
239 /* calculate a3x1 et a3x2 */
240 __asm__ __volatile__ ( " \n\
242 # mm6 = ( 2, -5, 5, -2 ) \n\
243 movd 1(%2), %%mm0 \n\
244 movd 5(%2), %%mm2 \n\
245 punpcklbw %%mm7,%%mm0 \n\
246 punpcklbw %%mm7,%%mm2 \n\
247 pmaddwd %%mm6, %%mm0 \n\
248 pmaddwd %%mm6, %%mm2 \n"
249 "movq %%mm0, %%mm1 \n\
251 "paddd %%mm1, %%mm0 # mm0 = a3x1 \n\
253 "movq %%mm2, %%mm1 \n\
255 "paddd %%mm1, %%mm2 # mm2 = a3x2 \n\
257 " : "=r"(a3x1), "=r"(a3x2) : "r"(i_v) );
259 a3x1 = 2 * ( i_v[1] - i_v[4] ) + 5 * ( i_v[3] - i_v[2] );
260 a3x2 = 2 * ( i_v[5] - i_v[8] ) + 5 * ( i_v[7] - i_v[6] );
263 if( a3x1 < 0) a3x1 = -a3x1; /* abs( a3x1 ) */
264 if( a3x2 < 0) a3x2 = -a3x2; /* abs( a3x2 ) */
266 a3x0_ = PP_MIN3( a3x0, a3x1, a3x2 );
268 d = 5 *( a3x0 - a3x0_ ) / 8; /* always > 0 */
270 i_delta = ( i_v[4] - i_v[5] ) / 2;
271 /* clip into [0, i_delta] or [i_delta, 0] */
274 if( !b_neg ) /* since true d has sgn(d) = - sgn( a3x0 ) */
277 if( d < i_delta ) d = i_delta;
286 if( d > i_delta ) d = i_delta;
296 static inline void pp_deblock_DCMode( u8 *p_v, /* = int i_v[10] */
301 i_p0 = PP_ABS( p_v[1] - p_v[0] ) < i_QP ? p_v[0] : p_v[1];
302 i_p9 = PP_ABS( p_v[8] - p_v[9] ) < i_QP ? p_v[9] : p_v[8];
304 /* mm0 = 8 pix unmodified
305 -We will process first 4 pixel
306 mm0 = 8 pix unmodified
307 mm1 = for the first part of the 4 first pix
308 (v1) -> (p0) -> ... ( word )
313 = for the commoin part between first and last pix
314 (v2) -> (v3) -> ... ( word )
319 = for the last part of the 4 last pix
320 (v5) -> (v6) -> ... ( word )
325 mm2 = acu for first new pix
326 mm3 = acu for last pix
331 __asm__ __volatile__ (
332 "pxor %%mm7, %%mm7 \n\
333 movq 1(%0), %%mm0 # get 8 pix \n\
334 # unpack into mm1 \n\
335 movq %%mm0, %%mm1 \n\
336 punpcklbw %%mm7, %%mm1 \n\
337 # get p_0 and i_p9 \n\
342 movq %%mm1, %%mm3 # p_v[5-8] = v[1-4] !! \n\
343 movq %%mm1, %%mm2 \n\
344 psllw $2, %%mm2 # p_v[1-4] = 4*v[1-4] \n\
347 por %%mm5, %%mm1 # mm1 =( p0, v1, v2 ,v3)\n\
349 paddw %%mm1, %%mm2 \n\
350 paddw %%mm1, %%mm2 \n\
353 por %%mm5, %%mm1 # mm1 =( p0, p0, v1, v2)\n\
355 paddw %%mm1, %%mm2 \n\
356 paddw %%mm1, %%mm2 \n\
359 por %%mm5, %%mm1 # mm1 =( p0, p0, p0, v1)\n\
361 paddw %%mm1, %%mm2 \n\
364 por %%mm5, %%mm1 # mm1 =( p0, p0, p0, p0)\n\
366 paddw %%mm1, %%mm2 \n\
367 # Now last part a little borring\n\
368 # last part for mm2, beginig for mm3
369 movq %%mm0, %%mm1 \n\
371 punpcklbw %%mm7, %%mm1 # mm1 =( v2, v3, v4, v5 )\n\
372 paddw %%mm1, %%mm2 \n\
373 paddw %%mm1, %%mm2 \n\
374 paddw %%mm1, %%mm3 \n\
377 movq %%mm0, %%mm1 \n\
379 punpcklbw %%mm7, %%mm1 # mm1 =( v3, v4, v5, v6 )\n\
381 paddw %%mm1, %%mm2 \n\
382 paddw %%mm1, %%mm3 \n\
384 movq %%mm0, %%mm1 \n\
386 punpcklbw %%mm7, %%mm1 # mm1 =( v4, v5, v6, v7) \n\
387 paddw %%mm1, %%mm2 \n\
388 paddw %%mm1, %%mm3 \n\
389 paddw %%mm1, %%mm3 \n\
391 movq %%mm0, %%mm1 \n\
393 punpcklbw %%mm7, %%mm1 # mm1 =( v5, v6, v7, v8) \n\
394 paddw %%mm1, %%mm2 \n\
396 paddw %%mm1, %%mm3 \n\
397 # Now last part for last 4 pix \n\
399 movq %%mm0, %%mm1 \n\
400 punpckhbw %%mm7, %%mm1 # mm1 = ( v5, v6, v7, v8) \n\
403 por %%mm6, %%mm1 # mm1 =( v6, v7, v8, p9 )\n\
