]> git.sesse.net Git - x264/blobdiff - encoder/encoder.c
macroblock-level ratecontrol: improved vbv strictness, and improved quality when...
[x264] / encoder / encoder.c
index 1e7552cdc760d09aab5b2eaa4ed1374c395142ab..6ed7862e13991c314251de81d023b5b02106cb6b 100644 (file)
  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
  *****************************************************************************/
 
-#include <stdlib.h>
 #include <stdio.h>
 #include <string.h>
-
 #include <math.h>
 
+#ifdef __WIN32__
+#include <windows.h>
+#define pthread_t               HANDLE
+#define pthread_create(t,u,f,d) *(t)=CreateThread(NULL,0,f,d,0,NULL)
+#define pthread_join(t,s)       { WaitForSingleObject(t,INFINITE); \
+                                  CloseHandle(t); } 
+#define HAVE_PTHREAD 1
+
+#elif defined(SYS_BEOS)
+#include <kernel/OS.h>
+#define pthread_t               thread_id
+#define pthread_create(t,u,f,d) { *(t)=spawn_thread(f,"",10,d); \
+                                  resume_thread(*(t)); }
+#define pthread_join(t,s)       wait_for_thread(t,(long*)s)
+#define HAVE_PTHREAD 1
+
+#elif HAVE_PTHREAD
+#include <pthread.h>
+#endif
+
 #include "common/common.h"
 #include "common/cpu.h"
 
 #include "ratecontrol.h"
 #include "macroblock.h"
 
+#if VISUALIZE
+#include "common/visualize.h"
+#endif
+
 //#define DEBUG_MB_TYPE
 //#define DEBUG_DUMP_FRAME
 //#define DEBUG_BENCHMARK
@@ -64,31 +86,6 @@ static int64_t i_mtime_filter = 0;
  ******************************* x264 libs **********************************
  *
  ****************************************************************************/
-static int64_t x264_sqe( uint8_t *pix1, int i_pix_stride, uint8_t *pix2, int i_pix2_stride, int i_width, int i_height )
-{
-    int64_t i_sqe = 0;
-
-    int x, y;
-
-    for( y = 0; y < i_height; y++ )
-    {
-        for( x = 0; x < i_width; x++ )
-        {
-            int tmp;
-
-            tmp = pix1[y*i_pix_stride+x] - pix2[y*i_pix2_stride+x];
-
-            i_sqe += tmp * tmp;
-        }
-    }
-    return i_sqe;
-}
-
-static float x264_mse( int64_t i_sqe, int64_t i_size )
-{
-    return (double)i_sqe / ((double)65025.0 * (double)i_size);
-}
-
 static float x264_psnr( int64_t i_sqe, int64_t i_size )
 {
     double f_mse = (double)i_sqe / ((double)65025.0 * (double)i_size);
@@ -132,6 +129,7 @@ static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
 
     sh->i_type      = i_type;
     sh->i_first_mb  = 0;
+    sh->i_last_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
     sh->i_pps_id    = pps->i_id;
 
     sh->i_frame_num = i_frame;
@@ -149,7 +147,14 @@ static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
 
     sh->i_redundant_pic_cnt = 0;
 
-    sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
+    if( !h->mb.b_direct_auto_read )
+    {
+        if( h->mb.b_direct_auto_write )
+            sh->b_direct_spatial_mv_pred = ( h->stat.i_direct_score[1] > h->stat.i_direct_score[0] );
+        else
+            sh->b_direct_spatial_mv_pred = ( param->analyse.i_direct_mv_pred == X264_DIRECT_PRED_SPATIAL );
+    }
+    /* else b_direct_spatial_mv_pred was read from the 2pass statsfile */
 
     sh->b_num_ref_idx_override = 0;
     sh->i_num_ref_idx_l0_active = 1;
@@ -176,6 +181,7 @@ static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
 
     sh->i_cabac_init_idc = param->i_cabac_init_idc;
 
+    sh->i_qp = i_qp;
     sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
     sh->b_sp_for_swidth = 0;
     sh->i_qs_delta = 0;
@@ -183,7 +189,7 @@ static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
     /* If effective qp <= 15, deblocking would have no effect anyway */
     if( param->b_deblocking_filter
         && ( h->mb.b_variable_qp
-        || 15 < i_qp + X264_MAX(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
+        || 15 < i_qp + 2 * X264_MAX(param->i_deblocking_filter_alphac0, param->i_deblocking_filter_beta) ) )
     {
         sh->i_disable_deblocking_filter_idc = 0;
     }
@@ -293,7 +299,6 @@ static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal
         else
         {
             bs_write1( s, 0 );  /* adaptive_ref_pic_marking_mode_flag */
-            /* FIXME */
         }
     }
 
