CABAC trellis opts part 4: x86_64 asm
[x262.git] / common / frame.c
index 90783cd..3d82627 100644 (file)
@@ -1,7 +1,7 @@
 /*****************************************************************************
- * frame.c: h264 encoder library
+ * frame.c: frame handling
  *****************************************************************************
- * Copyright (C) 2003-2008 x264 project
+ * Copyright (C) 2003-2011 x264 project
  *
  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
  *          Loren Merritt <lorenm@u.washington.edu>
  * 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, int b_fdec )
+static int align_plane_size( int x, int disalign )
 {
-    x264_frame_t *frame;
+    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_I420:
+        case X264_CSP_YV12:
+            return X264_CSP_NV12;
+        case X264_CSP_NV16:
+        case X264_CSP_I422:
+        case X264_CSP_YV16:
+            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 i_stride, i_width, i_lines, luma_plane_count;
+    int i_padv = PADV << PARAM_INTERLACED;
     int align = h->param.cpu&X264_CPU_CACHELINE_64 ? 64 : h->param.cpu&X264_CPU_CACHELINE_32 ? 32 : 16;
+    int disalign = h->param.cpu&X264_CPU_ALTIVEC ? 1<<9 : 1<<10;
 
     CHECKED_MALLOCZERO( frame, sizeof(x264_frame_t) );
 
     /* 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<<h->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( int 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;
+        }
     }
-
-    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( int i = 1; i < 3; i++ )
+    else if( i_csp == X264_CSP_I444 )
     {
-        CHECKED_MALLOC( frame->buffer[i], chroma_plane_size );
-        frame->plane[i] = frame->buffer[i] + (frame->i_stride[i] * i_padv + PADH)/2;
+        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;
+
+    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++ )
@@ -66,7 +118,7 @@ x264_frame_t *x264_frame_new( x264_t *h, int b_fdec )
 
     frame->i_poc = -1;
     frame->i_type = X264_TYPE_AUTO;
-    frame->i_qpplus1 = 0;
+    frame->i_qpplus1 = X264_QP_AUTO;
     frame->i_pts = -1;
     frame->i_frame = -1;
     frame->i_frame_num = -1;
@@ -83,19 +135,47 @@ x264_frame_t *x264_frame_new( x264_t *h, int b_fdec )
 
     frame->orig = frame;
 
-    /* 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 )
+    if( i_csp == X264_CSP_NV12 || i_csp == X264_CSP_NV16 )
     {
-        CHECKED_MALLOC( frame->buffer[0], 4*luma_plane_size );
-        for( int 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 chroma_padv = i_padv >> (i_csp == X264_CSP_NV12);
+        int chroma_plane_size = (frame->i_stride[1] * (frame->i_lines[1] + 2*chroma_padv));
+        CHECKED_MALLOC( frame->buffer[1], chroma_plane_size * sizeof(pixel) );
+        frame->plane[1] = frame->buffer[1] + frame->i_stride[1] * chroma_padv + PADH;
+        if( PARAM_INTERLACED )
+        {
+            CHECKED_MALLOC( frame->buffer_fld[1], chroma_plane_size * sizeof(pixel) );
+            frame->plane_fld[1] = frame->buffer_fld[1] + frame->i_stride[1] * chroma_padv + PADH;
+        }
     }
-    else
+
+    /* all 4 luma planes allocated together, since the cacheline split code
+     * requires them to be in-phase wrt cacheline alignment. */
+
+    for( int p = 0; p < luma_plane_count; p++ )
     {
-        CHECKED_MALLOC( frame->buffer[0], luma_plane_size );
-        frame->filtered[0] = frame->plane[0] = frame->buffer[0] + frame->i_stride[0] * i_padv + PADH;
+        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. */
+            CHECKED_MALLOC( frame->buffer[p], 4*luma_plane_size * sizeof(pixel) );
+            if( PARAM_INTERLACED )
+                CHECKED_MALLOC( frame->buffer_fld[p], 4*luma_plane_size * sizeof(pixel) );
+            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
+        {
+            CHECKED_MALLOC( frame->buffer[p], luma_plane_size * sizeof(pixel) );
+            if( PARAM_INTERLACED )
+                CHECKED_MALLOC( frame->buffer_fld[p], luma_plane_size * sizeof(pixel) );
+            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;
+        }
     }
 
