X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=common%2Fframe.c;h=0ece66917c9e2d464b01f85c312c729dc63c4f33;hb=cde39046222b112261179144033e7a51430783d0;hp=a852c911df737d104f9a7e26bedf5d4bdfa1e4c8;hpb=26bed72be4941353b4c79eddf500ce39aaf75490;p=x264 diff --git a/common/frame.c b/common/frame.c index a852c911..0ece6691 100644 --- a/common/frame.c +++ b/common/frame.c @@ -1,10 +1,11 @@ /***************************************************************************** * frame.c: h264 encoder library ***************************************************************************** - * Copyright (C) 2003 Laurent Aimar - * $Id: frame.c,v 1.1 2004/06/03 19:27:06 fenrir Exp $ + * Copyright (C) 2003-2008 x264 project * * Authors: Laurent Aimar + * Loren Merritt + * Fiona Glaser * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -18,93 +19,51 @@ * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA. + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. *****************************************************************************/ -#include -#include -#ifndef _MSC_VER -#include -#endif - #include "common.h" -#define PADH 32 -#define PADV 32 +#define ALIGN(x,a) (((x)+((a)-1))&~((a)-1)) -x264_frame_t *x264_frame_new( x264_t *h ) +x264_frame_t *x264_frame_new( x264_t *h, int b_fdec ) { - x264_frame_t *frame = x264_malloc( sizeof(x264_frame_t) ); + x264_frame_t *frame; int i, j; int i_mb_count = h->mb.i_mb_count; - int i_stride; - int i_lines; + int i_stride, i_width, i_lines; int i_padv = PADV << h->param.b_interlaced; + int luma_plane_size; + int chroma_plane_size; + int align = h->param.cpu&X264_CPU_CACHELINE_64 ? 64 : h->param.cpu&X264_CPU_CACHELINE_32 ? 32 : 16; - if( !frame ) return NULL; - - memset( frame, 0, sizeof(x264_frame_t) ); + CHECKED_MALLOCZERO( frame, sizeof(x264_frame_t) ); /* allocate frame data (+64 for extra data for me) */ - i_stride = ( ( h->param.i_width + 15 ) & -16 )+ 2*PADH; - i_lines = ( ( h->param.i_height + 15 ) & -16 ); - if( h->param.b_interlaced ) - i_lines = ( i_lines + 31 ) & -32; + i_width = ALIGN( h->param.i_width, 16 ); + i_stride = ALIGN( i_width + 2*PADH, align ); + i_lines = ALIGN( h->param.i_height, 16<param.b_interlaced ); frame->i_plane = 3; for( i = 0; i < 3; i++ ) { - int i_divh = 1; - int i_divw = 1; - if( i > 0 ) - { - if( h->param.i_csp == X264_CSP_I420 ) - i_divh = i_divw = 2; - else if( h->param.i_csp == X264_CSP_I422 ) - i_divw = 2; - } - frame->i_stride[i] = i_stride / i_divw; - frame->i_lines[i] = i_lines / i_divh; - CHECKED_MALLOC( frame->buffer[i], - frame->i_stride[i] * ( frame->i_lines[i] + 2*i_padv / i_divh ) ); - - frame->plane[i] = ((uint8_t*)frame->buffer[i]) + - frame->i_stride[i] * i_padv / i_divh + PADH / i_divw; + frame->i_stride[i] = ALIGN( i_stride >> !!i, align ); + frame->i_width[i] = i_width >> !!i; + frame->i_lines[i] = i_lines >> !!i; } - frame->i_stride[3] = 0; - frame->i_lines[3] = 0; - frame->buffer[3] = NULL; - frame->plane[3] = NULL; - frame->filtered[0] = frame->plane[0]; - for( i = 0; i < 3; i++ ) - { - CHECKED_MALLOC( frame->buffer[4+i], - frame->i_stride[0] * ( frame->i_lines[0] + 2*i_padv ) ); - frame->filtered[i+1] = ((uint8_t*)frame->buffer[4+i]) + - frame->i_stride[0] * i_padv + PADH; - } - - if( h->frames.b_have_lowres ) + luma_plane_size = (frame->i_stride[0] * (frame->i_lines[0] + 2*i_padv)); + chroma_plane_size = (frame->i_stride[1] * (frame->i_lines[1] + 2*i_padv)); + for( i = 1; i < 3; i++ ) { - frame->i_stride_lowres = frame->i_stride[0]/2 + PADH; - frame->i_lines_lowres = frame->i_lines[0]/2; - for( i = 0; i < 4; i++ ) - { - CHECKED_MALLOC( frame->buffer_lowres[i], - frame->i_stride_lowres * ( frame->i_lines[0]/2 + 2*i_padv ) ); - frame->lowres[i] = ((uint8_t*)frame->buffer_lowres[i]) + - frame->i_stride_lowres * i_padv + PADH; - } + CHECKED_MALLOC( frame->buffer[i], chroma_plane_size ); + frame->plane[i] = frame->buffer[i] + (frame->i_stride[i] * i_padv + PADH)/2; } - if( h->param.analyse.i_me_method == X264_ME_ESA ) - { - CHECKED_MALLOC( frame->buffer[7], - 2 * frame->i_stride[0] * (frame->i_lines[0] + 2*i_padv) * sizeof(uint16_t) ); - frame->integral = (uint16_t*)frame->buffer[7] + frame->i_stride[0] * i_padv + PADH; - } + for( i = 0; i < h->param.i_bframe + 2; i++ ) + for( j = 0; j < h->param.i_bframe + 2; j++ ) + CHECKED_MALLOC( frame->i_row_satds[i][j], i_lines/16 * sizeof(int) ); frame->i_poc = -1; frame->i_type = X264_TYPE_AUTO; @@ -113,70 +72,175 @@ x264_frame_t *x264_frame_new( x264_t *h ) frame->i_frame = -1; frame->i_frame_num = -1; frame->i_lines_completed = -1; + frame->b_fdec = b_fdec; + frame->orig = frame; - CHECKED_MALLOC( frame->mb_type, i_mb_count * sizeof(int8_t)); - CHECKED_MALLOC( frame->mv[0], 2*16 * i_mb_count * sizeof(int16_t) ); - CHECKED_MALLOC( frame->ref[0], 4 * i_mb_count * sizeof(int8_t) ); - if( h->param.i_bframe ) + /* all 4 luma planes allocated together, since the cacheline split code + * requires them to be in-phase wrt cacheline alignment. */ + if( h->param.analyse.i_subpel_refine && b_fdec ) { - CHECKED_MALLOC( frame->mv[1], 2*16 * i_mb_count * sizeof(int16_t) ); - CHECKED_MALLOC( frame->ref[1], 4 * i_mb_count * sizeof(int8_t) ); + CHECKED_MALLOC( frame->buffer[0], 4*luma_plane_size ); + for( i = 0; i < 4; i++ ) + frame->filtered[i] = frame->buffer[0] + i*luma_plane_size + frame->i_stride[0] * i_padv + PADH; + frame->plane[0] = frame->filtered[0]; } else { - frame->mv[1] = NULL; - frame->ref[1] = NULL; + CHECKED_MALLOC( frame->buffer[0], luma_plane_size ); + frame->filtered[0] = frame->plane[0] = frame->buffer[0] + frame->i_stride[0] * i_padv + PADH; } - CHECKED_MALLOC( frame->i_row_bits, i_lines/16 * sizeof(int) ); - CHECKED_MALLOC( frame->i_row_qp, i_lines/16 * sizeof(int) ); - for( i = 0; i < h->param.i_bframe + 2; i++ ) - for( j = 0; j < h->param.i_bframe + 2; j++ ) - CHECKED_MALLOC( frame->i_row_satds[i][j], i_lines/16 * sizeof(int) ); + frame->b_duplicate = 0; + + if( b_fdec ) /* fdec frame */ + { + CHECKED_MALLOC( frame->mb_type, i_mb_count * sizeof(int8_t)); + CHECKED_MALLOC( frame->mv[0], 2*16 * i_mb_count * sizeof(int16_t) ); + CHECKED_MALLOC( frame->ref[0], 4 * i_mb_count * sizeof(int8_t) ); + if( h->param.i_bframe ) + { + CHECKED_MALLOC( frame->mv[1], 2*16 * i_mb_count * sizeof(int16_t) ); + CHECKED_MALLOC( frame->ref[1], 4 * i_mb_count * sizeof(int8_t) ); + } + else + { + frame->mv[1] = NULL; + frame->ref[1] = NULL; + } + CHECKED_MALLOC( frame->i_row_bits, i_lines/16 * sizeof(int) ); + CHECKED_MALLOC( frame->i_row_qp, i_lines/16 * sizeof(int) ); + if( h->param.analyse.i_me_method >= X264_ME_ESA ) + { + CHECKED_MALLOC( frame->buffer[3], + frame->i_stride[0] * (frame->i_lines[0] + 2*i_padv) * sizeof(uint16_t) << h->frames.b_have_sub8x8_esa ); + frame->integral = (uint16_t*)frame->buffer[3] + frame->i_stride[0] * i_padv + PADH; + } + } + else /* fenc frame */ + { + if( h->frames.b_have_lowres ) + { + frame->i_width_lowres = frame->i_width[0]/2; + frame->i_stride_lowres = ALIGN( frame->i_width_lowres + 2*PADH, align ); + frame->i_lines_lowres = frame->i_lines[0]/2; + + luma_plane_size = frame->i_stride_lowres * (frame->i_lines[0]/2 + 2*i_padv); + + CHECKED_MALLOC( frame->buffer_lowres[0], 4 * luma_plane_size ); + for( i = 0; i < 4; i++ ) + frame->lowres[i] = frame->buffer_lowres[0] + (frame->i_stride_lowres * i_padv + PADH) + i * luma_plane_size; + + for( j = 0; j <= !!h->param.i_bframe; j++ ) + for( i = 0; i <= h->param.i_bframe; i++ ) + { + CHECKED_MALLOCZERO( frame->lowres_mvs[j][i], 2*h->mb.i_mb_count*sizeof(int16_t) ); + CHECKED_MALLOC( frame->lowres_mv_costs[j][i], h->mb.i_mb_count*sizeof(int) ); + } + CHECKED_MALLOC( frame->i_propagate_cost, (i_mb_count+3) * sizeof(uint16_t) ); + for( j = 0; j <= h->param.i_bframe+1; j++ ) + for( i = 0; i <= h->param.i_bframe+1; i++ ) + { + CHECKED_MALLOC( frame->lowres_costs[j][i], (i_mb_count+3) * sizeof(uint16_t) ); + CHECKED_MALLOC( frame->lowres_inter_types[j][i], (i_mb_count+3)/4 * sizeof(uint8_t) ); + } + frame->i_intra_cost = frame->lowres_costs[0][0]; + memset( frame->i_intra_cost, -1, (i_mb_count+3) * sizeof(uint16_t) ); + } + if( h->param.rc.i_aq_mode ) + { + CHECKED_MALLOC( frame->f_qp_offset, h->mb.i_mb_count * sizeof(float) ); + CHECKED_MALLOC( frame->f_qp_offset_aq, h->mb.i_mb_count * sizeof(float) ); + if( h->frames.b_have_lowres ) + /* shouldn't really be initialized, just silences a valgrind false-positive in x264_mbtree_propagate_cost_sse2 */ + CHECKED_MALLOCZERO( frame->i_inv_qscale_factor, (h->mb.i_mb_count+3) * sizeof(uint16_t) ); + } + } - pthread_mutex_init( &frame->mutex, NULL ); - pthread_cond_init( &frame->cv, NULL ); + if( x264_pthread_mutex_init( &frame->mutex, NULL ) ) + goto fail; + if( x264_pthread_cond_init( &frame->cv, NULL ) ) + goto fail; return frame; fail: - x264_frame_delete( frame ); + x264_free( frame ); return NULL; } void x264_frame_delete( x264_frame_t *frame ) { int i, j; - for( i = 0; i < 8; i++ ) - x264_free( frame->buffer[i] ); - for( i = 0; i < 4; i++ ) - x264_free( frame->buffer_lowres[i] ); - for( i = 0; i < X264_BFRAME_MAX+2; i++ ) - for( j = 0; j < X264_BFRAME_MAX+2; j++ ) - x264_free( frame->i_row_satds[i][j] ); - x264_free( frame->i_row_bits ); - x264_free( frame->i_row_qp ); - x264_free( frame->mb_type ); - x264_free( frame->mv[0] ); - x264_free( frame->mv[1] ); - x264_free( frame->ref[0] ); - x264_free( frame->ref[1] ); - pthread_mutex_destroy( &frame->mutex ); - pthread_cond_destroy( &frame->cv ); + /* Duplicate frames are blank copies of real frames (including pointers), + * so freeing those pointers would cause a double free later. */ + if( !frame->b_duplicate ) + { + for( i = 0; i < 4; i++ ) + x264_free( frame->buffer[i] ); + for( i = 0; i < 4; i++ ) + x264_free( frame->buffer_lowres[i] ); + for( i = 0; i < X264_BFRAME_MAX+2; i++ ) + for( j = 0; j < X264_BFRAME_MAX+2; j++ ) + x264_free( frame->i_row_satds[i][j] ); + for( j = 0; j < 2; j++ ) + for( i = 0; i <= X264_BFRAME_MAX; i++ ) + { + x264_free( frame->lowres_mvs[j][i] ); + x264_free( frame->lowres_mv_costs[j][i] ); + } + x264_free( frame->i_propagate_cost ); + for( j = 0; j <= X264_BFRAME_MAX+1; j++ ) + for( i = 0; i <= X264_BFRAME_MAX+1; i++ ) + { + x264_free( frame->lowres_costs[j][i] ); + x264_free( frame->lowres_inter_types[j][i] ); + } + x264_free( frame->f_qp_offset ); + x264_free( frame->f_qp_offset_aq ); + x264_free( frame->i_inv_qscale_factor ); + x264_free( frame->i_row_bits ); + x264_free( frame->i_row_qp ); + x264_free( frame->mb_type ); + x264_free( frame->mv[0] ); + x264_free( frame->mv[1] ); + x264_free( frame->ref[0] ); + x264_free( frame->ref[1] ); + x264_pthread_mutex_destroy( &frame->mutex ); + x264_pthread_cond_destroy( &frame->cv ); + } x264_free( frame ); } -void x264_frame_copy_picture( x264_t *h, x264_frame_t *dst, x264_picture_t *src ) +int x264_frame_copy_picture( x264_t *h, x264_frame_t *dst, x264_picture_t *src ) { int i_csp = src->img.i_csp & X264_CSP_MASK; + int i; + if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 ) + { + x264_log( h, X264_LOG_ERROR, "Arg invalid CSP\n" ); + return -1; + } + dst->i_type = src->i_type; dst->i_qpplus1 = src->i_qpplus1; - dst->i_pts = src->i_pts; + dst->i_pts = dst->i_dts = src->i_pts; + dst->param = src->param; - if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX ) - x264_log( h, X264_LOG_ERROR, "Arg invalid CSP\n" ); - else - h->csp.convert[i_csp]( &h->mc, dst, &src->img, h->param.i_width, h->param.i_height ); + for( i=0; i<3; i++ ) + { + int s = (i_csp == X264_CSP_YV12 && i) ? i^3 : i; + uint8_t *plane = src->img.plane[s]; + int stride = src->img.i_stride[s]; + int width = h->param.i_width >> !!i; + int height = h->param.i_height >> !!i; + if( src->img.i_csp & X264_CSP_VFLIP ) + { + plane += (height-1)*stride; + stride = -stride; + } + h->mc.plane_copy( dst->plane[i], dst->i_stride[i], plane, stride, width, height ); + } + return 0; } @@ -234,19 +298,20 @@ void x264_frame_expand_border( x264_t *h, x264_frame_t *frame, int mb_y, int b_e void x264_frame_expand_border_filtered( x264_t *h, x264_frame_t *frame, int mb_y, int b_end ) { - /* during filtering, 8 extra pixels were filtered on each edge. + /* during filtering, 8 extra pixels were filtered on each edge, + * but up to 3 of the horizontal ones may be wrong. we want to expand border from the last filtered pixel */ int b_start = !mb_y; int stride = frame->i_stride[0]; - int width = 16*h->sps->i_mb_width + 16; + int width = 16*h->sps->i_mb_width + 8; int height = b_end ? (16*(h->sps->i_mb_height - mb_y) >> h->sh.b_mbaff) + 16 : 16; - int padh = PADH - 8; + int padh = PADH - 4; int padv = PADV - 8; int i; for( i = 1; i < 4; i++ ) { // buffer: 8 luma, to match the hpel filter - uint8_t *pix = frame->filtered[i] + (16*mb_y - (8 << h->sh.b_mbaff)) * stride - 8; + uint8_t *pix = frame->filtered[i] + (16*mb_y - (8 << h->sh.b_mbaff)) * stride - 4; if( h->sh.b_mbaff ) { plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end ); @@ -263,7 +328,7 @@ void x264_frame_expand_border_lowres( x264_frame_t *frame ) { int i; for( i = 0; i < 4; i++ ) - plane_expand_border( frame->lowres[i], frame->i_stride_lowres, frame->i_stride_lowres - 2*PADH, frame->i_lines_lowres, PADH, PADV, 1, 1 ); + plane_expand_border( frame->lowres[i], frame->i_stride_lowres, frame->i_width_lowres, frame->i_lines_lowres, PADH, PADV, 1, 1 ); } void x264_frame_expand_border_mod16( x264_t *h, x264_frame_t *frame ) @@ -274,8 +339,8 @@ void x264_frame_expand_border_mod16( x264_t *h, x264_frame_t *frame ) int i_subsample = i ? 1 : 0; int i_width = h->param.i_width >> i_subsample; int i_height = h->param.i_height >> i_subsample; - int i_padx = ( h->sps->i_mb_width * 16 - h->param.i_width ) >> i_subsample; - int i_pady = ( h->sps->i_mb_height * 16 - h->param.i_height ) >> i_subsample; + int i_padx = (h->sps->i_mb_width * 16 - h->param.i_width) >> i_subsample; + int i_pady = (h->sps->i_mb_height * 16 - h->param.i_height) >> i_subsample; if( i_padx ) { @@ -298,20 +363,20 @@ void x264_frame_expand_border_mod16( x264_t *h, x264_frame_t *frame ) /* cavlc + 8x8 transform stores nnz per 16 coeffs for the purpose of * entropy coding, but per 64 coeffs for the purpose of deblocking */ -void munge_cavlc_nnz_row( x264_t *h, int mb_y, uint8_t (*buf)[16] ) +static void munge_cavlc_nnz_row( x264_t *h, int mb_y, uint8_t (*buf)[16] ) { uint32_t (*src)[6] = (uint32_t(*)[6])h->mb.non_zero_count + mb_y * h->sps->i_mb_width; int8_t *transform = h->mb.mb_transform_size + mb_y * h->sps->i_mb_width; - int x; + int x, nnz; for( x=0; xsps->i_mb_width; x++ ) { memcpy( buf+x, src+x, 16 ); if( transform[x] ) { - if( src[x][0] ) src[x][0] = 0x01010101; - if( src[x][1] ) src[x][1] = 0x01010101; - if( src[x][2] ) src[x][2] = 0x01010101; - if( src[x][3] ) src[x][3] = 0x01010101; + nnz = src[x][0] | src[x][1]; + src[x][0] = src[x][1] = ((uint16_t)nnz ? 0x0101 : 0) + (nnz>>16 ? 0x01010000 : 0); + nnz = src[x][2] | src[x][3]; + src[x][2] = src[x][3] = ((uint16_t)nnz ? 0x0101 : 0) + (nnz>>16 ? 0x01010000 : 0); } } } @@ -339,82 +404,86 @@ static void munge_cavlc_nnz( x264_t *h, int mb_y, uint8_t (*buf)[16], void (*fun /* Deblocking filter */ - -static const int i_alpha_table[52] = +static const uint8_t i_alpha_table[52+12*2] = { + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 4, 4, 5, 6, 7, 8, 9, 10, 12, 13, 15, 17, 20, 22, 25, 28, 32, 36, 40, 45, 50, 56, 63, 71, 80, 90,101,113,127,144,162,182,203,226, - 255, 255 + 255,255, + 255,255,255,255,255,255,255,255,255,255,255,255, }; -static const int i_beta_table[52] = +static const uint8_t i_beta_table[52+12*2] = { + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14, 14, 15, 15, 16, 16, 17, 17, - 18, 18 + 18, 18, + 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, }; -static const int i_tc0_table[52][3] = -{ - { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, - { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, - { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 0 }, { 0, 0, 1 }, - { 0, 0, 1 }, { 0, 0, 1 }, { 0, 0, 1 }, { 0, 1, 1 }, { 0, 1, 1 }, { 1, 1, 1 }, - { 1, 1, 1 }, { 1, 1, 1 }, { 1, 1, 1 }, { 1, 1, 2 }, { 1, 1, 2 }, { 1, 1, 2 }, - { 1, 1, 2 }, { 1, 2, 3 }, { 1, 2, 3 }, { 2, 2, 3 }, { 2, 2, 4 }, { 2, 3, 4 }, - { 2, 3, 4 }, { 3, 3, 5 }, { 3, 4, 6 }, { 3, 4, 6 }, { 4, 5, 7 }, { 4, 5, 8 }, - { 4, 6, 9 }, { 5, 7,10 }, { 6, 8,11 }, { 6, 8,13 }, { 7,10,14 }, { 8,11,16 }, - { 