@@ -302,16 +307,6 @@ static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal
         bs_write_ue( s, sh->i_cabac_init_idc );
     }
     bs_write_se( s, sh->i_qp_delta );      /* slice qp delta */
-#if 0
-    if( sh->i_type == SLICE_TYPE_SP || sh->i_type == SLICE_TYPE_SI )
-    {
-        if( sh->i_type == SLICE_TYPE_SP )
-        {
-            bs_write1( s, sh->b_sp_for_swidth );
-        }
-        bs_write_se( s, sh->i_qs_delta );
-    }
-#endif
 
     if( sh->pps->b_deblocking_filter_control )
     {
@@ -332,48 +327,66 @@ static void x264_slice_header_write( bs_t *s, x264_slice_header_t *sh, int i_nal
  *
  ****************************************************************************/
 
-/****************************************************************************
- * x264_encoder_open:
- ****************************************************************************/
-x264_t *x264_encoder_open   ( x264_param_t *param )
+static int x264_validate_parameters( x264_t *h )
 {
-    x264_t *h = x264_malloc( sizeof( x264_t ) );
-    int i, i_slice;
-
-    /* Create a copy of param */
-    memcpy( &h->param, param, sizeof( x264_param_t ) );
-    if( h->param.rc.psz_stat_out )
-        h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
-    if( h->param.rc.psz_stat_in )
-        h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
-    if( h->param.rc.psz_rc_eq )
-        h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
-
-    /* Check parameters validity */
-    if( param->i_width <= 0  || param->i_height <= 0 )
+    if( h->param.i_width <= 0 || h->param.i_height <= 0 )
     {
         x264_log( h, X264_LOG_ERROR, "invalid width x height (%dx%d)\n",
-                  param->i_width, param->i_height );
-        free( h );
-        return NULL;
+                  h->param.i_width, h->param.i_height );
+        return -1;
     }
 
-    if( param->i_width % 16 != 0 || param->i_height % 16 != 0 )
+    if( h->param.i_width % 2 || h->param.i_height % 2 )
     {
-        x264_log( h, X264_LOG_ERROR, "width %% 16 != 0 or height %% 16 != 0 (%dx%d)\n",
-                 param->i_width, param->i_height );
-        free( h );
-        return NULL;
+        x264_log( h, X264_LOG_ERROR, "width or height not divisible by 2 (%dx%d)\n",
+                  h->param.i_width, h->param.i_height );
+        return -1;
     }
-    if( param->i_csp != X264_CSP_I420 )
+    if( h->param.i_csp != X264_CSP_I420 )
     {
         x264_log( h, X264_LOG_ERROR, "invalid CSP (only I420 supported)\n" );
-        free( h );
-        return NULL;
+        return -1;
     }
 
-    /* Fix parameters values */
-    h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 15 );
+    h->param.i_threads = x264_clip3( h->param.i_threads, 1, X264_SLICE_MAX );
+    h->param.i_threads = X264_MIN( h->param.i_threads, (h->param.i_height + 15) / 16 );
+#if !(HAVE_PTHREAD)
+    if( h->param.i_threads > 1 )
+    {
+        x264_log( h, X264_LOG_WARNING, "not compiled with pthread support!\n");
+        x264_log( h, X264_LOG_WARNING, "multislicing anyway, but you won't see any speed gain.\n" );
+    }
+#endif
+
+    if( h->param.rc.b_cbr )
+        h->param.rc.i_rf_constant = 0;
+    if( h->param.rc.i_rf_constant > 0 )
+        h->param.rc.i_qp_constant = h->param.rc.i_rf_constant;
+    h->param.rc.i_rf_constant = x264_clip3( h->param.rc.i_rf_constant, 0, 51 );
+    h->param.rc.i_qp_constant = x264_clip3( h->param.rc.i_qp_constant, 0, 51 );
+    if( !h->param.rc.b_cbr && h->param.rc.i_qp_constant == 0 )
+    {
+        h->mb.b_lossless = 1;
+        h->param.analyse.b_transform_8x8 = 0;
+        h->param.i_cqm_preset = X264_CQM_FLAT;
+        h->param.psz_cqm_file = NULL;
+        h->param.rc.f_ip_factor = 1;
+        h->param.rc.f_pb_factor = 1;
+        h->param.analyse.b_psnr = 0;
+        h->param.analyse.i_chroma_qp_offset = 0;
+        h->param.analyse.i_trellis = 0;
+        h->param.analyse.b_fast_pskip = 0;
+        h->param.analyse.i_noise_reduction = 0;
+    }
+
+    if( ( h->param.i_width % 16 || h->param.i_height % 16 ) && !h->mb.b_lossless )
+    {
+        x264_log( h, X264_LOG_WARNING, 
+                  "width or height not divisible by 16 (%dx%d), compression will suffer.\n",
+                  h->param.i_width, h->param.i_height );
+    }
+
+    h->param.i_frame_reference = x264_clip3( h->param.i_frame_reference, 1, 16 );
     if( h->param.i_keyint_max <= 0 )
         h->param.i_keyint_max = 1;
     h->param.i_keyint_min = x264_clip3( h->param.i_keyint_min, 1, h->param.i_keyint_max/2+1 );
@@ -381,26 +394,115 @@ x264_t *x264_encoder_open   ( x264_param_t *param )
     h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_BFRAME_MAX );
     h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
     h->param.b_bframe_pyramid = h->param.b_bframe_pyramid && h->param.i_bframe > 1;
-    h->frames.i_delay = h->param.i_bframe;
-    h->frames.i_max_ref0 = h->param.i_frame_reference;
-    h->frames.i_max_ref1 = h->param.b_bframe_pyramid ? 2
-                            : h->param.i_bframe ? 1 : 0;
-    h->frames.i_max_dpb = h->frames.i_max_ref0 + h->frames.i_max_ref1 + 1;
+    h->param.b_bframe_adaptive = h->param.b_bframe_adaptive && h->param.i_bframe > 0;
+    h->mb.b_direct_auto_write = h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
+                                && h->param.i_bframe
+                                && ( h->param.rc.b_stat_write || !h->param.rc.b_stat_read );
 
