X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=common%2Fframe.c;h=ffecd2cadb3db7528850279299b06332ddc39510;hb=9f422c0cd9c0abcd6a7abb10b51f8be883c39b2b;hp=e81f6239334063dc4e2910a06e8247f1a52b14f2;hpb=6eab44d4295801c28184ec13f03f9727c60129cc;p=x264 diff --git a/common/frame.c b/common/frame.c index e81f6239..ffecd2ca 100644 --- a/common/frame.c +++ b/common/frame.c @@ -1,7 +1,7 @@ /***************************************************************************** - * frame.c: h264 encoder library + * frame.c: frame handling ***************************************************************************** - * Copyright (C) 2003-2008 x264 project + * Copyright (C) 2003-2015 x264 project * * Authors: Laurent Aimar * Loren Merritt @@ -20,130 +20,279 @@ * 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., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. + * + * This program is also available under a commercial proprietary license. + * For more information, contact us at licensing@x264.com. *****************************************************************************/ #include "common.h" -#define ALIGN(x,a) (((x)+((a)-1))&~((a)-1)) +static int align_stride( int x, int align, int disalign ) +{ + x = ALIGN( x, align ); + if( !(x&(disalign-1)) ) + x += align; + return x; +} -x264_frame_t *x264_frame_new( x264_t *h ) +static int align_plane_size( int x, int disalign ) { - x264_frame_t *frame; - int i, j; + if( !(x&(disalign-1)) ) + x += 128; + return x; +} +static int x264_frame_internal_csp( int external_csp ) +{ + switch( external_csp & X264_CSP_MASK ) + { + case X264_CSP_NV12: + case X264_CSP_NV21: + case X264_CSP_I420: + case X264_CSP_YV12: + return X264_CSP_NV12; + case X264_CSP_NV16: + case X264_CSP_I422: + case X264_CSP_YV16: + case X264_CSP_V210: + return X264_CSP_NV16; + case X264_CSP_I444: + case X264_CSP_YV24: + case X264_CSP_BGR: + case X264_CSP_BGRA: + case X264_CSP_RGB: + return X264_CSP_I444; + default: + return X264_CSP_NONE; + } +} + +static x264_frame_t *x264_frame_new( x264_t *h, int b_fdec ) +{ + x264_frame_t *frame; + int i_csp = x264_frame_internal_csp( h->param.i_csp ); int i_mb_count = h->mb.i_mb_count; - 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; + int i_stride, i_width, i_lines, luma_plane_count; + int i_padv = PADV << PARAM_INTERLACED; + int align = 16; +#if ARCH_X86 || ARCH_X86_64 + if( h->param.cpu&X264_CPU_CACHELINE_64 ) + align = 64; + else if( h->param.cpu&X264_CPU_CACHELINE_32 || h->param.cpu&X264_CPU_AVX ) + align = 32; +#endif +#if ARCH_PPC + int disalign = 1<<9; +#else + int disalign = 1<<10; +#endif CHECKED_MALLOCZERO( frame, sizeof(x264_frame_t) ); + PREALLOC_INIT /* allocate frame data (+64 for extra data for me) */ - 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 ); + i_width = h->mb.i_mb_width*16; + i_lines = h->mb.i_mb_height*16; + i_stride = align_stride( i_width + 2*PADH, align, disalign ); - frame->i_plane = 3; - for( i = 0; i < 3; i++ ) + if( i_csp == X264_CSP_NV12 || i_csp == X264_CSP_NV16 ) { - frame->i_stride[i] = ALIGN( i_stride >> !!i, align ); - frame->i_width[i] = i_width >> !!i; - frame->i_lines[i] = i_lines >> !!i; + luma_plane_count = 1; + frame->i_plane = 2; + for( int i = 0; i < 2; i++ ) + { + frame->i_width[i] = i_width >> i; + frame->i_lines[i] = i_lines >> (i && i_csp == X264_CSP_NV12); + frame->i_stride[i] = i_stride; + } } + else if( i_csp == X264_CSP_I444 ) + { + luma_plane_count = 3; + frame->i_plane = 3; + for( int i = 0; i < 3; i++ ) + { + frame->i_width[i] = i_width; + frame->i_lines[i] = i_lines; + frame->i_stride[i] = i_stride; + } + } + else + goto fail; - 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_csp = i_csp; + frame->i_width_lowres = frame->i_width[0]/2; + frame->i_lines_lowres = frame->i_lines[0]/2; + frame->i_stride_lowres = align_stride( frame->i_width_lowres + 2*PADH, align, disalign<<1 ); + + for( int i = 0; i < h->param.i_bframe + 2; i++ ) + for( int j = 0; j < h->param.i_bframe + 2; j++ ) + PREALLOC( frame->i_row_satds[i][j], i_lines/16 * sizeof(int) ); + + frame->i_poc = -1; + frame->i_type = X264_TYPE_AUTO; + frame->i_qpplus1 = X264_QP_AUTO; + frame->i_pts = -1; + frame->i_frame = -1; + frame->i_frame_num = -1; + frame->i_lines_completed = -1; + frame->b_fdec = b_fdec; + frame->i_pic_struct = PIC_STRUCT_AUTO; + frame->i_field_cnt = -1; + frame->i_duration = + frame->i_cpb_duration = + frame->i_dpb_output_delay = + frame->i_cpb_delay = 0; + frame->i_coded_fields_lookahead = + frame->i_cpb_delay_lookahead = -1; + + frame->orig = frame; + + if( i_csp == X264_CSP_NV12 || i_csp == X264_CSP_NV16 ) { - CHECKED_MALLOC( frame->buffer[i], chroma_plane_size ); - frame->plane[i] = frame->buffer[i] + (frame->i_stride[i] * i_padv + PADH)/2; + int chroma_padv = i_padv >> (i_csp == X264_CSP_NV12); + int chroma_plane_size = (frame->i_stride[1] * (frame->i_lines[1] + 2*chroma_padv)); + PREALLOC( frame->buffer[1], chroma_plane_size * sizeof(pixel) ); + if( PARAM_INTERLACED ) + PREALLOC( frame->buffer_fld[1], chroma_plane_size * sizeof(pixel) ); } + /* 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 ) + + for( int p = 0; p < luma_plane_count; p++ ) { - 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]; + int luma_plane_size = align_plane_size( frame->i_stride[p] * (frame->i_lines[p] + 2*i_padv), disalign ); + if( h->param.analyse.i_subpel_refine && b_fdec ) + { + /* FIXME: Don't allocate both buffers in non-adaptive MBAFF. */ + PREALLOC( frame->buffer[p], 4*luma_plane_size * sizeof(pixel) ); + if( PARAM_INTERLACED ) + PREALLOC( frame->buffer_fld[p], 4*luma_plane_size * sizeof(pixel) ); + } + else + { + PREALLOC( frame->buffer[p], luma_plane_size * sizeof(pixel) ); + if( PARAM_INTERLACED ) + PREALLOC( frame->buffer_fld[p], luma_plane_size * sizeof(pixel) ); + } } - else + + frame->b_duplicate = 0; + + if( b_fdec ) /* fdec frame */ { - CHECKED_MALLOC( frame->buffer[0], luma_plane_size); - frame->plane[0] = frame->buffer[0] + frame->i_stride[0] * i_padv + PADH; + PREALLOC( frame->mb_type, i_mb_count * sizeof(int8_t) ); + PREALLOC( frame->mb_partition, i_mb_count * sizeof(uint8_t) ); + PREALLOC( frame->mv[0], 2*16 * i_mb_count * sizeof(int16_t) ); + PREALLOC( frame->mv16x16, 2*(i_mb_count+1) * sizeof(int16_t) ); + PREALLOC( frame->ref[0], 4 * i_mb_count * sizeof(int8_t) ); + if( h->param.i_bframe ) + { + PREALLOC( frame->mv[1], 2*16 * i_mb_count * sizeof(int16_t) ); + PREALLOC( frame->ref[1], 4 * i_mb_count * sizeof(int8_t) ); + } + else + { + frame->mv[1] = NULL; + frame->ref[1] = NULL; + } + PREALLOC( frame->i_row_bits, i_lines/16 * sizeof(int) ); + PREALLOC( frame->f_row_qp, i_lines/16 * sizeof(float) ); + PREALLOC( frame->f_row_qscale, i_lines/16 * sizeof(float) ); + if( h->param.analyse.i_me_method >= X264_ME_ESA ) + PREALLOC( frame->buffer[3], frame->i_stride[0] * (frame->i_lines[0] + 2*i_padv) * sizeof(uint16_t) << h->frames.b_have_sub8x8_esa ); + if( PARAM_INTERLACED ) + PREALLOC( frame->field, i_mb_count * sizeof(uint8_t) ); + if( h->param.analyse.b_mb_info ) + PREALLOC( frame->effective_qp, i_mb_count * sizeof(uint8_t) ); } - - if( h->frames.b_have_lowres ) + else /* fenc frame */ { - 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; + if( h->frames.b_have_lowres ) + { + int luma_plane_size = align_plane_size( frame->i_stride_lowres * (frame->i_lines[0]/2 + 2*PADV), disalign ); - luma_plane_size = frame->i_stride_lowres * ( frame->i_lines[0]/2 + 2*i_padv ); + PREALLOC( frame->buffer_lowres[0], 4 * luma_plane_size * sizeof(pixel) ); - 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( int j = 0; j <= !!h->param.i_bframe; j++ ) + for( int i = 0; i <= h->param.i_bframe; i++ ) + { + PREALLOC( frame->lowres_mvs[j][i], 2*h->mb.i_mb_count*sizeof(int16_t) ); + PREALLOC( frame->lowres_mv_costs[j][i], h->mb.i_mb_count*sizeof(int) ); + } + PREALLOC( frame->i_propagate_cost, (i_mb_count+7) * sizeof(uint16_t) ); + for( int j = 0; j <= h->param.i_bframe+1; j++ ) + for( int i = 0; i <= h->param.i_bframe+1; i++ ) + PREALLOC( frame->lowres_costs[j][i], (i_mb_count+3) * sizeof(uint16_t) ); - 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_intra_cost, i_mb_count * sizeof(uint16_t) ); - memset( frame->i_intra_cost, -1, i_mb_count * sizeof(uint16_t) ); - CHECKED_MALLOC( frame->i_propagate_cost, i_mb_count * sizeof(uint32_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 * sizeof(uint16_t) ); - CHECKED_MALLOC( frame->lowres_inter_types[j][i], i_mb_count * sizeof(uint8_t) ); - } + } + if( h->param.rc.i_aq_mode ) + { + PREALLOC( frame->f_qp_offset, h->mb.i_mb_count * sizeof(float) ); + PREALLOC( frame->f_qp_offset_aq, h->mb.i_mb_count * sizeof(float) ); + if( h->frames.b_have_lowres ) + PREALLOC( frame->i_inv_qscale_factor, (h->mb.i_mb_count+3) * sizeof(uint16_t) ); + } } - if( h->param.analyse.i_me_method >= X264_ME_ESA ) + PREALLOC_END( frame->base ); + + if( i_csp == X264_CSP_NV12 || i_csp == X264_CSP_NV16 ) { - 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; + int