405 paddw %%mm1, %%mm3 \n\
406 paddw %%mm1, %%mm3 \n\
409 por %%mm6, %%mm1 # mm1 =( v7, v8, p9, p9)\n\
411 paddw %%mm1, %%mm3 \n\
412 paddw %%mm1, %%mm3 \n\
415 por %%mm6, %%mm1 # mm1 =( v8, p9, p9, p9 )\n\
417 paddw %%mm1, %%mm3 \n\
420 por %%mm6, %%mm1 # mm1 =( p9, p9, p9, p9 )\n\
422 paddw %%mm1, %%mm3 \n\
426 packuswb %%mm3, %%mm2 \n\
427 movq %%mm2, 1(%0) \n\
429 ": : "r"(p_v), "r"(i_p0), "r"(i_p9) : "memory" );
432 for( i = 1; i < 9; i++ )
434 v[i] = p_v[i]; /* save 8 pix that will be modified */
437 p_v[1] = ( 6 * i_p0 + 4 * v[1]
438 + 2 *( v[2] + v[3]) + v[4] + v[5]) >> 4;
440 p_v[2] = ( 4 * i_p0 + 2 * v[1] + 4 * v[2]
441 + 2 *( v[3] + v[4]) + v[5] + v[6]) >> 4;
443 p_v[3] = ( 2 * i_p0 + 2 * (v[1] + v[2]) + 4 * v[3]
444 + 2 *( v[4] + v[5]) + v[6] + v[7]) >> 4;
446 p_v[4] = ( i_p0 + v[1] + 2 * (v[2] + v[3]) + 4 * v[4]
447 + 2 *( v[5] + v[6]) + v[7] + v[8]) >> 4;
449 p_v[5] = ( v[1] + v[2] + 2 * (v[3] + v[4]) + 4 * v[5]
450 + 2 *( v[6] + v[7]) + v[8] + i_p9) >> 4;
452 p_v[6] = ( v[2] + v[3] + 2 * (v[4] + v[5]) + 4 * v[6]
453 + 2 *( v[7] + v[8]) + 2 * i_p9) >> 4;
455 p_v[7] = ( v[3] + v[4] + 2 * (v[5] + v[6]) + 4 * v[7]
456 + 2 * v[8] + 4 * i_p9) >> 4;
458 p_v[8] = ( v[4] + v[5] + 2 * (v[6] + v[7]) + 4 * v[8]
466 /*****************************************************************************/
467 /*---------------------------------------------------------------------------*/
469 /* ---------- filter Vertical lines so follow horizontal edges -------- */
471 /*---------------------------------------------------------------------------*/
472 /*****************************************************************************/
474 void E_( pp_deblock_V )( u8 *p_plane,
475 int i_width, int i_height, int i_stride,
476 QT_STORE_T *p_QP_store, int i_QP_stride,
481 int i_QP_scale; /* use to do ( ? >> i_QP_scale ) */
486 i_QP_scale = b_chroma ? 5 : 4 ;
488 for( y = 8; y < i_height - 4; y += 8 )
490 p_v = p_plane + ( y - 5 )* i_stride;
491 for( x = 0; x < i_width; x++ )
493 /* First get 10 vert pix to use them without i_stride */
494 for( i = 0; i < 10; i++ )
496 i_v[i] = p_v[i*i_stride + x];
499 i_QP = p_QP_store[(y>>i_QP_scale)*i_QP_stride+
501 /* XXX QP is for v5 */
502 if( pp_deblock_isDC_mode( i_v ) )
504 if( pp_deblock_isMinMaxOk( i_v, i_QP ) )
506 pp_deblock_DCMode( i_v, i_QP );
511 pp_deblock_DefaultMode( i_v, i_stride, i_QP );
514 /* Copy back, XXX only 1-8 were modified */
515 for( i = 1; i < 9; i++ )
517 p_v[i*i_stride + x] = i_v[i];
525 /*****************************************************************************/
526 /*---------------------------------------------------------------------------*/
528 /* --------- filter Horizontal lines so follow vertical edges -------- */
530 /*---------------------------------------------------------------------------*/
531 /*****************************************************************************/
533 void E_( pp_deblock_H )( u8 *p_plane,
534 int i_width, int i_height, int i_stride,
535 QT_STORE_T *p_QP_store, int i_QP_stride,
543 i_QP_scale = b_chroma ? 5 : 4 ;
545 for( y = 0; y < i_height; y++ )