     h->param.i_deblocking_filter_alphac0 = x264_clip3( h->param.i_deblocking_filter_alphac0, -6, 6 );
     h->param.i_deblocking_filter_beta    = x264_clip3( h->param.i_deblocking_filter_beta, -6, 6 );
 
-    h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, -1, 2 );
-
-    h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 5 );
-    if( h->param.analyse.inter & X264_ANALYSE_PSUB8x8 )
-        h->param.analyse.inter |= X264_ANALYSE_PSUB16x16;
+    h->param.i_cabac_init_idc = x264_clip3( h->param.i_cabac_init_idc, 0, 2 );
+
+    if( h->param.i_cqm_preset < X264_CQM_FLAT || h->param.i_cqm_preset > X264_CQM_CUSTOM )
+        h->param.i_cqm_preset = X264_CQM_FLAT;
+
+    if( h->param.analyse.i_me_method < X264_ME_DIA ||
+        h->param.analyse.i_me_method > X264_ME_ESA )
+        h->param.analyse.i_me_method = X264_ME_HEX;
+    if( h->param.analyse.i_me_range < 4 )
+        h->param.analyse.i_me_range = 4;
+    if( h->param.analyse.i_me_range > 16 && h->param.analyse.i_me_method <= X264_ME_HEX )
+        h->param.analyse.i_me_range = 16;
+    h->param.analyse.i_subpel_refine = x264_clip3( h->param.analyse.i_subpel_refine, 1, 6 );
+    h->param.analyse.b_bframe_rdo = h->param.analyse.b_bframe_rdo && h->param.analyse.i_subpel_refine >= 6;
+    h->param.analyse.b_mixed_references = h->param.analyse.b_mixed_references && h->param.i_frame_reference > 1;
+    h->param.analyse.inter &= X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|
+                              X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
+    h->param.analyse.intra &= X264_ANALYSE_I4x4|X264_ANALYSE_I8x8;
+    if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
+        h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;
+    if( !h->param.analyse.b_transform_8x8 )
+    {
+        h->param.analyse.inter &= ~X264_ANALYSE_I8x8;
+        h->param.analyse.intra &= ~X264_ANALYSE_I8x8;
+    }
+    h->param.analyse.i_chroma_qp_offset = x264_clip3(h->param.analyse.i_chroma_qp_offset, -12, 12);
+    if( !h->param.b_cabac )
+        h->param.analyse.i_trellis = 0;
+    h->param.analyse.i_trellis = x264_clip3( h->param.analyse.i_trellis, 0, 2 );
+    h->param.analyse.i_noise_reduction = x264_clip3( h->param.analyse.i_noise_reduction, 0, 1<<16 );
+
+    {
+        const x264_level_t *l = x264_levels;
+        while( l->level_idc != 0 && l->level_idc != h->param.i_level_idc )
+            l++;
+        if( l->level_idc == 0 )
+        {
+            x264_log( h, X264_LOG_ERROR, "invalid level_idc: %d\n", h->param.i_level_idc );
+            return -1;
+        }
+        if( h->param.analyse.i_mv_range <= 0 )
+            h->param.analyse.i_mv_range = l->mv_range;
+        else
+            h->param.analyse.i_mv_range = x264_clip3(h->param.analyse.i_mv_range, 32, 2048);
+    }
 
     if( h->param.rc.f_qblur < 0 )
         h->param.rc.f_qblur = 0;
     if( h->param.rc.f_complexity_blur < 0 )
         h->param.rc.f_complexity_blur = 0;
 