     frame->b_duplicate = 0;
@@ -105,6 +185,9 @@ x264_frame_t *x264_frame_new( x264_t *h, int b_fdec )
         CHECKED_MALLOC( frame->mb_type, i_mb_count * sizeof(int8_t));
         CHECKED_MALLOC( frame->mb_partition, i_mb_count * sizeof(uint8_t));
         CHECKED_MALLOC( frame->mv[0], 2*16 * i_mb_count * sizeof(int16_t) );
+        CHECKED_MALLOC( frame->mv16x16, 2*(i_mb_count+1) * sizeof(int16_t) );
+        M32( frame->mv16x16[0] ) = 0;
+        frame->mv16x16++;
         CHECKED_MALLOC( frame->ref[0], 4 * i_mb_count * sizeof(int8_t) );
         if( h->param.i_bframe )
         {
@@ -117,25 +200,24 @@ x264_frame_t *x264_frame_new( x264_t *h, int b_fdec )
             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) );
+        CHECKED_MALLOC( frame->f_row_qp, i_lines/16 * sizeof(float) );
+        CHECKED_MALLOC( frame->f_row_qscale, i_lines/16 * sizeof(float) );
         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;
         }
+        if( PARAM_INTERLACED )
+            CHECKED_MALLOC( frame->field, i_mb_count * sizeof(uint8_t) );
     }
     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*PADV);
+            int luma_plane_size = align_plane_size( frame->i_stride_lowres * (frame->i_lines[0]/2 + 2*PADV), disalign );
 
-            CHECKED_MALLOC( frame->buffer_lowres[0], 4 * luma_plane_size );
+            CHECKED_MALLOC( frame->buffer_lowres[0], 4 * luma_plane_size * sizeof(pixel) );
             for( int i = 0; i < 4; i++ )
                 frame->lowres[i] = frame->buffer_lowres[0] + (frame->i_stride_lowres * PADV + PADH) + i * luma_plane_size;
 
@@ -145,13 +227,10 @@ x264_frame_t *x264_frame_new( x264_t *h, int b_fdec )
                     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) );
+            CHECKED_MALLOC( 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++ )
-                {
                     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) );
         }
@@ -184,7 +263,10 @@ void x264_frame_delete( x264_frame_t *frame )
     if( !frame->b_duplicate )
     {
         for( int i = 0; i < 4; i++ )
+        {
             x264_free( frame->buffer[i] );
+            x264_free( frame->buffer_fld[i] );
+        }
         for( int i = 0; i < 4; i++ )
             x264_free( frame->buffer_lowres[i] );
         for( int i = 0; i < X264_BFRAME_MAX+2; i++ )
@@ -199,19 +281,20 @@ void x264_frame_delete( x264_frame_t *frame )
         x264_free( frame->i_propagate_cost );
         for( int j = 0; j <= X264_BFRAME_MAX+1; j++ )
             for( int 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->f_row_qp );
+        x264_free( frame->f_row_qscale );
+        x264_free( frame->field );
         x264_free( frame->mb_type );
         x264_free( frame->mb_partition );
         x264_free( frame->mv[0] );
         x264_free( frame->mv[1] );
+        if( frame->mv16x16 )
+            x264_free( frame->mv16x16-1 );
         x264_free( frame->ref[0] );
         x264_free( frame->ref[1] );
         x264_pthread_mutex_destroy( &frame->mutex );
@@ -220,85 +303,218 @@ void x264_frame_delete( x264_frame_t *frame )
     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;
-    if( i_csp != X264_CSP_I420 && i_csp != X264_CSP_YV12 )
+    if( i_csp <= X264_CSP_NONE || i_csp >= X264_CSP_MAX )
+    {
+        x264_log( h, X264_LOG_ERROR, "Invalid input colorspace\n" );
+        return -1;
+    }
+
+#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, "Arg invalid CSP\n" );
+        x264_log( h, X264_LOG_ERROR, "This build of x264 requires 8-bit input. Rebuild to support high depth input.\n" );
         return -1;
     }
+#endif
 
     dst->i_type     = src->i_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;
 