9,12,18 }, {10,13,20 }, {11,15,23 }, {13,17,25 } +static const int8_t i_tc0_table[52+12*2][4] = +{ + {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, + {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, + {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, + {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, + {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 1 }, + {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 }, + {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, + {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 }, + {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 }, + {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 }, + {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-1,13,17,25 }, + {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, + {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, }; +#define alpha_table(x) i_alpha_table[(x)+12] +#define beta_table(x) i_beta_table[(x)+12] +#define tc0_table(x) i_tc0_table[(x)+12] /* From ffmpeg */ -static inline int clip_uint8( int a ) -{ - if (a&(~255)) - return (-a)>>31; - else - return a; -} - static inline void deblock_luma_c( uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0 ) { int i, d; - for( i = 0; i < 4; i++ ) { - if( tc0[i] < 0 ) { + for( i = 0; i < 4; i++ ) + { + if( tc0[i] < 0 ) + { pix += 4*ystride; continue; } - for( d = 0; d < 4; d++ ) { + for( d = 0; d < 4; d++ ) + { const int p2 = pix[-3*xstride]; const int p1 = pix[-2*xstride]; const int p0 = pix[-1*xstride]; const int q0 = pix[ 0*xstride]; const int q1 = pix[ 1*xstride]; const int q2 = pix[ 2*xstride]; - - if( abs( p0 - q0 ) < alpha && - abs( p1 - p0 ) < beta && - abs( q1 - q0 ) < beta ) { - + + if( abs( p0 - q0 ) < alpha && abs( p1 - p0 ) < beta && abs( q1 - q0 ) < beta ) + { int tc = tc0[i]; int delta; - - if( abs( p2 - p0 ) < beta ) { + if( abs( p2 - p0 ) < beta ) + { pix[-2*xstride] = p1 + x264_clip3( (( p2 + ((p0 + q0 + 1) >> 1)) >> 1) - p1, -tc0[i], tc0[i] ); - tc++; + tc++; } - if( abs( q2 - q0 ) < beta ) { + if( abs( q2 - q0 ) < beta ) + { pix[ 1*xstride] = q1 + x264_clip3( (( q2 + ((p0 + q0 + 1) >> 1)) >> 1) - q1, -tc0[i], tc0[i] ); tc++; } - + delta = x264_clip3( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); - pix[-1*xstride] = clip_uint8( p0 + delta ); /* p0' */ - pix[ 0*xstride] = clip_uint8( q0 - delta ); /* q0' */ + pix[-1*xstride] = x264_clip_uint8( p0 + delta ); /* p0' */ + pix[ 0*xstride] = x264_clip_uint8( q0 - delta ); /* q0' */ } pix += ystride; } @@ -422,7 +491,7 @@ static inline void deblock_luma_c( uint8_t *pix, int xstride, int ystride, int a } static void deblock_v_luma_c( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ) { - deblock_luma_c( pix, stride, 1, alpha, beta, tc0 ); + deblock_luma_c( pix, stride, 1, alpha, beta, tc0 ); } static void deblock_h_luma_c( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ) { @@ -432,43 +501,45 @@ static void deblock_h_luma_c( uint8_t *pix, int stride, int alpha, int beta, int static inline void deblock_chroma_c( uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0 ) { int i, d; - for( i = 0; i < 4; i++ ) { + for( i = 0; i < 4; i++ ) + { const int tc = tc0[i]; - if( tc <= 0 ) { + if( tc <= 0 ) + { pix += 2*ystride; continue; } - for( d = 0; d < 2; d++ ) { + for( d = 0; d < 2; d++ ) + { const int p1 = pix[-2*xstride]; const int p0 = pix[-1*xstride]; const int q0 = pix[ 0*xstride]; const int q1 = pix[ 1*xstride]; - if( abs( p0 - q0 ) < alpha && - abs( p1 - p0 ) < beta && - abs( q1 - q0 ) < beta ) { - + if( abs( p0 - q0 ) < alpha && abs( p1 - p0 ) < beta && abs( q1 - q0 ) < beta ) + { int delta = x264_clip3( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); - pix[-1*xstride] = clip_uint8( p0 + delta ); /* p0' */ - pix[ 0*xstride] = clip_uint8( q0 - delta ); /* q0' */ + pix[-1*xstride] = x264_clip_uint8( p0 + delta ); /* p0' */ + pix[ 0*xstride] = x264_clip_uint8( q0 - delta ); /* q0' */ } pix += ystride; } } } static void deblock_v_chroma_c( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ) -{ +{ deblock_chroma_c( pix, stride, 1, alpha, beta, tc0 ); } static void deblock_h_chroma_c( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ) -{ +{ deblock_chroma_c( pix, 1, stride, alpha, beta, tc0 ); } static inline void deblock_luma_intra_c( uint8_t *pix, int xstride, int ystride, int alpha, int beta ) { int d; - for( d = 0; d < 16; d++ ) { + for( d = 0; d < 16; d++ ) + { const int p2 = pix[-3*xstride]; const int p1 = pix[-2*xstride]; const int p0 = pix[-1*xstride]; @@ -476,35 +547,31 @@ static inline void deblock_luma_intra_c( uint8_t *pix, int xstride, int ystride, const int q1 = pix[ 1*xstride]; const int q2 = pix[ 2*xstride]; - if( abs( p0 - q0 ) < alpha && - abs( p1 - p0 ) < beta && - abs( q1 - q0 ) < beta ) { - - if(abs( p0 - q0 ) < ((alpha >> 2) + 2) ){ - if( abs( p2 - p0 ) < beta) + if( abs( p0 - q0 ) < alpha && abs( p1 - p0 ) < beta && abs( q1 - q0 ) < beta ) + { + if(abs( p0 - q0 ) < ((alpha >> 2) + 2) ) + { + if( abs( p2 - p0 ) < beta ) /* p0', p1', p2' */ { const int p3 = pix[-4*xstride]; - /* p0', p1', p2' */ pix[-1*xstride] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; pix[-2*xstride] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; pix[-3*xstride] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; - } else { - /* p0' */ - pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2; } - if( abs( q2 - q0 ) < beta) + else /* p0' */ + pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2; + if( abs( q2 - q0 ) < beta ) /* q0', q1', q2' */ { const int q3 = pix[3*xstride]; - /* q0', q1', q2' */ pix[0*xstride] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; pix[1*xstride] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; pix[2*xstride] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; - } else { - /* q0' */ - pix[0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2; } - }else{ - /* p0', q0' */ + else /* q0' */ + pix[0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2; + } + else /* p0', q0' */ + { pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2; pix[ 0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2; } @@ -513,59 +580,72 @@ static inline void deblock_luma_intra_c( uint8_t *pix, int xstride, int ystride, } } static void deblock_v_luma_intra_c( uint8_t *pix, int stride, int alpha, int beta ) -{ +{ deblock_luma_intra_c( pix, stride, 1, alpha, beta ); } static void deblock_h_luma_intra_c( uint8_t *pix, int stride, int alpha, int beta ) -{ +{ deblock_luma_intra_c( pix, 1, stride, alpha, beta ); } static inline void deblock_chroma_intra_c( uint8_t *pix, int xstride, int ystride, int alpha, int beta ) -{ - int d; - for( d = 0; d < 8; d++ ) { +{ + int d; + for( d = 0; d < 8; d++ ) + { const int p1 = pix[-2*xstride]; const int p0 = pix[-1*xstride]; const int q0 = pix[ 0*xstride]; const int q1 = pix[ 1*xstride]; - if( abs( p0 - q0 ) < alpha && - abs( p1 - p0 ) < beta && - abs( q1 - q0 ) < beta ) { - + if( abs( p0 - q0 ) < alpha && abs( p1 - p0 ) < beta && abs( q1 - q0 ) < beta ) + { pix[-1*xstride] = (2*p1 + p0 + q1 + 2) >> 2; /* p0' */ pix[ 0*xstride] = (2*q1 + q0 + p1 + 2) >> 2; /* q0' */ } - pix += ystride; } } static void deblock_v_chroma_intra_c( uint8_t *pix, int stride, int alpha, int beta ) -{ +{ deblock_chroma_intra_c( pix, stride, 1, alpha, beta ); } static void deblock_h_chroma_intra_c( uint8_t *pix, int stride, int alpha, int beta ) -{ +{ deblock_chroma_intra_c( pix, 1, stride, alpha, beta ); } -static inline void deblock_edge( x264_t *h, uint8_t *pix, int i_stride, int bS[4], int i_qp, int b_chroma, - x264_deblock_inter_t pf_inter, x264_deblock_intra_t pf_intra ) +static inline void deblock_edge( x264_t *h, uint8_t *pix1, uint8_t *pix2, int i_stride, uint8_t bS[4], int i_qp, int b_chroma, x264_deblock_inter_t pf_inter ) { - int i; - const int index_a = x264_clip3( i_qp + h->sh.i_alpha_c0_offset, 0, 51 ); - const int alpha = i_alpha_table[index_a]; - const int beta = i_beta_table[x264_clip3( i_qp + h->sh.i_beta_offset, 0, 51 )]; - - if( bS[0] < 4 ) { - int8_t tc[4]; - for(i=0; i<4; i++) - tc[i] = (bS[i] ? i_tc0_table[index_a][bS[i] - 1] : -1) + b_chroma; - pf_inter( pix, i_stride, alpha, beta, tc ); - } else { - pf_intra( pix, i_stride, alpha, beta ); - } + const int index_a = i_qp + h->sh.i_alpha_c0_offset; + const int alpha = alpha_table(index_a); + const int beta = beta_table(i_qp + h->sh.i_beta_offset); + int8_t tc[4]; + + if( !alpha || !beta ) + return; + + tc[0] = tc0_table(index_a)[bS[0]] + b_chroma; + tc[1] = tc0_table(index_a)[bS[1]] + b_chroma; + tc[2] = tc0_table(index_a)[bS[2]] + b_chroma; + tc[3] = tc0_table(index_a)[bS[3]] + b_chroma; + + pf_inter( pix1, i_stride, alpha, beta, tc ); + if( b_chroma ) + pf_inter( pix2, i_stride, alpha, beta, tc ); +} + +static inline void deblock_edge_intra( x264_t *h, uint8_t *pix1, uint8_t *pix2, int i_stride, uint8_t bS[4], int i_qp, int b_chroma, x264_deblock_intra_t pf_intra ) +{ + const int alpha = alpha_table(i_qp + h->sh.i_alpha_c0_offset); + const int beta = beta_table(i_qp + h->sh.i_beta_offset); + + if( !alpha || !beta ) + return; + + pf_intra( pix1, i_stride, alpha, beta ); + if( b_chroma ) + pf_intra( pix2, i_stride, alpha, beta ); } void x264_frame_deblock_row( x264_t *h, int mb_y ) @@ -574,161 +654,172 @@ void x264_frame_deblock_row( x264_t *h, int mb_y ) const int s4x4 = 4 * h->mb.i_mb_stride; const int b_interlaced = h->sh.b_mbaff; const int mvy_limit = 4 >> b_interlaced; + const int qp_thresh = 15 - X264_MIN(h->sh.i_alpha_c0_offset, h->sh.i_beta_offset) - X264_MAX(0, h->param.analyse.i_chroma_qp_offset); + const int no_sub8x8 = !(h->param.analyse.inter & X264_ANALYSE_PSUB8x8); int mb_x; - - int i_stride2[3] = { h->fdec->i_stride[0] << b_interlaced, - h->fdec->i_stride[1] << b_interlaced, - h->fdec->i_stride[2] << b_interlaced }; + int stridey = h->fdec->i_stride[0]; + int stride2y = stridey << b_interlaced; + int strideuv = h->fdec->i_stride[1]; + int stride2uv = strideuv << b_interlaced; if( !h->pps->b_cabac && h->pps->b_transform_8x8_mode ) munge_cavlc_nnz( h, mb_y, h->mb.nnz_backup, munge_cavlc_nnz_row ); - for( mb_x = 0; mb_x < h->sps->i_mb_width; ) + for( mb_x = 0; mb_x < h->sps->i_mb_width; mb_x += (~b_interlaced | mb_y)&1, mb_y ^= b_interlaced ) { const int mb_xy = mb_y * h->mb.i_mb_stride + mb_x; const int mb_8x8 = 2 * s8x8 * mb_y + 2 * mb_x; const int mb_4x4 = 4 * s4x4 * mb_y + 4 * mb_x; const int b_8x8_transform = h->mb.mb_transform_size[mb_xy]; - const int i_edge_end = (h->mb.type[mb_xy] == P_SKIP) ? 1 : 4; - int i_edge, i_dir; - - int i_pix_y[3] = { 16*mb_y*h->fdec->i_stride[0] + 16*mb_x, - 8*mb_y*h->fdec->i_stride[1] + 8*mb_x, - 8*mb_y*h->fdec->i_stride[2] + 8*mb_x }; + const int i_qp = h->mb.qp[mb_xy]; + int i_edge_end = (h->mb.type[mb_xy] == P_SKIP) ? 