+    /* ensure the booleans are 0 or 1 so they can be used in math */
+#define BOOLIFY(x) h->param.x = !!h->param.x
+    BOOLIFY( b_cabac );
+    BOOLIFY( b_deblocking_filter );
+    BOOLIFY( analyse.b_transform_8x8 );
+    BOOLIFY( analyse.b_weighted_bipred );
+    BOOLIFY( analyse.b_bidir_me );
+    BOOLIFY( analyse.b_chroma_me );
+    BOOLIFY( analyse.b_fast_pskip );
+    BOOLIFY( rc.b_cbr );
+    BOOLIFY( rc.b_stat_write );
+    BOOLIFY( rc.b_stat_read );
+#undef BOOLIFY
+
+    return 0;
+}
+
+/****************************************************************************
+ * x264_encoder_open:
+ ****************************************************************************/
+x264_t *x264_encoder_open   ( x264_param_t *param )
+{
+    x264_t *h = x264_malloc( sizeof( x264_t ) );
+    int i;
+
+    memset( h, 0, sizeof( x264_t ) );
+
+    /* Create a copy of param */
+    memcpy( &h->param, param, sizeof( x264_param_t ) );
+
+    if( x264_validate_parameters( h ) < 0 )
+    {
+        x264_free( h );
+        return NULL;
+    }
+
+    if( h->param.psz_cqm_file )
+        if( x264_cqm_parse_file( h, h->param.psz_cqm_file ) < 0 )
+        {
+            x264_free( h );
+            return NULL;
+        }
+
+    if( h->param.rc.psz_stat_out )
+        h->param.rc.psz_stat_out = strdup( h->param.rc.psz_stat_out );
+    if( h->param.rc.psz_stat_in )
+        h->param.rc.psz_stat_in = strdup( h->param.rc.psz_stat_in );
+    if( h->param.rc.psz_rc_eq )
+        h->param.rc.psz_rc_eq = strdup( h->param.rc.psz_rc_eq );
+
     /* VUI */
     if( h->param.vui.i_sar_width > 0 && h->param.vui.i_sar_height > 0 )
     {
@@ -445,12 +547,13 @@ x264_t *x264_encoder_open   ( x264_param_t *param )
 
     /* Init x264_t */
     h->out.i_nal = 0;
-    h->out.i_bitstream = 1000000; /* FIXME estimate max size (idth/height) */
+    h->out.i_bitstream = X264_MAX( 1000000, h->param.i_width * h->param.i_height * 1.7
+        * ( h->param.rc.b_cbr ? pow( 0.5, h->param.rc.i_qp_min )
+          : pow( 0.5, h->param.rc.i_qp_constant ) * X264_MAX( 1, h->param.rc.f_ip_factor )));
     h->out.p_bitstream = x264_malloc( h->out.i_bitstream );
 
     h->i_frame = 0;
     h->i_frame_num = 0;
-    h->i_poc   = 0;
     h->i_idr_pic_id = 0;
 
     h->sps = &h->sps_array[0];
@@ -458,11 +561,22 @@ x264_t *x264_encoder_open   ( x264_param_t *param )
 
     h->pps = &h->pps_array[0];
     x264_pps_init( h->pps, 0, &h->param, h->sps);
+
+    x264_validate_levels( h );
+
+    x264_cqm_init( h );
     
     h->mb.i_mb_count = h->sps->i_mb_width * h->sps->i_mb_height;
 
     /* Init frames. */
-    for( i = 0; i < X264_BFRAME_MAX + 1; i++ )
+    h->frames.i_delay = h->param.i_bframe;
+    h->frames.i_max_ref0 = h->param.i_frame_reference;
+    h->frames.i_max_ref1 = h->sps->vui.i_num_reorder_frames;
+    h->frames.i_max_dpb  = h->sps->vui.i_max_dec_frame_buffering + 1;
+    h->frames.b_have_lowres = !h->param.rc.b_stat_read
+        && ( h->param.rc.b_cbr || h->param.rc.i_rf_constant || h->param.b_bframe_adaptive );
+
+    for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
     {
         h->frames.current[i] = NULL;
         h->frames.next[i]    = NULL;
@@ -479,20 +593,19 @@ x264_t *x264_encoder_open   ( x264_param_t *param )
     h->frames.reference[h->frames.i_max_dpb] = NULL;
     h->frames.i_last_idr = - h->param.i_keyint_max;
     h->frames.i_input    = 0;
+    h->frames.last_nonb  = NULL;
 
     h->i_ref0 = 0;
     h->i_ref1 = 0;
 
     h->fdec = h->frames.reference[0];
 
-    /* init mb cache */
     x264_macroblock_cache_init( h );
-
-    /* init cabac adaptive model */
-    x264_cabac_model_init( &h->cabac );
+    x264_rdo_init( );
 