-    for( int i = 0; i < 3; i++ )
+    uint8_t *pix[3];
+    int stride[3];
+    if ( i_csp >= X264_CSP_BGR )
     {
-        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 )
+         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 ? 2 : 0;
+         h->mc.plane_copy_deinterleave_rgb( dst->plane[1], dst->i_stride[1],
+                                            dst->plane[b], dst->i_stride[b],
+                                            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 );
+    }
+    else
+    {
+        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 += (height-1)*stride;
-            stride = -stride;
+            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 if( i_csp == X264_CSP_I420 || i_csp == X264_CSP_I422 || i_csp == X264_CSP_YV12 || i_csp == X264_CSP_YV16 )
+        {
+            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 //if( i_csp == X264_CSP_I444 || i_csp == X264_CSP_YV24 )
+        {
+            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 );
         }
-        h->mc.plane_copy( dst->plane[i], dst->i_stride[i], plane, stride, width, height );
     }
     return 0;
 }
 
+static void ALWAYS_INLINE pixel_memset( pixel *dst, pixel *src, int len, int size )
+{
+    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;
 
+    /* Align the input pointer if it isn't already */
+    if( (intptr_t)dstp & (WORD_SIZE - 1) )
+    {
+        if( size <= 2 && ((intptr_t)dstp & 3) )
+        {
+            if( size == 1 && ((intptr_t)dstp & 1) )
+                dstp[i++] = v1;
+            if( (intptr_t)dstp & 2 )
+            {
+                M16( dstp+i ) = v2;
+                i += 2;
+            }
+        }
+        if( WORD_SIZE == 8 && (intptr_t)dstp & 4 )
+        {
+            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;
+
+    /* Finish up the last few bytes */
+    if( size <= 2 )
+    {
+        if( i < len - 1 )
+        {
+            M16( dstp+i ) = v2;
+            i += 2;
+        }
+        if( size == 1 && i != len )
+            dstp[i] = v1;
+    }
+}
 
-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 )
+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 )
 {
 #define PPIXEL(x, y) ( pix + (x) + (y)*i_stride )
     for( int y = 0; y < i_height; y++ )
     {
         /* left band */
-        memset( PPIXEL(-i_padh, y), PPIXEL(0, y)[0], i_padh );
+        pixel_memset( PPIXEL(-i_padh, y), PPIXEL(0, y), i_padh>>b_chroma, sizeof(pixel)<<b_chroma );
         /* right band */
-        memset( PPIXEL(i_width, y), PPIXEL(i_width-1, y)[0], i_padh );
+        pixel_memset( PPIXEL(i_width, y), PPIXEL(i_width-1-b_chroma, y), i_padh>>b_chroma, sizeof(pixel)<<b_chroma );
     }
     /* upper band */
     if( b_pad_top )
         for( int y = 0; y < i_padv; y++ )
-            memcpy( PPIXEL(-i_padh, -y-1), PPIXEL(-i_padh, 0), i_width+2*i_padh );
+            memcpy( PPIXEL(-i_padh, -y-1), PPIXEL(-i_padh, 0), (i_width+2*i_padh) * sizeof(pixel) );
     /* lower band */
     if( b_pad_bottom )
         for( int y = 0; y < i_padv; y++ )
-            memcpy( PPIXEL(-i_padh, i_height+y), PPIXEL(-i_padh, i_height-1), i_width+2*i_padh );
+            memcpy( PPIXEL(-i_padh, i_height+y), PPIXEL(-i_padh, i_height-1), (i_width+2*i_padh) * sizeof(pixel) );
 #undef PPIXEL
 }
 
 void x264_frame_expand_border( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
 {
     int b_start = !mb_y;
-    if( mb_y & h->sh.b_mbaff )
+    if( mb_y & SLICE_MBAFF )
         return;
     for( int i = 0; i < frame->i_plane; i++ )
     {
+        int h_shift = i && CHROMA_H_SHIFT;
+        int v_shift = i && CHROMA_V_SHIFT;
         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;
+        int width = 16*h->mb.i_mb_width;
+        int height = (b_end ? 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
-        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 )
+            height += 4 >> (v_shift + SLICE_MBAFF);
+        pixel *pix;
+        if( SLICE_MBAFF )
         {
-            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 );
+            // border samples for each field are extended separately
+            pix = frame->plane_fld[i] + X264_MAX(0, (16*mb_y-4)*stride >> v_shift);
+            plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end, h_shift );
+            plane_expand_border( pix+stride, stride*2, width, height, padh, padv, b_start, b_end, h_shift );
+
+            height = (b_end ? 16*(h->mb.i_mb_height - mb_y) : 32) >> v_shift;
+            if( b_end && !b_start )
+                height += 4 >> v_shift;
+            pix = frame->plane[i] + X264_MAX(0, (16*mb_y-4)*stride >> v_shift);
+            plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end, h_shift );
         }
         else
         {
-            plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end );
+            pix = frame->plane[i] + X264_MAX(0, (16*mb_y-4)*stride >> v_shift);
+            plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end, h_shift );
         }
     }
 }
@@ -309,58 +525,83 @@ void x264_frame_expand_border_filtered( x264_t *h, x264_frame_t *frame, int mb_y
      * 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 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 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 )
+    for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
+        for( int i = 1; i < 4; i++ )
         {
-            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 stride = frame->i_stride[p];
+            // buffer: 8 luma, to match the hpel filter
+            pixel *pix;
+            if( SLICE_MBAFF )
+            {
+                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 );
+            }
+
+            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 );
         }
-        else
-            plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end );
-    }
 }
 