chroma_padv = i_padv >> (i_csp == X264_CSP_NV12); + frame->plane[1] = frame->buffer[1] + frame->i_stride[1] * chroma_padv + PADH; + if( PARAM_INTERLACED ) + frame->plane_fld[1] = frame->buffer_fld[1] + frame->i_stride[1] * chroma_padv + PADH; } - frame->i_poc = -1; - frame->i_type = X264_TYPE_AUTO; - frame->i_qpplus1 = 0; - frame->i_pts = -1; - frame->i_frame = -1; - frame->i_frame_num = -1; - frame->i_lines_completed = -1; + for( int p = 0; p < luma_plane_count; p++ ) + { + int luma_plane_size = align_plane_size( frame->i_stride[p] * (frame->i_lines[p] + 2*i_padv), disalign ); + if( h->param.analyse.i_subpel_refine && b_fdec ) + { + for( int i = 0; i < 4; i++ ) + { + frame->filtered[p][i] = frame->buffer[p] + i*luma_plane_size + frame->i_stride[p] * i_padv + PADH; + frame->filtered_fld[p][i] = frame->buffer_fld[p] + i*luma_plane_size + frame->i_stride[p] * i_padv + PADH; + } + frame->plane[p] = frame->filtered[p][0]; + frame->plane_fld[p] = frame->filtered_fld[p][0]; + } + else + { + frame->filtered[p][0] = frame->plane[p] = frame->buffer[p] + frame->i_stride[p] * i_padv + PADH; + frame->filtered_fld[p][0] = frame->plane_fld[p] = frame->buffer_fld[p] + frame->i_stride[p] * i_padv + PADH; + } + } - 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 ) + if( 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) ); + M32( frame->mv16x16[0] ) = 0; + frame->mv16x16++; + + if( h->param.analyse.i_me_method >= X264_ME_ESA ) + frame->integral = (uint16_t*)frame->buffer[3] + frame->i_stride[0] * i_padv + PADH; } else { - frame->mv[1] = NULL; - frame->ref[1] = NULL; - } + if( h->frames.b_have_lowres ) + { + int luma_plane_size = align_plane_size( frame->i_stride_lowres * (frame->i_lines[0]/2 + 2*PADV), disalign ); + for( int i = 0; i < 4; i++ ) + frame->lowres[i] = frame->buffer_lowres[0] + (frame->i_stride_lowres * PADV + PADH) + i * luma_plane_size; - 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) ); + for( int j = 0; j <= !!h->param.i_bframe; j++ ) + for( int i = 0; i <= h->param.i_bframe; i++ ) + memset( frame->lowres_mvs[j][i], 0, 2*h->mb.i_mb_count*sizeof(int16_t) ); - if( h->param.rc.i_aq_mode ) - { - CHECKED_MALLOC( frame->f_qp_offset, h->mb.i_mb_count * sizeof(float) ); - if( h->frames.b_have_lowres ) - CHECKED_MALLOC( frame->i_inv_qscale_factor, h->mb.i_mb_count * sizeof(uint16_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 ) + /* shouldn't really be initialized, just silences a valgrind false-positive in x264_mbtree_propagate_cost_sse2 */ + memset( frame->i_inv_qscale_factor, 0, (h->mb.i_mb_count+3) * sizeof(uint16_t) ); + } } if( x264_pthread_mutex_init( &frame->mutex, NULL ) ) @@ -151,6 +300,10 @@ x264_frame_t *x264_frame_new( x264_t *h ) if( x264_pthread_cond_init( &frame->cv, NULL ) ) goto fail; +#if HAVE_OPENCL + frame->opencl.ocl = h->opencl.ocl; +#endif + return frame; fail: @@ -160,729 +313,365 @@ fail: void x264_frame_delete( x264_frame_t *frame ) { - int i, j; - 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++ ) + /* Duplicate frames are blank copies of real frames (including pointers), + * so freeing those pointers would cause a double free later. */ + if( !frame->b_duplicate ) + { + x264_free( frame->base ); + + if( frame->param && frame->param->param_free ) + frame->param->param_free( frame->param ); + if( frame->mb_info_free ) + frame->mb_info_free( frame->mb_info ); + if( frame->extra_sei.sei_free ) { - x264_free( frame->lowres_mvs[j][i] ); - x264_free( frame->lowres_mv_costs[j][i] ); + for( int i = 0; i < frame->extra_sei.num_payloads; i++ ) + frame->extra_sei.sei_free( frame->extra_sei.payloads[i].payload ); + frame->extra_sei.sei_free( frame->extra_sei.payloads ); } - x264_free( frame->f_qp_offset ); - x264_free( frame->i_inv_qscale_factor ); - x264_free( frame->i_intra_cost ); - 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_pthread_mutex_destroy( &frame->mutex ); + x264_pthread_cond_destroy( &frame->cv ); +#if HAVE_OPENCL + x264_opencl_frame_delete( frame ); +#endif + } x264_free( frame ); } +static int get_plane_ptr( x264_t *h, x264_picture_t *src, uint8_t **pix, int *stride, int plane, int xshift, int yshift ) +{ + int width = h->param.i_width >> xshift; + int height = h->param.i_height >> yshift; + *pix = src->img.plane[plane]; + *stride = src->img.i_stride[plane]; + if( src->img.i_csp & X264_CSP_VFLIP ) + { + *pix += (height-1) * *stride; + *stride = -*stride; + } + if( width > abs(*stride) ) + { + x264_log( h, X264_LOG_ERROR, "Input picture width (%d) is greater than stride (%d)\n", width, *stride ); + return -1; + } + return 0; +} + +#define get_plane_ptr(...) do{ if( get_plane_ptr(__VA_ARGS__) < 0 ) return -1; }while(0) + 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 ) + if( dst->i_csp != x264_frame_internal_csp( i_csp ) ) { - x264_log( h, X264_LOG_ERROR, "Arg invalid CSP\n" ); + x264_log( h, X264_LOG_ERROR, "Invalid input colorspace\n" ); return -1; } - dst->i_type = src->i_type; - dst->i_qpplus1 = src->i_qpplus1; - dst->i_pts = src->i_pts; +#if HIGH_BIT_DEPTH + if( !