     /* init CPU functions */
     x264_predict_16x16_init( h->param.cpu, h->predict_16x16 );
+    x264_predict_8x8c_init( h->param.cpu, h->predict_8x8c );
     x264_predict_8x8_init( h->param.cpu, h->predict_8x8 );
     x264_predict_4x4_init( h->param.cpu, h->predict_4x4 );
 
@@ -500,29 +613,17 @@ x264_t *x264_encoder_open   ( x264_param_t *param )
     x264_dct_init( h->param.cpu, &h->dctf );
     x264_mc_init( h->param.cpu, &h->mc );
     x264_csp_init( h->param.cpu, h->param.i_csp, &h->csp );
+    x264_quant_init( h, h->param.cpu, &h->quantf );
+    x264_deblock_init( h->param.cpu, &h->loopf );
+
+    memcpy( h->pixf.mbcmp,
+            ( h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1 ) ? h->pixf.sad : h->pixf.satd,
+            sizeof(h->pixf.mbcmp) );
 
     /* rate control */
     if( x264_ratecontrol_new( h ) < 0 )
         return NULL;
 
-    h->i_last_intra_size = 0;
-    h->i_last_inter_size = 0;
-
-    /* stat */
-    for( i_slice = 0; i_slice < 5; i_slice++ )
-    {
-        h->stat.i_slice_count[i_slice] = 0;
-        h->stat.i_slice_size[i_slice] = 0;
-        h->stat.i_slice_qp[i_slice] = 0;
-
-        h->stat.i_sqe_global[i_slice] = 0;
-        h->stat.f_psnr_average[i_slice] = 0.0;
-        h->stat.f_psnr_mean_y[i_slice] = h->stat.f_psnr_mean_u[i_slice] = h->stat.f_psnr_mean_v[i_slice] = 0.0;
-        
-        for( i = 0; i < 18; i++ )
-            h->stat.i_mb_count[i_slice][i] = 0;
-    }
-
     x264_log( h, X264_LOG_INFO, "using cpu capabilities %s%s%s%s%s%s\n",
              param->cpu&X264_CPU_MMX ? "MMX " : "",
              param->cpu&X264_CPU_MMXEXT ? "MMXEXT " : "",
@@ -531,9 +632,36 @@ x264_t *x264_encoder_open   ( x264_param_t *param )
              param->cpu&X264_CPU_3DNOW ? "3DNow! " : "",
              param->cpu&X264_CPU_ALTIVEC ? "Altivec " : "" );
 
+    h->thread[0] = h;
+    h->i_thread_num = 0;
+    for( i = 1; i < param->i_threads; i++ )
+        h->thread[i] = x264_malloc( sizeof(x264_t) );
+
     return h;
 }
 
+/****************************************************************************
+ * x264_encoder_reconfig:
+ ****************************************************************************/
+int x264_encoder_reconfig( x264_t *h, x264_param_t *param )
+{
+    h->param.i_bframe_bias = param->i_bframe_bias;
+    h->param.i_deblocking_filter_alphac0 = param->i_deblocking_filter_alphac0;
+    h->param.i_deblocking_filter_beta    = param->i_deblocking_filter_beta;
+    h->param.analyse.i_me_method = param->analyse.i_me_method;
+    h->param.analyse.i_me_range = param->analyse.i_me_range;
+    h->param.analyse.i_subpel_refine = param->analyse.i_subpel_refine;
+    h->param.analyse.i_trellis = param->analyse.i_trellis;
+    h->param.analyse.intra = param->analyse.intra;
+    h->param.analyse.inter = param->analyse.inter;
+
+    memcpy( h->pixf.mbcmp,
+            ( h->mb.b_lossless || h->param.analyse.i_subpel_refine <= 1 ) ? h->pixf.sad : h->pixf.satd,
+            sizeof(h->pixf.mbcmp) );
+
+    return x264_validate_parameters( h );
+}
+
 /* internal usage */
 static void x264_nal_start( x264_t *h, int i_type, int i_ref_idc )
 {
@@ -572,7 +700,7 @@ int x264_encoder_headers( x264_t *h, x264_nal_t **pp_nal, int *pi_nal )
     {
         /* identify ourself */
         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
-        x264_sei_version_write( &h->out.bs );
+        x264_sei_version_write( h, &h->out.bs );
         x264_nal_end( h );
 