 void x264_frame_expand_border_lowres( x264_frame_t *frame )
 {
     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 );
+        plane_expand_border( frame->lowres[i], frame->i_stride_lowres, frame->i_width_lowres, frame->i_lines_lowres, PADH, PADV, 1, 1, 0 );
+}
+
+void x264_frame_expand_border_chroma( x264_t *h, x264_frame_t *frame, int plane )
+{
+    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_expand_border_mod16( x264_t *h, x264_frame_t *frame )
 {
     for( int i = 0; i < frame->i_plane; i++ )
     {
-        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_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;
 
         if( i_padx )
         {
             for( int 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 );
+                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)<<h_shift );
         }
         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&h->param.b_interlaced)-1)*frame->i_stride[i]],
-                        i_width + i_padx );
+                        &frame->plane[i][(i_height-(~y&PARAM_INTERLACED)-1)*frame->i_stride[i]],
+                        (i_width + i_padx) * sizeof(pixel) );
         }
     }
 }
 
+void x264_expand_border_mbpair( x264_t *h, int mb_x, int mb_y )
+{
+    for( int i = 0; i < h->fenc->i_plane; i++ )
+    {
+        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) );
+    }
+}
+
 /* threading */
 void x264_frame_cond_broadcast( x264_frame_t *frame, int i_lines_completed )
 {
@@ -439,6 +680,7 @@ x264_frame_t *x264_frame_pop_unused( x264_t *h, int b_fdec )
     frame->b_intra_calculated = 0;
     frame->b_scenecut = 1;
     frame->b_keyframe = 0;
+    frame->b_corrupt = 0;
 
     memset( frame->weight, 0, sizeof(frame->weight) );
     memset( frame->f_weighted_cost_delta, 0, sizeof(frame->f_weighted_cost_delta) );
@@ -468,35 +710,18 @@ x264_frame_t *x264_frame_pop_blank_unused( x264_t *h )
     return frame;
 }
 
-void x264_frame_sort( x264_frame_t **list, int b_dts )
-{
-    int b_ok;
-    do {
-        b_ok = 1;
-        for( int 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 );
-}
-
-void x264_weight_scale_plane( x264_t *h, uint8_t *dst, int i_dst_stride, uint8_t *src, int i_src_stride,
+void x264_weight_scale_plane( x264_t *h, pixel *dst, int i_dst_stride, pixel *src, int 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 )
     {
-        for( int x = 0; x < i_width; x += 16 )
+        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;
@@ -513,7 +738,7 @@ void x264_frame_delete_list( x264_frame_t **list )
     x264_free( list );
 }
 
-int x264_synch_frame_list_init( x264_synch_frame_list_t *slist, int max_size )
+int x264_sync_frame_list_init( x264_sync_frame_list_t *slist, int max_size )
 {
     if( max_size < 0 )
         return -1;
@@ -529,7 +754,7 @@ fail:
     return -1;
 }
 
-void x264_synch_frame_list_delete( x264_synch_frame_list_t *slist )
+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 );
@@ -537,7 +762,7 @@ void x264_synch_frame_list_delete( x264_synch_frame_list_t *slist )
     x264_frame_delete_list( slist->list );
 }
 
-void x264_synch_frame_list_push( x264_synch_frame_list_t *slist, x264_frame_t *frame )
+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 )
@@ -546,3 +771,16 @@ void x264_synch_frame_list_push( x264_synch_frame_list_t *slist, x264_frame_t *f
     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;
+}