547 p_v = p_plane + y * i_stride - 5;
548 for( x = 8; x < i_width - 4; x += 8 )
550 /* p_v point 5 pix before a block boundary */
551 /* XXX QP is for v5 */
552 i_QP = p_QP_store[(y>>i_QP_scale)*i_QP_stride+
554 if( pp_deblock_isDC_mode( p_v + x ) )
556 if( pp_deblock_isMinMaxOk( p_v+ x, i_QP ) )
558 pp_deblock_DCMode( p_v+x, i_QP );
563 pp_deblock_DefaultMode( p_v+x, i_stride, i_QP );
572 /*****************************************************************************
574 * Internals functions common to pp_Dering_Y pp_Dering_C
576 *****************************************************************************/
578 static inline void pp_dering_MinMax( u8 *p_block, int i_stride,
579 int *pi_min, int *pi_max )
586 /* First we will extract min/max for each pix on vertical line
587 and next extract global min/max */
588 __asm__ __volatile__(
590 "leal (%2,%3), %%eax \n\
591 movq (%2), %%mm0 #load line \n\
592 movq %%mm0, %%mm1 \n"
594 MMXEXT_GET_LMINMAX( (%%eax), %%mm0, %%mm1, %%mm7 )
595 MMXEXT_GET_LMINMAX( (%%eax, %3), %%mm0, %%mm1, %%mm7 )
596 MMXEXT_GET_LMINMAX( (%%eax, %3,2), %%mm0, %%mm1, %%mm7 )
597 MMXEXT_GET_LMINMAX( (%2, %3, 4), %%mm0, %%mm1, %%mm7 )
598 "leal (%%eax,%3,4), %%eax \n"
600 MMXEXT_GET_LMINMAX( (%%eax), %%mm0, %%mm1, %%mm7 )
601 MMXEXT_GET_LMINMAX( (%%eax, %3), %%mm0, %%mm1, %%mm7 )
602 MMXEXT_GET_LMINMAX( (%%eax, %3,2), %%mm0, %%mm1, %%mm7 )
603 MMXEXT_GET_PMIN( %%mm0, %%mm7 )
604 MMXEXT_GET_PMAX( %%mm1, %%mm7 )
605 "movd %%mm0, %%eax \n\
608 movd %%mm1, %%eax \n\
612 : : "r"(pi_min), "r"(pi_max), "r"(p_block), "r"(i_stride) : "%eax", "memory" );
615 i_min = 255; i_max = 0;
617 for( y = 0; y < 8; y++ )
619 for( x = 0; x < 8; x++ )
621 if( i_min > p_block[x] ) i_min = p_block[x];
622 if( i_max < p_block[x] ) i_max = p_block[x];
632 static inline void pp_dering_BinIndex( u8 *p_block, int i_stride, int i_thr,
638 for( y = 0; y < 10; y++ )
641 for( x = 0; x < 10; x++ )
643 if( p_block[x] > i_thr )
648 i_bin |= (~i_bin) << 16; /* for detect also three 0 */
649 *p_bin = i_bin&( i_bin >> 1 )&( i_bin << 1 );
656 static inline void pp_dering_Filter( u8 *p_block, int i_stride,
670 for( y = 0; y < 8; y++ )
672 i_bin = p_bin[y] & p_bin[y+1] & p_bin[y+2]; /* To be optimised */
673 i_bin |= i_bin >> 16; /* detect 0 or 1 */
675 for( x = 0; x < 8; x++ )
677 if( i_bin&0x02 ) /* 0x02 since 10 index but want 1-9 */
683 i_f = p_block[x - i_stride - 1] +
684 ( p_block[x - i_stride ] << 1)+
685 p_block[x - i_stride + 1] +
687 ( p_block[x - 1] << 1 )+
688 ( p_block[x ] << 2 )+
689 ( p_block[x + 1] << 1 )+
691 p_block[x + i_stride - 1] +
692 ( p_block[x + i_stride ] << 1 ) +
693 p_block[x + i_stride + 1];
695 i_f = ( 8 + i_f ) >> 4;
697 /* Clamp this value */
699 if( i_f - p_block[x] > ( i_QP_2 ) )
701 i_flt[y][x] = p_block[x] + i_QP_2;
704 if( i_f - p_block[x] < -i_QP_2 )
706 i_flt[y][x] = p_block[x] - i_QP_2;
715 i_flt[y][x] = p_block[x];
722 for( y = 0; y < 8; y++ )
724 for( x = 0; x < 8; x++ )
726 p_sav[x] = i_flt[y][x];
733 /*****************************************************************************/