         /* generate sequence parameters */
@@ -613,7 +741,7 @@ static x264_frame_t *x264_frame_get( x264_frame_t *list[X264_BFRAME_MAX+1] )
 {
     x264_frame_t *frame = list[0];
     int i;
-    for( i = 0; i < X264_BFRAME_MAX && list[i]; i++ )
+    for( i = 0; list[i]; i++ )
         list[i] = list[i+1];
     return frame;
 }
@@ -623,7 +751,7 @@ static void x264_frame_sort( x264_frame_t *list[X264_BFRAME_MAX+1], int b_dts )
     int i, b_ok;
     do {
         b_ok = 1;
-        for( i = 0; i < X264_BFRAME_MAX && list[i+1]; i++ )
+        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;
@@ -631,9 +759,7 @@ static void x264_frame_sort( x264_frame_t *list[X264_BFRAME_MAX+1], int b_dts )
                              : dtime > 0;
             if( swap )
             {
-                x264_frame_t *tmp = list[i+1];
-                list[i+1] = list[i];
-                list[i] = tmp;
+                XCHG( x264_frame_t*, list[i], list[i+1] );
                 b_ok = 0;
             }
         }
@@ -673,10 +799,7 @@ static inline void x264_reference_build_list( x264_t *h, int i_poc, int i_slice_
         {
             if( h->fref0[i]->i_poc < h->fref0[i+1]->i_poc )
             {
-                x264_frame_t *tmp = h->fref0[i+1];
-
-                h->fref0[i+1] = h->fref0[i];
-                h->fref0[i] = tmp;
+                XCHG( x264_frame_t*, h->fref0[i], h->fref0[i+1] );
                 b_ok = 0;
                 break;
             }
@@ -690,10 +813,7 @@ static inline void x264_reference_build_list( x264_t *h, int i_poc, int i_slice_
         {
             if( h->fref1[i]->i_poc > h->fref1[i+1]->i_poc )
             {
-                x264_frame_t *tmp = h->fref1[i+1];
-
-                h->fref1[i+1] = h->fref1[i];
-                h->fref1[i] = tmp;
+                XCHG( x264_frame_t*, h->fref1[i], h->fref1[i+1] );
                 b_ok = 0;
                 break;
             }
@@ -714,8 +834,9 @@ static inline void x264_reference_build_list( x264_t *h, int i_poc, int i_slice_
             }
     }
 
-    h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
     h->i_ref1 = X264_MIN( h->i_ref1, h->frames.i_max_ref1 );
+    h->i_ref0 = X264_MIN( h->i_ref0, h->frames.i_max_ref0 );
+    h->i_ref0 = X264_MIN( h->i_ref0, 16 - h->i_ref1 );
 }
 
 static inline void x264_reference_update( x264_t *h )
@@ -740,18 +861,13 @@ static inline void x264_reference_update( x264_t *h )
 
     /* move lowres copy of the image to the ref frame */
     for( i = 0; i < 4; i++)
-    {
-        uint8_t *tmp = h->fdec->lowres[i];
-        h->fdec->lowres[i] = h->fenc->lowres[i];
-        h->fenc->lowres[i] = tmp;
-    }
+        XCHG( uint8_t*, h->fdec->lowres[i], h->fenc->lowres[i] );
 
     /* adaptive B decision needs a pointer, since it can't use the ref lists */
     if( h->sh.i_type != SLICE_TYPE_B )
         h->frames.last_nonb = h->fdec;
 
     /* move frame in the buffer */
-    /* FIXME: override to forget earliest pts, not earliest dts */
     h->fdec = h->frames.reference[h->frames.i_max_dpb-1];
     for( i = h->frames.i_max_dpb-1; i > 0; i-- )
     {
@@ -808,52 +924,36 @@ static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_slice_type,
         /* Nothing to do ? */
     }
 
-    /* get adapative cabac model if needed */
-    if( h->param.b_cabac )
-    {
-        if( h->param.i_cabac_init_idc == -1 )
-        {
-            h->sh.i_cabac_init_idc = x264_cabac_model_get( &h->cabac, i_slice_type );
-        }
-    }
-
     x264_macroblock_slice_init( h );
 }
 
-static inline void x264_slice_write( x264_t *h, int i_nal_type, int i_nal_ref_idc )
+static int x264_slice_write( x264_t *h )
 {
     int i_skip;
     int mb_xy;
     int i;
 
-    /* Init stats */
-    h->stat.frame.i_hdr_bits  =
-    h->stat.frame.i_itex_bits =
-    h->stat.frame.i_ptex_bits =
-    h->stat.frame.i_misc_bits =
-    h->stat.frame.i_intra_cost =
-    h->stat.frame.i_inter_cost = 0;
-    for( i = 0; i < 18; i++ )
-        h->stat.frame.i_mb_count[i] = 0;
+    /* init stats */
+    memset( &h->stat.frame, 0, sizeof(h->stat.frame) );
 
     /* Slice */
-    x264_nal_start( h, i_nal_type, i_nal_ref_idc );
+    x264_nal_start( h, h->i_nal_type, h->i_nal_ref_idc );
 
     /* Slice header */
-    x264_slice_header_write( &h->out.bs, &h->sh, i_nal_ref_idc );
+    x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
     if( h->param.b_cabac )
     {
         /* alignment needed */
         bs_align_1( &h->out.bs );
 