1 : 4; + uint8_t *pixy = h->fdec->plane[0] + 16*mb_y*stridey + 16*mb_x; + uint8_t *pixu = h->fdec->plane[1] + 8*mb_y*strideuv + 8*mb_x; + uint8_t *pixv = h->fdec->plane[2] + 8*mb_y*strideuv + 8*mb_x; if( b_interlaced && (mb_y&1) ) { - i_pix_y[0] -= 15*h->fdec->i_stride[0]; - i_pix_y[1] -= 7*h->fdec->i_stride[1]; - i_pix_y[2] -= 7*h->fdec->i_stride[2]; + pixy -= 15*stridey; + pixu -= 7*strideuv; + pixv -= 7*strideuv; } x264_prefetch_fenc( h, h->fdec, mb_x, mb_y ); - /* i_dir == 0 -> vertical edge - * i_dir == 1 -> horizontal edge */ - for( i_dir = 0; i_dir < 2; i_dir++ ) - { - int i_start = (i_dir ? (mb_y <= b_interlaced) : (mb_x == 0)); - int i_qp, i_qpn; - - for( i_edge = i_start; i_edge < i_edge_end; i_edge++ ) - { - int mbn_xy, mbn_8x8, mbn_4x4; - int bS[4]; /* filtering strength */ - - if( b_8x8_transform && (i_edge&1) ) - continue; - - mbn_xy = i_edge > 0 ? mb_xy : ( i_dir == 0 ? mb_xy - 1 : mb_xy - h->mb.i_mb_stride ); - mbn_8x8 = i_edge > 0 ? mb_8x8 : ( i_dir == 0 ? mb_8x8 - 2 : mb_8x8 - 2 * s8x8 ); - mbn_4x4 = i_edge > 0 ? mb_4x4 : ( i_dir == 0 ? mb_4x4 - 4 : mb_4x4 - 4 * s4x4 ); - - if( b_interlaced && i_edge == 0 && i_dir == 1 ) - { - mbn_xy -= h->mb.i_mb_stride; - mbn_8x8 -= 2 * s8x8; - mbn_4x4 -= 4 * s4x4; - } - - /* *** Get bS for each 4px for the current edge *** */ - if( IS_INTRA( h->mb.type[mb_xy] ) || IS_INTRA( h->mb.type[mbn_xy] ) ) - { - bS[0] = bS[1] = bS[2] = bS[3] = ( i_edge == 0 && !(b_interlaced && i_dir) ? 4 : 3 ); - } - else - { - int i; - for( i = 0; i < 4; i++ ) - { - int x = i_dir == 0 ? i_edge : i; - int y = i_dir == 0 ? i : i_edge; - int xn = (x - (i_dir == 0 ? 1 : 0 ))&0x03; - int yn = (y - (i_dir == 0 ? 0 : 1 ))&0x03; - - if( h->mb.non_zero_count[mb_xy][block_idx_xy[x][y]] != 0 || - h->mb.non_zero_count[mbn_xy][block_idx_xy[xn][yn]] != 0 ) - { - bS[i] = 2; - } - else - { - /* FIXME: A given frame may occupy more than one position in - * the reference list. So we should compare the frame numbers, - * not the indices in the ref list. - * No harm yet, as we don't generate that case.*/ - - int i8p= mb_8x8+(x/2)+(y/2)*s8x8; - int i8q= mbn_8x8+(xn/2)+(yn/2)*s8x8; - int i4p= mb_4x4+x+y*s4x4; - int i4q= mbn_4x4+xn+yn*s4x4; - int l; - - bS[i] = 0; - - for( l = 0; l < 1 + (h->sh.i_type == SLICE_TYPE_B); l++ ) - { - if( h->mb.ref[l][i8p] != h->mb.ref[l][i8q] || - abs( h->mb.mv[l][i4p][0] - h->mb.mv[l][i4q][0] ) >= 4 || - abs( h->mb.mv[l][i4p][1] - h->mb.mv[l][i4q][1] ) >= mvy_limit ) - { - bS[i] = 1; - break; - } - } - } - } - } + if( i_qp <= qp_thresh ) + i_edge_end = 1; + + #define FILTER_DIR(intra, i_dir)\ + {\ + /* Y plane */\ + i_qpn= h->mb.qp[mbn_xy];\ + if( i_dir == 0 )\ + {\ + /* vertical edge */\ + deblock_edge##intra( h, pixy + 4*i_edge, NULL,\ + stride2y, bS, (i_qp+i_qpn+1) >> 1, 0,\ + h->loopf.deblock_h_luma##intra );\ + if( !(i_edge & 1) )\ + {\ + /* U/V planes */\ + int i_qpc = (h->chroma_qp_table[i_qp] + h->chroma_qp_table[i_qpn] + 1) >> 1;\ + deblock_edge##intra( h, pixu + 2*i_edge, pixv + 2*i_edge,\ + stride2uv, bS, i_qpc, 1,\ + h->loopf.deblock_h_chroma##intra );\ + }\ + }\ + else\ + {\ + /* horizontal edge */\ + deblock_edge##intra( h, pixy + 4*i_edge*stride2y, NULL,\ + stride2y, bS, (i_qp+i_qpn+1) >> 1, 0,\ + h->loopf.deblock_v_luma##intra );\ + /* U/V planes */\ + if( !(i_edge & 1) )\ + {\ + int i_qpc = (h->chroma_qp_table[i_qp] + h->chroma_qp_table[i_qpn] + 1) >> 1;\ + deblock_edge##intra( h, pixu + 2*i_edge*stride2uv, pixv + 2*i_edge*stride2uv,\ + stride2uv, bS, i_qpc, 1,\ + h->loopf.deblock_v_chroma##intra );\ + }\ + }\ + } - /* *** filter *** */ - /* Y plane */ - i_qp = h->mb.qp[mb_xy]; - i_qpn= h->mb.qp[mbn_xy]; + #define DEBLOCK_STRENGTH(i_dir)\ + {\ + /* *** Get bS for each 4px for the current edge *** */\ + if( IS_INTRA( h->mb.type[mb_xy] ) || IS_INTRA( h->mb.type[mbn_xy]) )\ + M32( bS ) = 0x03030303;\ + else\ + {\ + M32( bS ) = 0x00000000;\ + for( i = 0; i < 4; i++ )\ + {\ + int x = i_dir == 0 ? i_edge : i;\ + int y = i_dir == 0 ? i : i_edge;\ + int xn = i_dir == 0 ? (x - 1)&0x03 : x;\ + int yn = i_dir == 0 ? y : (y - 1)&0x03;\ + if( h->mb.non_zero_count[mb_xy][x+y*4] != 0 ||\ + h->mb.non_zero_count[mbn_xy][xn+yn*4] != 0 )\ + bS[i] = 2;\ + else if(!(i_edge&no_sub8x8))\ + {\ + if((i&no_sub8x8) && bS[i-1] != 2)\ + bS[i] = bS[i-1];\ + else\ + {\ + int i8p= mb_8x8+(x>>1)+(y>>1)*s8x8;\ + int i8q= mbn_8x8+(xn>>1)+(yn>>1)*s8x8;\ + int i4p= mb_4x4+x+y*s4x4;\ + int i4q= mbn_4x4+xn+yn*s4x4;\ + int refs_equal;\ + /* We don't use duplicate refs in B-frames, so we can take this shortcut for now. */ \ + if( h->sh.i_type == SLICE_TYPE_B || h->mb.ref[0][i8p] < 0 || h->mb.ref[0][i8q] < 0 )\ + refs_equal = h->mb.ref[0][i8p] == h->mb.ref[0][i8q];\ + else if( !h->mb.b_interlaced )\ + refs_equal = h->fref0[h->mb.ref[0][i8p]]->i_poc == h->fref0[h->mb.ref[0][i8q]]->i_poc;\ + else\ + refs_equal = h->fref0[h->mb.ref[0][i8p]>>1]->i_poc == h->fref0[h->mb.ref[0][i8q]>>1]->i_poc\ + && (h->mb.ref[0][i8p]&1) == (h->mb.ref[0][i8q]&1);\ + if((!refs_equal ||\ + abs( h->mb.mv[0][i4p][0] - h->mb.mv[0][i4q][0] ) >= 4 ||\ + abs( h->mb.mv[0][i4p][1] - h->mb.mv[0][i4q][1] ) >= mvy_limit ) ||\ + (h->sh.i_type == SLICE_TYPE_B &&\ + (h->mb.ref[1][i8p] != h->mb.ref[1][i8q] ||\ + abs( h->mb.mv[1][i4p][0] - h->mb.mv[1][i4q][0] ) >= 4 ||\ + abs( h->mb.mv[1][i4p][1] - h->mb.mv[1][i4q][1] ) >= mvy_limit )))\ + {\ + bS[i] = 1;\ + }\ + }\ + }\ + }\ + }\ + } - if( i_dir == 0 ) - { - /* vertical edge */ - deblock_edge( h, &h->fdec->plane[0][i_pix_y[0] + 4*i_edge], - i_stride2[0], bS, (i_qp+i_qpn+1) >> 1, 0, - h->loopf.deblock_h_luma, h->loopf.deblock_h_luma_intra ); - if( !