734 /*---------------------------------------------------------------------------*/
736 /* ----------------- Dering filter on Y and C blocks ----------------- */
738 /*---------------------------------------------------------------------------*/
739 /*****************************************************************************/
741 void E_( pp_dering_Y )( u8 *p_plane,
742 int i_width, int i_height, int i_stride,
743 QT_STORE_T *p_QP_store, int i_QP_stride )
746 int i_max[4], i_min[4], i_range[4];
748 int i_max_range, i_kmax;
753 /* We process 4 blocks/loop*/
754 for( y = 8; y < i_height-8; y += 16 )
762 p_block[0] = p_plane + y * i_stride + 8;
763 p_block[1] = p_block[0] + 8;
764 p_block[2] = p_block[0] + ( i_stride << 3 );
765 p_block[3] = p_block[2] + 8;
767 for( x = 8; x < i_width-8; x += 16 )
769 /* 1: Calculate threshold */
770 /* Calculate max/min for each block */
771 pp_dering_MinMax( p_block[0], i_stride, &i_min[0], &i_max[0] );
772 pp_dering_MinMax( p_block[1], i_stride, &i_min[1], &i_max[1] );
773 pp_dering_MinMax( p_block[2], i_stride, &i_min[2], &i_max[2] );
774 pp_dering_MinMax( p_block[3], i_stride, &i_min[3], &i_max[3] );
775 /* Calculate range, max_range and thr */
776 i_max_range = 0; i_kmax = 0;
777 for( k = 0; k <= 4; k++ )
779 i_range[k] = i_max[k] - i_min[k];
780 i_thr[k] = ( i_max[k] + i_min[k] + 1 )/2;
781 if( i_max_range < i_max[k])
783 i_max_range = i_max[k];
787 /* Now rearrange thr */
788 if( i_max_range > 64 )
790 for( k = 1; k < 5; k++ )
792 if( i_range[k] < 16 )
797 if( i_range[k] < 32 )
799 i_thr[k] = i_thr[i_kmax];
805 for( k = 1; k < 5; k++ )
807 if( i_range[k] < 16 )
813 /* 2: Index acquisition 10x10 ! so " -i_stride - 1"*/
814 pp_dering_BinIndex( p_block[0] - i_stride - 1, i_stride,
815 i_thr[0], i_bin[0] );
816 pp_dering_BinIndex( p_block[1] - i_stride - 1, i_stride,
817 i_thr[1], i_bin[1] );
818 pp_dering_BinIndex( p_block[2] - i_stride - 1, i_stride,
819 i_thr[2], i_bin[2] );
820 pp_dering_BinIndex( p_block[3] - i_stride - 1, i_stride,
821 i_thr[3], i_bin[3] );
824 /* 3: adaptive smoothing */
825 /* since we begin at (8,8) QP can be different for each block */
826 p_QP = &( p_QP_store[( y >> 4) * i_QP_stride + (x >> 4)] );
828 pp_dering_Filter( p_block[0], i_stride,
831 pp_dering_Filter( p_block[1], i_stride,
834 pp_dering_Filter( p_block[2], i_stride,
835 i_bin[2], p_QP[i_QP_stride] );
837 pp_dering_Filter( p_block[3], i_stride,
838 i_bin[3], p_QP[i_QP_stride+1] );
849 void E_( pp_dering_C )( u8 *p_plane,
850 int i_width, int i_height, int i_stride,
851 QT_STORE_T *p_QP_store, int i_QP_stride )
861 for( y = 8; y < i_height-8; y += 8 )
864 p_block = p_plane + y * i_stride + 8;
865 for( x = 8; x < i_width-8; x += 8 )
868 /* 1: Calculate threshold */
869 /* Calculate max/min for each block */
870 pp_dering_MinMax( p_block, i_stride,
873 i_thr = ( i_max + i_min + 1 )/2;
875 /* 2: Index acquisition 10x10 */
876 /* point on 10x10 in wich we have our 8x8 block */
877 pp_dering_BinIndex( p_block - i_stride -1, i_stride,
881 /* 3: adaptive smoothing */
882 pp_dering_Filter( p_block, i_stride,
884 p_QP_store[(y>>5)*i_QP_stride+ (x>>5)]);