         /* init cabac */
-        x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.pps->i_pic_init_qp + h->sh.i_qp_delta, h->sh.i_cabac_init_idc );
+        x264_cabac_context_init( &h->cabac, h->sh.i_type, h->sh.i_qp, h->sh.i_cabac_init_idc );
         x264_cabac_encode_init ( &h->cabac, &h->out.bs );
     }
-    h->mb.i_last_qp = h->pps->i_pic_init_qp + h->sh.i_qp_delta;
+    h->mb.i_last_qp = h->sh.i_qp;
     h->mb.i_last_dqp = 0;
 
-    for( mb_xy = 0, i_skip = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
+    for( mb_xy = h->sh.i_first_mb, i_skip = 0; mb_xy < h->sh.i_last_mb; mb_xy++ )
     {
         const int i_mb_y = mb_xy / h->sps->i_mb_width;
         const int i_mb_x = mb_xy % h->sps->i_mb_width;
@@ -877,38 +977,24 @@ static inline void x264_slice_write( x264_t *h, int i_nal_type, int i_nal_ref_id
         TIMER_STOP( i_mtime_encode );
 
         TIMER_START( i_mtime_write );
-        if( IS_SKIP( h->mb.i_type ) )
+        if( h->param.b_cabac )
         {
-            if( h->param.b_cabac )
-            {
-                if( mb_xy > 0 )
-                {
-                    /* not end_of_slice_flag */
-                    x264_cabac_encode_terminal( &h->cabac, 0 );
-                }
+            if( mb_xy > h->sh.i_first_mb )
+                x264_cabac_encode_terminal( &h->cabac, 0 );
 
+            if( IS_SKIP( h->mb.i_type ) )
                 x264_cabac_mb_skip( h, 1 );
-            }
             else
             {
-                i_skip++;
+                if( h->sh.i_type != SLICE_TYPE_I )
+                    x264_cabac_mb_skip( h, 0 );
+                x264_macroblock_write_cabac( h, &h->cabac );
             }
         }
         else
         {
-            if( h->param.b_cabac )
-            {
-                if( mb_xy > 0 )
-                {
-                    /* not end_of_slice_flag */
-                    x264_cabac_encode_terminal( &h->cabac, 0 );
-                }
-                if( h->sh.i_type != SLICE_TYPE_I )
-                {
-                    x264_cabac_mb_skip( h, 0 );
-                }
-                x264_macroblock_write_cabac( h, &h->out.bs );
-            }
+            if( IS_SKIP( h->mb.i_type ) )
+                i_skip++;
             else
             {
                 if( h->sh.i_type != SLICE_TYPE_I )
@@ -921,10 +1007,38 @@ static inline void x264_slice_write( x264_t *h, int i_nal_type, int i_nal_ref_id
         }
         TIMER_STOP( i_mtime_write );
 
+#if VISUALIZE
+        if( h->param.b_visualize )
+            x264_visualize_mb( h );
+#endif
+
         /* save cache */
         x264_macroblock_cache_save( h );
 
+        /* accumulate mb stats */
         h->stat.frame.i_mb_count[h->mb.i_type]++;
+        if( !IS_SKIP(h->mb.i_type) && !IS_INTRA(h->mb.i_type) && !IS_DIRECT(h->mb.i_type) )
+        {
+            if( h->mb.i_partition != D_8x8 )
+                h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_partition ] ] += 4;
+            else
+                for( i = 0; i < 4; i++ )
+                    h->stat.frame.i_mb_count_size[ x264_mb_partition_pixel_table[ h->mb.i_sub_partition[i] ] ] ++;
+            if( h->param.i_frame_reference > 1 )
+            {
+                for( i = 0; i < 4; i++ )
+                {
+                    int i_ref = h->mb.cache.ref[0][ x264_scan8[4*i] ];
+                    if( i_ref >= 0 )
+                        h->stat.frame.i_mb_count_ref[i_ref] ++;
+                }
+            }
+        }
+        if( h->mb.i_cbp_luma && !IS_INTRA(h->mb.i_type) )
+        {
+            h->stat.frame.i_mb_count_8x8dct[0] ++;
+            h->stat.frame.i_mb_count_8x8dct[1] += h->mb.b_transform_8x8;
+        }
 
         if( h->mb.b_variable_qp )
             x264_ratecontrol_mb(h, bs_pos(&h->out.bs) - mb_spos);
@@ -942,16 +1056,8 @@ static inline void x264_slice_write( x264_t *h, int i_nal_type, int i_nal_ref_id
 
     if( h->param.b_cabac )
     {
-        int i_cabac_word;
         x264_cabac_encode_flush( &h->cabac );
-        /* TODO cabac stuffing things (p209) */
-        i_cabac_word = (((3 * h->cabac.i_sym_cnt - 3 * 96 * h->sps->i_mb_width * h->sps->i_mb_height)/32) - bs_pos( &h->out.bs)/8)/3;
 