(src->img.i_csp & X264_CSP_HIGH_DEPTH) ) + { + x264_log( h, X264_LOG_ERROR, "This build of x264 requires high depth input. Rebuild to support 8-bit input.\n" ); + return -1; + } +#else + if( src->img.i_csp & X264_CSP_HIGH_DEPTH ) + { + x264_log( h, X264_LOG_ERROR, "This build of x264 requires 8-bit input. Rebuild to support high depth input.\n" ); + return -1; + } +#endif - for( i=0; i<3; i++ ) + if( BIT_DEPTH != 10 && i_csp == X264_CSP_V210 ) { - 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 ); + x264_log( h, X264_LOG_ERROR, "v210 input is only compatible with bit-depth of 10 bits\n" ); + return -1; } - return 0; -} + if( src->i_type < X264_TYPE_AUTO || src->i_type > X264_TYPE_KEYFRAME ) + { + x264_log( h, X264_LOG_WARNING, "forced frame type (%d) at %d is unknown\n", src->i_type, h->frames.i_input ); + dst->i_forced_type = X264_TYPE_AUTO; + } + else + dst->i_forced_type = src->i_type; + dst->i_type = dst->i_forced_type; + dst->i_qpplus1 = src->i_qpplus1; + dst->i_pts = dst->i_reordered_pts = src->i_pts; + dst->param = src->param; + dst->i_pic_struct = src->i_pic_struct; + dst->extra_sei = src->extra_sei; + dst->opaque = src->opaque; + dst->mb_info = h->param.analyse.b_mb_info ? src->prop.mb_info : NULL; + dst->mb_info_free = h->param.analyse.b_mb_info ? src->prop.mb_info_free : NULL; + + uint8_t *pix[3]; + int stride[3]; + if( i_csp == X264_CSP_V210 ) + { + stride[0] = src->img.i_stride[0]; + pix[0] = src->img.plane[0]; -static void plane_expand_border( uint8_t *pix, int i_stride, int i_width, int i_height, int i_padh, int i_padv, int b_pad_top, int b_pad_bottom ) -{ -#define PPIXEL(x, y) ( pix + (x) + (y)*i_stride ) - int y; - for( y = 0; y < i_height; y++ ) + h->mc.plane_copy_deinterleave_v210( dst->plane[0], dst->i_stride[0], + dst->plane[1], dst->i_stride[1], + (uint32_t *)pix[0], stride[0]/sizeof(uint32_t), h->param.i_width, h->param.i_height ); + } + else if( i_csp >= X264_CSP_BGR ) { - /* left band */ - memset( PPIXEL(-i_padh, y), PPIXEL(0, y)[0], i_padh ); - /* right band */ - memset( PPIXEL(i_width, y), PPIXEL(i_width-1, y)[0], i_padh ); + stride[0] = src->img.i_stride[0]; + pix[0] = src->img.plane[0]; + if( src->img.i_csp & X264_CSP_VFLIP ) + { + pix[0] += (h->param.i_height-1) * stride[0]; + stride[0] = -stride[0]; + } + int b = i_csp==X264_CSP_RGB; + h->mc.plane_copy_deinterleave_rgb( dst->plane[1+b], dst->i_stride[1+b], + dst->plane[0], dst->i_stride[0], + dst->plane[2-b], dst->i_stride[2-b], + (pixel*)pix[0], stride[0]/sizeof(pixel), i_csp==X264_CSP_BGRA ? 4 : 3, h->param.i_width, h->param.i_height ); } - /* upper band */ - if( b_pad_top ) - for( y = 0; y < i_padv; y++ ) - memcpy( PPIXEL(-i_padh, -y-1), PPIXEL(-i_padh, 0), i_width+2*i_padh ); - /* lower band */ - if( b_pad_bottom ) - for( y = 0; y < i_padv; y++ ) - memcpy( PPIXEL(-i_padh, i_height+y), PPIXEL(-i_padh, i_height-1), i_width+2*i_padh ); -#undef PPIXEL -} - -void x264_frame_expand_border( x264_t *h, x264_frame_t *frame, int mb_y, int b_end ) -{ - int i; - int b_start = !mb_y; - if( mb_y & h->sh.b_mbaff ) - return; - for( i = 0; i < frame->i_plane; i++ ) + else { - int stride = frame->i_stride[i]; - int width = 16*h->sps->i_mb_width >> !!i; - int height = (b_end ? 16*(h->sps->i_mb_height - mb_y) >> h->sh.b_mbaff : 16) >> !!i; - int padh = PADH >> !!i; - int padv = PADV >> !!i; - // buffer: 2 chroma, 3 luma (rounded to 4) because deblocking goes beyond the top of the mb - uint8_t *pix = frame->plane[i] + X264_MAX(0, (16*mb_y-4)*stride >> !!i); - if( b_end && !b_start ) - height += 4 >> (!!i + h->sh.b_mbaff); - if( h->sh.b_mbaff ) + int v_shift = CHROMA_V_SHIFT; + get_plane_ptr( h, src, &pix[0], &stride[0], 0, 0, 0 ); + h->mc.plane_copy( dst->plane[0], dst->i_stride[0], (pixel*)pix[0], + stride[0]/sizeof(pixel), h->param.i_width, h->param.i_height ); + if( i_csp == X264_CSP_NV12 || i_csp == X264_CSP_NV16 ) { - plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end ); - plane_expand_border( pix+stride, stride*2, width, height, padh, padv, b_start, b_end ); + get_plane_ptr( h, src, &pix[1], &stride[1], 1, 0, v_shift ); + h->mc.plane_copy( dst->plane[1], dst->i_stride[1], (pixel*)pix[1], + stride[1]/sizeof(pixel), h->param.i_width, h->param.i_height>>v_shift ); } - else + else if( i_csp == X264_CSP_NV21 ) { - plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end ); + get_plane_ptr( h, src, &pix[1], &stride[1], 1, 0, v_shift ); + h->mc.plane_copy_swap( dst->plane[1], dst->i_stride[1], (pixel*)pix[1], + stride[1]/sizeof(pixel), h->param.i_width>>1, h->param.i_height>>v_shift ); } - } -} - -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, - * 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 + 8; - int height = b_end ? (16*(h->sps->i_mb_height - mb_y) >> h->sh.b_mbaff) + 16 : 16; - 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 - 4; - if( h->sh.b_mbaff ) + else if( i_csp == X264_CSP_I420 || i_csp == X264_CSP_I422 || i_csp == X264_CSP_YV12 || i_csp == X264_CSP_YV16 ) { - plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end ); - plane_expand_border( pix+stride, stride*2, width, height, padh, padv, b_start, b_end ); + int uv_swap = i_csp == X264_CSP_YV12 || i_csp == X264_CSP_YV16; + get_plane_ptr( h, src, &pix[1], &stride[1], uv_swap ? 2 : 1, 1, v_shift ); + get_plane_ptr( h, src, &pix[2], &stride[2], uv_swap ? 1 : 2, 1, v_shift ); + h->mc.plane_copy_interleave( dst->plane[1], dst->i_stride[1], + (pixel*)pix[1], stride[1]/sizeof(pixel), + (pixel*)pix[2], stride[2]/sizeof(pixel), + h->param.i_width>>1, h->param.i_height>>v_shift ); } - else + else //if( i_csp == X264_CSP_I444 || i_csp == X264_CSP_YV24 ) { - plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end ); + get_plane_ptr( h, src, &pix[1], &stride[1], i_csp==X264_CSP_I444 ? 1 : 2, 0, 0 ); + get_plane_ptr( h, src, &pix[2], &stride[2], i_csp==X264_CSP_I444 ? 2 : 1, 0, 0 ); + h->mc.plane_copy( dst->plane[1], dst->i_stride[1], (pixel*)pix[1], + stride[1]/sizeof(pixel), h->param.i_width, h->param.i_height ); + h->mc.plane_copy( dst->plane[2], dst->i_stride[2], (pixel*)pix[2], + stride[2]/sizeof(pixel), h->param.i_width, h->param.i_height ); } } + return 0; } -void x264_frame_expand_border_lowres( x264_frame_t *frame ) +static void ALWAYS_INLINE pixel_memset( pixel *dst, pixel *src, int len, int size ) { - 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 ); -} + uint8_t *dstp = (uint8_t*)dst; + uint32_t v1 = *src; + uint32_t v2 = size == 1 ? v1 + (v1 << 8) : M16( src ); + uint32_t v4 = size <= 2 ? v2 + (v2 << 16) : M32( src ); + int i = 0; + len *= size; -void x264_frame_expand_border_mod16( x264_t *h, x264_frame_t *frame ) -{ - int i, y; - for( i = 0; i < frame->i_plane; i++ ) + /* Align the input pointer if it isn't already */ + if( (intptr_t)dstp & (WORD_SIZE - 1) ) { - 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; - - if( i_padx ) + if( size <= 2 && ((intptr_t)dstp & 3) ) { - for( y = 0; y < i_height; y++ ) - memset( &frame->plane[i][y*frame->i_stride[i] + i_width], - frame->plane[i][y*frame->i_stride[i] + i_width - 1], - i_padx ); + if( size == 1 && ((intptr_t)dstp & 1) ) + dstp[i++] = v1; + if( (intptr_t)dstp & 2 ) + { + M16( dstp+i ) = v2; + i += 2; + } } - if( i_pady ) + if( WORD_SIZE == 8 && (intptr_t)dstp & 4 ) { - //FIXME interlace? or just let it pad using the wrong field - for( y = i_height; y < i_height + i_pady; y++ ) - memcpy( &frame->plane[i][y*frame->i_stride[i]], - &frame->plane[i][(i_height-1)*frame->i_stride[i]], - i_width + i_padx ); + M32( dstp+i ) = v4; + i += 4; } } -} + /* Main copy loop */ + if( WORD_SIZE == 8 ) + { + uint64_t v8 = v4 + ((uint64_t)v4<<32); + for( ; i < len - 7; i+=8 ) + M64( dstp+i ) = v8; + } + for( ; i < len - 3; i+=4 ) + M32( dstp+i ) = v4; -/* cavlc + 8x8 transform stores nnz per 16 coeffs for the purpose of - * entropy coding, but per 64 coeffs for the purpose of deblocking */ -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, nnz; - for( x=0; xsps->i_mb_width; x++ ) + /* Finish up the last few bytes */ + if( size <= 2 ) { - memcpy( buf+x, src+x, 16 ); - if( transform[x] ) + if( i < len - 1 ) { - 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); + M16( dstp+i ) = v2; + i += 2; } + if( size == 1 && i != len ) + dstp[i] = v1; } } -static void restore_cavlc_nnz_row( x264_t *h, int mb_y, uint8_t (*buf)[16] ) +static void ALWAYS_INLINE plane_expand_border( pixel *pix, int i_stride, int i_width, int i_height, int i_padh, int i_padv, int b_pad_top, int b_pad_bottom, int b_chroma ) { - uint8_t (*dst)[24] = h->mb.non_zero_count + mb_y * h->sps->i_mb_width; - int x; - for( x=0; xsps->i_mb_width; x++ ) - memcpy( dst+x, buf+x, 16 ); -} - -static void munge_cavlc_nnz( x264_t *h, int mb_y, uint8_t (*buf)[16], void (*func)(x264_t*, int, uint8_t (*)[16]) ) -{ - func( h, mb_y, buf ); - if( mb_y > 0 ) - func( h, mb_y-1, buf + h->sps->i_mb_width ); - if( h->sh.b_mbaff ) - { - func( h, mb_y+1, buf + h->sps->i_mb_width * 2 ); - if( mb_y > 0 ) - func( h, mb_y-2, buf + h->sps->i_mb_width * 3 ); - } -} - - -/* Deblocking filter */ -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, -}; -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, -}; -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 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++ ) +#define PPIXEL(x, y) ( pix + (x) + (y)*i_stride ) + for( int y = 0; y < i_height; y++ ) { - if( tc0[i] < 0 ) - { - pix += 4*ystride; - continue; - } - 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 ) - { - int tc = tc0[i]; - int delta; - if( abs( p2 - p0 ) < beta ) - { - pix[-2*xstride] = p1 + x264_clip3( (( p2 + ((p0 + q0 + 1) >> 1)) >> 1) - p1, -tc0[i], tc0[i] ); - tc++; - } - 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] = x264_clip_uint8( p0 + delta ); /* p0' */ - pix[ 0*xstride] = x264_clip_uint8( q0 - delta ); /* q0' */ - } - pix += ystride; - } + /* left band */ + pixel_memset( PPIXEL(-i_padh, y), PPIXEL(0, y), i_padh>>b_chroma, sizeof(pixel)<>b_chroma, sizeof(pixel)<mb.i_mb_height - (1 << SLICE_MBAFF); + int b_start = mb_y == h->i_threadslice_start; + int b_end = mb_y == h->i_threadslice_end - (1 << SLICE_MBAFF); + if( mb_y & SLICE_MBAFF ) + return; + for( int i = 0; i < frame->i_plane; i++ ) { - const int tc = tc0[i]; - if( tc <= 0 ) + int h_shift = i && CHROMA_H_SHIFT; + int v_shift = i && CHROMA_V_SHIFT; + int stride = frame->i_stride[i]; + int width = 16*h->mb.i_mb_width; + int height = (pad_bot ? 16*(h->mb.i_mb_height - mb_y) >> SLICE_MBAFF : 16) >> v_shift; + int padh = PADH; + int padv = PADV >> v_shift; + // buffer: 2 chroma, 3 luma (rounded to 4) because deblocking goes beyond the top of the mb + if( b_end && !b_start ) + height += 4 >> (v_shift + SLICE_MBAFF); + pixel *pix; + int starty = 16*mb_y - 4*!b_start; + if( SLICE_MBAFF ) { - pix += 2*ystride; - continue; + // border samples for each field are extended separately + pix = frame->plane_fld[i] + (starty*stride >> v_shift); + plane_expand_border( pix, stride*2, width, height, padh, padv, pad_top, pad_bot, h_shift ); + plane_expand_border( pix+stride, stride*2, width, height, padh, padv, pad_top, pad_bot, h_shift ); + + height = (pad_bot ? 16*(h->mb.i_mb_height - mb_y) : 32) >> v_shift; + if( b_end && !b_start ) + height += 4 >> v_shift; + pix = frame->plane[i] + (starty*stride >> v_shift); + plane_expand_border( pix, stride, width, height, padh, padv, pad_top, pad_bot, h_shift ); } - for( d = 0; d < 2; d++ ) + else { - 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 ) - { - int delta = x264_clip3( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); - pix[-1*xstride] = x264_clip_uint8( p0 + delta ); /* p0' */ - pix[ 0*xstride] = x264_clip_uint8( q0 - delta ); /* q0' */ - } - pix += ystride; + pix = frame->plane[i] + (starty*stride >> v_shift); + plane_expand_border( pix, stride, width, height, padh, padv, pad_top, pad_bot, h_shift ); } } } -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 ) +void x264_frame_expand_border_filtered( x264_t *h, x264_frame_t *frame, int mb_y, int b_end ) { - int 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]; - 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 ) + /* 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 width = 16*h->mb.i_mb_width + 8; + int height = b_end ? (16*(h->mb.i_mb_height - mb_y) >> SLICE_MBAFF) + 16 : 16; + int padh = PADH - 4; + int padv = PADV - 8; + for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ ) + for( int i = 1; i < 4; i++ ) { - if(abs( p0 - q0 ) < ((alpha >> 2) + 2) ) + int stride = frame->i_stride[p]; + // buffer: 8 luma, to match the hpel filter + pixel *pix; + if( SLICE_MBAFF ) { - if( abs( p2 - p0 ) < beta ) /* p0', p1', p2' */ - { - const int p3 = pix[-4*xstride]; - 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 ) /* q0', q1', q2' */ - { - const int q3 = pix[3*xstride]; - 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; + pix = frame->filtered_fld[p][i] + (16*mb_y - 16) * stride - 4; + plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end, 0 ); + plane_expand_border( pix+stride, stride*2, width, height, padh, padv, b_start, b_end, 0 ); } - else /* p0', q0' */ - { - pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2; - pix[ 0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2; - } - } - pix += 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++ ) - { - 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 ) - { - pix[-1*xstride] = (2*p1 + p0 + q1 + 2) >> 2; /* p0' */ - pix[ 0*xstride] = (2*q1 + q0 + p1 + 2) >> 2; /* q0' */ + pix = frame->filtered[p][i] + (16*mb_y - 8) * stride - 4; + plane_expand_border( pix, stride, width, height << SLICE_MBAFF, padh, padv, b_start, b_end, 0 ); } - 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 *pix1, uint8_t *pix2, int i_stride, uint8_t bS[4], int i_qp, int b_chroma, x264_deblock_inter_t pf_inter ) +void x264_frame_expand_border_lowres( x264_frame_t *frame ) { - 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 ); + for( int i = 0; i < 4; i++ ) + plane_expand_border( frame->lowres[i], frame->i_stride_lowres, frame->i_width_lowres, frame->i_lines_lowres, PADH, PADV, 1, 1, 0 ); } -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 ) +void x264_frame_expand_border_chroma( x264_t *h, x264_frame_t *frame, int plane ) { - 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 ); + int v_shift = CHROMA_V_SHIFT; + plane_expand_border( frame->plane[plane], frame->i_stride[plane], 16*h->mb.i_mb_width, 16*h->mb.i_mb_height>>v_shift, + PADH, PADV>>v_shift, 1, 1, CHROMA_H_SHIFT ); } -void x264_frame_deblock_row( x264_t *h, int mb_y ) +void x264_frame_expand_border_mod16( x264_t *h, x264_frame_t *frame ) { - const int s8x8 = 2 * h->mb.i_mb_stride; - 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 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; mb_x += (~b_interlaced | mb_y)&1, mb_y ^= b_interlaced ) + for( int i = 0; i < frame->i_plane; i++ ) { - 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_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) ) - { - pixy -= 15*stridey; - pixu -= 7*strideuv; - pixv -= 7*strideuv; - } - - x264_prefetch_fenc( h, h->fdec, mb_x, mb_y ); - - 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 );\ - }\ - }\ - } + int i_width = h->param.i_width; + int h_shift = i && CHROMA_H_SHIFT; + int v_shift = i && CHROMA_V_SHIFT; + int i_height = h->param.i_height >> v_shift; + int i_padx = (h->mb.i_mb_width * 16 - h->param.i_width); + int i_pady = (h->mb.i_mb_height * 16 - h->param.i_height) >> v_shift; - #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]) )\ - *(uint32_t*)bS = 0x03030303;\ - else\ - {\ - *(uint32_t*)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\ - {\ - /* 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>>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;\ - if((h->mb.ref[0][i8p] != h->mb.ref[0][i8q] ||\ - 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_padx ) + { + for( int y = 0; y < i_height; y++ ) + pixel_memset( &frame->plane[i][y*frame->i_stride[i] + i_width], + &frame->plane[i][y*frame->i_stride[i] + i_width - 1-h_shift], + i_padx>>h_shift, sizeof(pixel)< 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;\ - DECLARE_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( *(uint32_t*)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( *(uint32_t*)bS )\ - FILTER_DIR( , i_dir);\ - }\ + if( i_pady ) + { + for( int y = i_height; y < i_height + i_pady; y++ ) + memcpy( &frame->plane[i][y*frame->i_stride[i]], + &frame->plane[i][(i_height-(~y&PARAM_INTERLACED)-1)*frame->i_stride[i]], + (i_width + i_padx) * sizeof(pixel) ); } - - DEBLOCK_DIR(0); - DEBLOCK_DIR(1); } - - if( !h->pps->b_cabac && h->pps->b_transform_8x8_mode ) - munge_cavlc_nnz( h, mb_y, h->mb.nnz_backup, restore_cavlc_nnz_row ); -} - -void x264_frame_deblock( x264_t *h ) -{ - int mb_y; - for( mb_y = 0; mb_y < h->sps->i_mb_height; mb_y += 1 + h->sh.b_mbaff ) - x264_frame_deblock_row( h, mb_y ); -} - -#ifdef HAVE_MMX -void x264_deblock_v_chroma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ); -void x264_deblock_h_chroma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ); -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 ); - -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 ); -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 ); - -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 -void x264_deblock_init( int cpu, x264_deblock_function_t *pf ) +void x264_expand_border_mbpair( x264_t *h, int mb_x, int mb_y ) { - pf->deblock_v_luma = deblock_v_luma_c; - pf->deblock_h_luma = deblock_h_luma_c; - pf->deblock_v_chroma = deblock_v_chroma_c; - pf->deblock_h_chroma = deblock_h_chroma_c; - pf->deblock_v_luma_intra = deblock_v_luma_intra_c; - pf->deblock_h_luma_intra = deblock_h_luma_intra_c; - pf->deblock_v_chroma_intra = deblock_v_chroma_intra_c; - pf->deblock_h_chroma_intra = deblock_h_chroma_intra_c; - -#ifdef HAVE_MMX - if( cpu&X264_CPU_MMXEXT ) + for( int i = 0; i < h->fenc->i_plane; i++ ) { - pf->deblock_v_chroma = x264_deblock_v_chroma_mmxext; - 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 - 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; - } + int v_shift = i && CHROMA_V_SHIFT; + int stride = h->fenc->i_stride[i]; + int height = h->param.i_height >> v_shift; + int pady = (h->mb.i_mb_height * 16 - h->param.i_height) >> v_shift; + pixel *fenc = h->fenc->plane[i] + 16*mb_x; + for( int y = height; y < height + pady; y++ ) + memcpy( fenc + y*stride, fenc + (height-1)*stride, 16*sizeof(pixel) ); } -#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 } - /* threading */ void x264_frame_cond_broadcast( x264_frame_t *frame, int i_lines_completed ) { @@ -900,6 +689,38 @@ void x264_frame_cond_wait( x264_frame_t *frame, int i_lines_completed ) x264_pthread_mutex_unlock( &frame->mutex ); } +void x264_threadslice_cond_broadcast( x264_t *h, int pass ) +{ + x264_pthread_mutex_lock( &h->mutex ); + h->i_threadslice_pass = pass; + if( pass > 0 ) + x264_pthread_cond_broadcast( &h->cv ); + x264_pthread_mutex_unlock( &h->mutex ); +} + +void x264_threadslice_cond_wait( x264_t *h, int pass ) +{ + x264_pthread_mutex_lock( &h->mutex ); + while( h->i_threadslice_pass < pass ) + x264_pthread_cond_wait( &h->cv, &h->mutex ); + x264_pthread_mutex_unlock( &h->mutex ); +} + +int x264_frame_new_slice( x264_t *h, x264_frame_t *frame ) +{ + if( h->param.i_slice_count_max ) + { + int slice_count; + if( h->param.b_sliced_threads ) + slice_count = x264_pthread_fetch_and_add( &frame->i_slice_count, 1, &frame->mutex ); + else + slice_count = frame->i_slice_count++; + if( slice_count >= h->param.i_slice_count_max ) + return -1; + } + return 0; +} + /* list operators */ void x264_frame_push( x264_frame_t **list, x264_frame_t *frame ) @@ -944,40 +765,125 @@ 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 ); + 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; + frame->b_corrupt = 0; + frame->i_slice_count = h->param.b_sliced_threads ? h->param.i_threads : 1; + + 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_sort( x264_frame_t **list, int b_dts ) +void x264_frame_push_blank_unused( x264_t *h, x264_frame_t *frame ) { - int i, b_ok; - do { - b_ok = 1; - for( i = 0; list[i+1]; i++ ) - { - int dtype = list[i]->i_type - list[i+1]->i_type; - int dtime = list[i]->i_frame - list[i+1]->i_frame; - int swap = b_dts ? dtype > 0 || ( dtype == 0 && dtime > 0 ) - : dtime > 0; - if( swap ) - { - XCHG( x264_frame_t*, list[i], list[i+1] ); - b_ok = 0; - } - } - } while( !b_ok ); + 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; +} + +void x264_weight_scale_plane( x264_t *h, pixel *dst, intptr_t i_dst_stride, pixel *src, intptr_t i_src_stride, + int i_width, int i_height, x264_weight_t *w ) +{ + /* Weight horizontal strips of height 16. This was found to be the optimal height + * in terms of the cache loads. */ + while( i_height > 0 ) + { + int x; + for( x = 0; x < i_width-8; x += 16 ) + w->weightfn[16>>2]( dst+x, i_dst_stride, src+x, i_src_stride, w, X264_MIN( i_height, 16 ) ); + if( x < i_width ) + w->weightfn[ 8>>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_sync_frame_list_init( x264_sync_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_sync_frame_list_delete( x264_sync_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_sync_frame_list_push( x264_sync_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 ); +} + +x264_frame_t *x264_sync_frame_list_pop( x264_sync_frame_list_t *slist ) +{ + x264_frame_t *frame; + x264_pthread_mutex_lock( &slist->mutex ); + while( !slist->i_size ) + x264_pthread_cond_wait( &slist->cv_fill, &slist->mutex ); + frame = slist->list[ --slist->i_size ]; + slist->list[ slist->i_size ] = NULL; + x264_pthread_cond_broadcast( &slist->cv_empty ); + x264_pthread_mutex_unlock( &slist->mutex ); + return frame; }