* H.26L/H.264/AVC/JVT/14496-10/... loop filter
* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
*
- * This file is part of FFmpeg.
+ * This file is part of Libav.
*
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
- * @file libavcodec/h264_loopfilter.c
+ * @file
* H.264 / AVC / MPEG4 part10 loop filter.
* @author Michael Niedermayer <michaelni@gmx.at>
*/
+#include "libavutil/intreadwrite.h"
#include "internal.h"
#include "dsputil.h"
#include "avcodec.h"
};
static void av_always_inline filter_mb_edgev( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h) {
- const unsigned int index_a = qp + h->slice_alpha_c0_offset;
+ const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
+ const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
const int alpha = alpha_table[index_a];
- const int beta = beta_table[qp + h->slice_beta_offset];
+ const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
if (alpha ==0 || beta == 0) return;
if( bS[0] < 4 ) {
tc[1] = tc0_table[index_a][bS[1]];
tc[2] = tc0_table[index_a][bS[2]];
tc[3] = tc0_table[index_a][bS[3]];
- h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
+ h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
} else {
- h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
+ h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
}
}
static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
- const unsigned int index_a = qp + h->slice_alpha_c0_offset;
+ const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
+ const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
const int alpha = alpha_table[index_a];
- const int beta = beta_table[qp + h->slice_beta_offset];
+ const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
if (alpha ==0 || beta == 0) return;
if( bS[0] < 4 ) {
tc[1] = tc0_table[index_a][bS[1]]+1;
tc[2] = tc0_table[index_a][bS[2]]+1;
tc[3] = tc0_table[index_a][bS[3]]+1;
- h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
+ h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
} else {
- h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
+ h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
}
}
-static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
- int i;
- int index_a = qp + h->slice_alpha_c0_offset;
+static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) {
+ const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
+ int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
int alpha = alpha_table[index_a];
- int beta = beta_table[qp + h->slice_beta_offset];
- for( i = 0; i < 8; i++, pix += stride) {
- const int bS_index = (i >> 1) * bsi;
-
- if( bS[bS_index] == 0 ) {
- continue;
- }
-
- if( bS[bS_index] < 4 ) {
- const int tc0 = tc0_table[index_a][bS[bS_index]];
- const int p0 = pix[-1];
- const int p1 = pix[-2];
- const int p2 = pix[-3];
- const int q0 = pix[0];
- const int q1 = pix[1];
- const int q2 = pix[2];
-
- if( FFABS( p0 - q0 ) < alpha &&
- FFABS( p1 - p0 ) < beta &&
- FFABS( q1 - q0 ) < beta ) {
- int tc = tc0;
- int i_delta;
-
- if( FFABS( p2 - p0 ) < beta ) {
- if(tc0)
- pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
- tc++;
- }
- if( FFABS( q2 - q0 ) < beta ) {
- if(tc0)
- pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
- tc++;
- }
-
- i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
- pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
- pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
- tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
- }
- }else{
- const int p0 = pix[-1];
- const int p1 = pix[-2];
- const int p2 = pix[-3];
-
- const int q0 = pix[0];
- const int q1 = pix[1];
- const int q2 = pix[2];
-
- if( FFABS( p0 - q0 ) < alpha &&
- FFABS( p1 - p0 ) < beta &&
- FFABS( q1 - q0 ) < beta ) {
+ int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
+ if (alpha ==0 || beta == 0) return;
- if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
- if( FFABS( p2 - p0 ) < beta)
- {
- const int p3 = pix[-4];
- /* p0', p1', p2' */
- pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
- pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
- pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
- } else {
- /* p0' */
- pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
- }
- if( FFABS( q2 - q0 ) < beta)
- {
- const int q3 = pix[3];
- /* q0', q1', q2' */
- pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
- pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
- pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
- } else {
- /* q0' */
- pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
- }
- }else{
- /* p0', q0' */
- pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
- pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
- }
- tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
- }
- }
+ if( bS[0] < 4 ) {
+ int8_t tc[4];
+ tc[0] = tc0_table[index_a][bS[0*bsi]];
+ tc[1] = tc0_table[index_a][bS[1*bsi]];
+ tc[2] = tc0_table[index_a][bS[2*bsi]];
+ tc[3] = tc0_table[index_a][bS[3*bsi]];
+ h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
+ } else {
+ h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
}
}
-static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
- int i;
- int index_a = qp + h->slice_alpha_c0_offset;
+static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) {
+ const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
+ int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
int alpha = alpha_table[index_a];
- int beta = beta_table[qp + h->slice_beta_offset];
- for( i = 0; i < 4; i++, pix += stride) {
- const int bS_index = i*bsi;
-
- if( bS[bS_index] == 0 ) {
- continue;
- }
+ int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
+ if (alpha ==0 || beta == 0) return;
- if( bS[bS_index] < 4 ) {
- const int tc = tc0_table[index_a][bS[bS_index]] + 1;
- const int p0 = pix[-1];
- const int p1 = pix[-2];
- const int q0 = pix[0];
- const int q1 = pix[1];
-
- if( FFABS( p0 - q0 ) < alpha &&
- FFABS( p1 - p0 ) < beta &&
- FFABS( q1 - q0 ) < beta ) {
- const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
-
- pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
- pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
- tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
- }
- }else{
- const int p0 = pix[-1];
- const int p1 = pix[-2];
- const int q0 = pix[0];
- const int q1 = pix[1];
-
- if( FFABS( p0 - q0 ) < alpha &&
- FFABS( p1 - p0 ) < beta &&
- FFABS( q1 - q0 ) < beta ) {
-
- pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
- pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
- tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
- }
- }
+ if( bS[0] < 4 ) {
+ int8_t tc[4];
+ tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
+ tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
+ tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
+ tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
+ h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
+ } else {
+ h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
}
}
static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
- const unsigned int index_a = qp + h->slice_alpha_c0_offset;
+ const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
+ const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
const int alpha = alpha_table[index_a];
- const int beta = beta_table[qp + h->slice_beta_offset];
+ const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
if (alpha ==0 || beta == 0) return;
if( bS[0] < 4 ) {
tc[1] = tc0_table[index_a][bS[1]];
tc[2] = tc0_table[index_a][bS[2]];
tc[3] = tc0_table[index_a][bS[3]];
- h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
+ h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
} else {
- h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
+ h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
}
}
static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
- const unsigned int index_a = qp + h->slice_alpha_c0_offset;
+ const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
+ const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
const int alpha = alpha_table[index_a];
- const int beta = beta_table[qp + h->slice_beta_offset];
+ const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
if (alpha ==0 || beta == 0) return;
if( bS[0] < 4 ) {
tc[1] = tc0_table[index_a][bS[1]]+1;
tc[2] = tc0_table[index_a][bS[2]]+1;
tc[3] = tc0_table[index_a][bS[3]]+1;
- h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
+ h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
} else {
- h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
+ h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
}
}
mb_xy = h->mb_xy;
- if(!h->top_type || !s->dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
+ if(!h->top_type || !h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
return;
}
return;
} else {
LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
- uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
int edges;
if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
edges = 4;
- bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;
+ AV_WN64A(bS[0][0], 0x0002000200020002ULL);
+ AV_WN64A(bS[0][2], 0x0002000200020002ULL);
+ AV_WN64A(bS[1][0], 0x0002000200020002ULL);
+ AV_WN64A(bS[1][2], 0x0002000200020002ULL);
} else {
int mask_edge1 = (3*(((5*mb_type)>>5)&1)) | (mb_type>>4); //(mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : (mb_type & MB_TYPE_16x8) ? 1 : 0;
int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[0] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0;
int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
- s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
+ h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
}
if( IS_INTRA(left_type) )
- bSv[0][0] = 0x0004000400040004ULL;
+ AV_WN64A(bS[0][0], 0x0004000400040004ULL);
if( IS_INTRA(h->top_type) )
- bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;
+ AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
#define FILTER(hv,dir,edge)\
- if(bSv[dir][edge]) {\
+ if(AV_RN64A(bS[dir][edge])) { \
filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
if(!(edge&1)) {\
filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
int j;
for(j=0; j<2; j++, mbn_xy += s->mb_stride){
- DECLARE_ALIGNED_8(int16_t, bS)[4];
+ DECLARE_ALIGNED(8, int16_t, bS)[4];
int qp;
if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
- *(uint64_t*)bS= 0x0003000300030003ULL;
+ AV_WN64A(bS, 0x0003000300030003ULL);
} else {
if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){
bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]);
( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
}
}else{
- DECLARE_ALIGNED_8(int16_t, bS)[4];
+ DECLARE_ALIGNED(8, int16_t, bS)[4];
int qp;
if( IS_INTRA(mb_type|mbm_type)) {
- *(uint64_t*)bS= 0x0003000300030003ULL;
+ AV_WN64A(bS, 0x0003000300030003ULL);
if ( (!IS_INTERLACED(mb_type|mbm_type))
|| ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
)
- *(uint64_t*)bS= 0x0004000400040004ULL;
+ AV_WN64A(bS, 0x0004000400040004ULL);
} else {
int i;
int mv_done;
if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
- *(uint64_t*)bS= 0x0001000100010001ULL;
+ AV_WN64A(bS, 0x0001000100010001ULL);
mv_done = 1;
}
else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
/* Calculate bS */
for( edge = 1; edge < edges; edge++ ) {
- DECLARE_ALIGNED_8(int16_t, bS)[4];
+ DECLARE_ALIGNED(8, int16_t, bS)[4];
int qp;
if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
continue;
if( IS_INTRA(mb_type)) {
- *(uint64_t*)bS= 0x0003000300030003ULL;
+ AV_WN64A(bS, 0x0003000300030003ULL);
} else {
int i;
int mv_done;
if( edge & mask_edge ) {
- *(uint64_t*)bS= 0;
+ AV_ZERO64(bS);
mv_done = 1;
}
else if( mask_par0 ) {
tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
//{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
if( dir == 0 ) {
- filter_mb_edgev( &img_y[4*edge], linesize, bS, qp, h );
+ filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, h );
if( (edge&1) == 0 ) {
- filter_mb_edgecv( &img_cb[2*edge], uvlinesize, bS, h->chroma_qp[0], h);
- filter_mb_edgecv( &img_cr[2*edge], uvlinesize, bS, h->chroma_qp[1], h);
+ filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
+ filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
}
} else {
filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
/* First vertical edge is different in MBAFF frames
* There are 8 different bS to compute and 2 different Qp
*/
- DECLARE_ALIGNED_8(int16_t, bS)[8];
+ DECLARE_ALIGNED(8, int16_t, bS)[8];
int qp[2];
int bqp[2];
int rqp[2];
int i;
first_vertical_edge_done = 1;
- if( IS_INTRA(mb_type) )
- *(uint64_t*)&bS[0]=
- *(uint64_t*)&bS[4]= 0x0004000400040004ULL;
- else {
+ if( IS_INTRA(mb_type) ) {
+ AV_WN64A(&bS[0], 0x0004000400040004ULL);
+ AV_WN64A(&bS[4], 0x0004000400040004ULL);
+ } else {
static const uint8_t offset[2][2][8]={
{
{7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},