(i_edge & 1) ) - { - /* U/V planes */ - int i_qpc = ( i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] + - i_chroma_qp_table[x264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1; - deblock_edge( h, &h->fdec->plane[1][i_pix_y[1] + 2*i_edge], - i_stride2[1], bS, i_qpc, 1, - h->loopf.deblock_h_chroma, h->loopf.deblock_h_chroma_intra ); - deblock_edge( h, &h->fdec->plane[2][i_pix_y[2] + 2*i_edge], - i_stride2[2], bS, i_qpc, 1, - h->loopf.deblock_h_chroma, h->loopf.deblock_h_chroma_intra ); - } - } - else - { - /* horizontal edge */ - deblock_edge( h, &h->fdec->plane[0][i_pix_y[0] + 4*i_edge*i_stride2[0]], - i_stride2[0], bS, (i_qp+i_qpn+1) >> 1, 0, - h->loopf.deblock_v_luma, h->loopf.deblock_v_luma_intra ); - /* U/V planes */ - if( !(i_edge & 1) ) - { - int i_qpc = ( i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] + - i_chroma_qp_table[x264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1; - deblock_edge( h, &h->fdec->plane[1][i_pix_y[1] + 2*i_edge*i_stride2[1]], - i_stride2[1], bS, i_qpc, 1, - h->loopf.deblock_v_chroma, h->loopf.deblock_v_chroma_intra ); - deblock_edge( h, &h->fdec->plane[2][i_pix_y[2] + 2*i_edge*i_stride2[2]], - i_stride2[2], bS, i_qpc, 1, - h->loopf.deblock_v_chroma, h->loopf.deblock_v_chroma_intra ); - } - } - } + /* i_dir == 0 -> vertical edge + * i_dir == 1 -> horizontal edge */ + #define DEBLOCK_DIR(i_dir)\ + {\ + int i_edge = (i_dir ? (mb_y <= b_interlaced) : (mb_x == 0));\ + int i_qpn, i, mbn_xy, mbn_8x8, mbn_4x4;\ + ALIGNED_4( uint8_t bS[4] ); /* filtering strength */\ + if( i_edge )\ + i_edge+= b_8x8_transform;\ + else\ + {\ + mbn_xy = i_dir == 0 ? mb_xy - 1 : mb_xy - h->mb.i_mb_stride;\ + mbn_8x8 = i_dir == 0 ? mb_8x8 - 2 : mb_8x8 - 2 * s8x8;\ + mbn_4x4 = i_dir == 0 ? mb_4x4 - 4 : mb_4x4 - 4 * s4x4;\ + if( b_interlaced && i_dir == 1 )\ + {\ + mbn_xy -= h->mb.i_mb_stride;\ + mbn_8x8 -= 2 * s8x8;\ + mbn_4x4 -= 4 * s4x4;\ + }\ + else if( IS_INTRA( h->mb.type[mb_xy] ) || IS_INTRA( h->mb.type[mbn_xy]) )\ + {\ + FILTER_DIR( _intra, i_dir );\ + goto end##i_dir;\ + }\ + DEBLOCK_STRENGTH(i_dir);\ + if( M32( bS ) )\ + FILTER_DIR( , i_dir);\ + end##i_dir:\ + i_edge += b_8x8_transform+1;\ + }\ + mbn_xy = mb_xy;\ + mbn_8x8 = mb_8x8;\ + mbn_4x4 = mb_4x4;\ + for( ; i_edge < i_edge_end; i_edge+=b_8x8_transform+1 )\ + {\ + DEBLOCK_STRENGTH(i_dir);\ + if( M32( bS ) )\ + FILTER_DIR( , i_dir);\ + }\ } - /* next mb */ - if( !b_interlaced || (mb_y&1) ) - mb_x++; - mb_y ^= b_interlaced; + DEBLOCK_DIR(0); + DEBLOCK_DIR(1); } if( !h->pps->b_cabac && h->pps->b_transform_8x8_mode ) @@ -748,19 +839,39 @@ void x264_deblock_h_chroma_mmxext( uint8_t *pix, int stride, int alpha, int beta void x264_deblock_v_chroma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta ); void x264_deblock_h_chroma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta ); -#ifdef ARCH_X86_64 void x264_deblock_v_luma_sse2( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ); void x264_deblock_h_luma_sse2( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ); -#else +void x264_deblock_v_luma_intra_sse2( uint8_t *pix, int stride, int alpha, int beta ); +void x264_deblock_h_luma_intra_sse2( uint8_t *pix, int stride, int alpha, int beta ); +#ifdef ARCH_X86 void x264_deblock_h_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ); void x264_deblock_v8_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ); +void x264_deblock_h_luma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta ); +void x264_deblock_v8_luma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta ); -void x264_deblock_v_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ) +static void x264_deblock_v_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ) { x264_deblock_v8_luma_mmxext( pix, stride, alpha, beta, tc0 ); x264_deblock_v8_luma_mmxext( pix+8, stride, alpha, beta, tc0+2 ); } +static void x264_deblock_v_luma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta ) +{ + x264_deblock_v8_luma_intra_mmxext( pix, stride, alpha, beta ); + x264_deblock_v8_luma_intra_mmxext( pix+8, stride, alpha, beta ); +} +#endif #endif + +#ifdef ARCH_PPC +void x264_deblock_v_luma_altivec( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ); +void x264_deblock_h_luma_altivec( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ); +#endif // ARCH_PPC + +#ifdef HAVE_ARMV6 +void x264_deblock_v_luma_neon( uint8_t *, int, int, int, int8_t * ); +void x264_deblock_h_luma_neon( uint8_t *, int, int, int, int8_t * ); +void x264_deblock_v_chroma_neon( uint8_t *, int, int, int, int8_t * ); +void x264_deblock_h_chroma_neon( uint8_t *, int, int, int, int8_t * ); #endif void x264_deblock_init( int cpu, x264_deblock_function_t *pf ) @@ -781,49 +892,59 @@ void x264_deblock_init( int cpu, x264_deblock_function_t *pf ) pf->deblock_h_chroma = x264_deblock_h_chroma_mmxext; pf->deblock_v_chroma_intra = x264_deblock_v_chroma_intra_mmxext; pf->deblock_h_chroma_intra = x264_deblock_h_chroma_intra_mmxext; - -#ifdef ARCH_X86_64 - if( cpu&X264_CPU_SSE2 ) +#ifdef ARCH_X86 + pf->deblock_v_luma = x264_deblock_v_luma_mmxext; + pf->deblock_h_luma = x264_deblock_h_luma_mmxext; + pf->deblock_v_luma_intra = x264_deblock_v_luma_intra_mmxext; + pf->deblock_h_luma_intra = x264_deblock_h_luma_intra_mmxext; +#endif + if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_STACK_MOD4) ) { pf->deblock_v_luma = x264_deblock_v_luma_sse2; pf->deblock_h_luma = x264_deblock_h_luma_sse2; + pf->deblock_v_luma_intra = x264_deblock_v_luma_intra_sse2; + pf->deblock_h_luma_intra = x264_deblock_h_luma_intra_sse2; } -#else - pf->deblock_v_luma = x264_deblock_v_luma_mmxext; - pf->deblock_h_luma = x264_deblock_h_luma_mmxext; -#endif } #endif + +#ifdef ARCH_PPC + if( cpu&X264_CPU_ALTIVEC ) + { + pf->deblock_v_luma = x264_deblock_v_luma_altivec; + pf->deblock_h_luma = x264_deblock_h_luma_altivec; + } +#endif // ARCH_PPC + +#ifdef HAVE_ARMV6 + if( cpu&X264_CPU_NEON ) + { + pf->deblock_v_luma = x264_deblock_v_luma_neon; + pf->deblock_h_luma = x264_deblock_h_luma_neon; + pf->deblock_v_chroma = x264_deblock_v_chroma_neon; + pf->deblock_h_chroma = x264_deblock_h_chroma_neon; + } +#endif } /* threading */ - -#ifdef HAVE_PTHREAD void x264_frame_cond_broadcast( x264_frame_t *frame, int i_lines_completed ) { - pthread_mutex_lock( &frame->mutex ); + x264_pthread_mutex_lock( &frame->mutex ); frame->i_lines_completed = i_lines_completed; - pthread_cond_broadcast( &frame->cv ); - pthread_mutex_unlock( &frame->mutex ); + x264_pthread_cond_broadcast( &frame->cv ); + x264_pthread_mutex_unlock( &frame->mutex ); } void x264_frame_cond_wait( x264_frame_t *frame, int i_lines_completed ) { - pthread_mutex_lock( &frame->mutex ); + x264_pthread_mutex_lock( &frame->mutex ); while( frame->i_lines_completed < i_lines_completed ) - pthread_cond_wait( &frame->cv, &frame->mutex ); - pthread_mutex_unlock( &frame->mutex ); + x264_pthread_cond_wait( &frame->cv, &frame->mutex ); + x264_pthread_mutex_unlock( &frame->mutex ); } -#else -void x264_frame_cond_broadcast( x264_frame_t *frame, int i_lines_completed ) -{} -void x264_frame_cond_wait( x264_frame_t *frame, int i_lines_completed ) -{} -#endif - - /* list operators */ void x264_frame_push( x264_frame_t **list, x264_frame_t *frame ) @@ -868,18 +989,48 @@ void x264_frame_push_unused( x264_t *h, x264_frame_t *frame ) assert( frame->i_reference_count > 0 ); frame->i_reference_count--; if( frame->i_reference_count == 0 ) - x264_frame_push( h->frames.unused, frame ); - assert( h->frames.unused[ sizeof(h->frames.unused) / sizeof(*h->frames.unused) - 1 ] == NULL ); + x264_frame_push( h->frames.unused[frame->b_fdec], frame ); } -x264_frame_t *x264_frame_pop_unused( x264_t *h ) +x264_frame_t *x264_frame_pop_unused( x264_t *h, int b_fdec ) { x264_frame_t *frame; - if( h->frames.unused[0] ) - frame = x264_frame_pop( h->frames.unused ); + if( h->frames.unused[b_fdec][0] ) + frame = x264_frame_pop( h->frames.unused[b_fdec] ); else - frame = x264_frame_new( h ); - assert( frame->i_reference_count == 0 ); + frame = x264_frame_new( h, b_fdec ); + if( !frame ) + return NULL; + frame->b_last_minigop_bframe = 0; + frame->i_reference_count = 1; + frame->b_intra_calculated = 0; + frame->b_scenecut = 1; + frame->b_keyframe = 0; + + memset( frame->weight, 0, sizeof(frame->weight) ); + memset( frame->f_weighted_cost_delta, 0, sizeof(frame->f_weighted_cost_delta) ); + + return frame; +} + +void x264_frame_push_blank_unused( x264_t *h, x264_frame_t *frame ) +{ + assert( frame->i_reference_count > 0 ); + frame->i_reference_count--; + if( frame->i_reference_count == 0 ) + x264_frame_push( h->frames.blank_unused, frame ); +} + +x264_frame_t *x264_frame_pop_blank_unused( x264_t *h ) +{ + x264_frame_t *frame; + if( h->frames.blank_unused[0] ) + frame = x264_frame_pop( h->frames.blank_unused ); + else + frame = x264_malloc( sizeof(x264_frame_t) ); + if( !frame ) + return NULL; + frame->b_duplicate = 1; frame->i_reference_count = 1; return frame; } @@ -903,3 +1054,63 @@ void x264_frame_sort( x264_frame_t **list, int b_dts ) } } while( !b_ok ); } + +void x264_weight_scale_plane( x264_t *h, uint8_t *dst, int i_dst_stride, uint8_t *src, int i_src_stride, + int i_width, int i_height, x264_weight_t *w ) +{ + int x; + /* Weight horizontal strips of height 16. This was found to be the optimal height + * in terms of the cache loads. */ + while( i_height > 0 ) + { + for( x = 0; x < i_width; x += 16 ) + w->weightfn[16>>2]( dst+x, i_dst_stride, src+x, i_src_stride, w, X264_MIN( i_height, 16 ) ); + i_height -= 16; + dst += 16 * i_dst_stride; + src += 16 * i_src_stride; + } +} + +void x264_frame_delete_list( x264_frame_t **list ) +{ + int i = 0; + if( !list ) + return; + while( list[i] ) + x264_frame_delete( list[i++] ); + x264_free( list ); +} + +int x264_synch_frame_list_init( x264_synch_frame_list_t *slist, int max_size ) +{ + if( max_size < 0 ) + return -1; + slist->i_max_size = max_size; + slist->i_size = 0; + CHECKED_MALLOCZERO( slist->list, (max_size+1) * sizeof(x264_frame_t*) ); + if( x264_pthread_mutex_init( &slist->mutex, NULL ) || + x264_pthread_cond_init( &slist->cv_fill, NULL ) || + x264_pthread_cond_init( &slist->cv_empty, NULL ) ) + return -1; + return 0; +fail: + return -1; +} + +void x264_synch_frame_list_delete( x264_synch_frame_list_t *slist ) +{ + x264_pthread_mutex_destroy( &slist->mutex ); + x264_pthread_cond_destroy( &slist->cv_fill ); + x264_pthread_cond_destroy( &slist->cv_empty ); + x264_frame_delete_list( slist->list ); +} + +void x264_synch_frame_list_push( x264_synch_frame_list_t *slist, x264_frame_t *frame ) +{ + x264_pthread_mutex_lock( &slist->mutex ); + while( slist->i_size == slist->i_max_size ) + x264_pthread_cond_wait( &slist->cv_empty, &slist->mutex ); + slist->list[ slist->i_size++ ] = frame; + x264_pthread_mutex_unlock( &slist->mutex ); + x264_pthread_cond_broadcast( &slist->cv_fill ); +}