-        while( i_cabac_word > 0 )
-        {
-            bs_write( &h->out.bs, 16, 0x0000 );
-            i_cabac_word--;
-        }
     }
     else
     {
@@ -967,6 +1073,86 @@ static inline void x264_slice_write( x264_t *h, int i_nal_type, int i_nal_ref_id
                               - h->stat.frame.i_itex_bits
                               - h->stat.frame.i_ptex_bits
                               - h->stat.frame.i_hdr_bits;
+
+    return 0;
+}
+
+static inline int x264_slices_write( x264_t *h )
+{
+    int i_frame_size;
+
+#if VISUALIZE
+    if( h->param.b_visualize )
+        x264_visualize_init( h );
+#endif
+
+    if( h->param.i_threads == 1 )
+    {
+        x264_ratecontrol_threads_start( h );
+        x264_slice_write( h );
+        i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
+    }
+    else
+    {
+        int i_nal = h->out.i_nal;
+        int i_bs_size = h->out.i_bitstream / h->param.i_threads;
+        int i;
+        /* duplicate contexts */
+        for( i = 0; i < h->param.i_threads; i++ )
+        {
+            x264_t *t = h->thread[i];
+            if( i > 0 )
+            {
+                memcpy( t, h, sizeof(x264_t) );
+                t->out.p_bitstream += i*i_bs_size;
+                bs_init( &t->out.bs, t->out.p_bitstream, i_bs_size );
+                t->i_thread_num = i;
+            }
+            t->sh.i_first_mb = (i    * h->sps->i_mb_height / h->param.i_threads) * h->sps->i_mb_width;
+            t->sh.i_last_mb = ((i+1) * h->sps->i_mb_height / h->param.i_threads) * h->sps->i_mb_width;
+            t->out.i_nal = i_nal + i;
+        }
+        x264_ratecontrol_threads_start( h );
+
+        /* dispatch */
+#if HAVE_PTHREAD
+        {
+            pthread_t handles[X264_SLICE_MAX];
+            void *status;
+            for( i = 0; i < h->param.i_threads; i++ )
+                pthread_create( &handles[i], NULL, (void*)x264_slice_write, (void*)h->thread[i] );
+            for( i = 0; i < h->param.i_threads; i++ )
+                pthread_join( handles[i], &status );
+        }
+#else
+        for( i = 0; i < h->param.i_threads; i++ )
+            x264_slice_write( h->thread[i] );
+#endif
+
+        /* merge contexts */
+        i_frame_size = h->out.nal[i_nal].i_payload;
+        for( i = 1; i < h->param.i_threads; i++ )
+        {
+            int j;
+            x264_t *t = h->thread[i];
+            h->out.nal[i_nal+i] = t->out.nal[i_nal+i];
+            i_frame_size += t->out.nal[i_nal+i].i_payload;
+            // all entries in stat.frame are ints
+            for( j = 0; j < sizeof(h->stat.frame) / sizeof(int); j++ )
+                ((int*)&h->stat.frame)[j] += ((int*)&t->stat.frame)[j];
+        }
+        h->out.i_nal = i_nal + h->param.i_threads;
+    }
+
+#if VISUALIZE
+    if( h->param.b_visualize )
+    {
+        x264_visualize_show( h );
+        x264_visualize_close( h );
+    }
+#endif
+
+    return i_frame_size;
 }
 
 /****************************************************************************
@@ -991,6 +1177,7 @@ int     x264_encoder_encode( x264_t *h,
     int     i_nal_type;
     int     i_nal_ref_idc;
     int     i_slice_type;
+    int     i_frame_size;
 
     int i;
 
@@ -1012,11 +1199,15 @@ int     x264_encoder_encode( x264_t *h,
 
         x264_frame_copy_picture( h, fenc, pic_in );
 
+        if( h->param.i_width % 16 || h->param.i_height % 16 )
+            x264_frame_expand_border_mod16( h, fenc );
+
         fenc->i_frame = h->frames.i_input++;
 
         x264_frame_put( h->frames.next, fenc );
 
-        x264_frame_init_lowres( h->param.cpu, fenc );
+        if( h->frames.b_have_lowres )
+            x264_frame_init_lowres( h->param.cpu, fenc );
 
         if( h->frames.i_input <= h->frames.i_delay )
         {
@@ -1108,8 +1299,9 @@ do_encode:
         i_slice_type = SLICE_TYPE_B;
     }
 
+    h->fdec->i_poc =
+    h->fenc->i_poc = 2 * (h->fenc->i_frame - h->frames.i_last_idr);
     h->fdec->i_type = h->fenc->i_type;
-    h->fdec->i_poc  = h->fenc->i_poc;
     h->fdec->i_frame = h->fenc->i_frame;
     h->fenc->b_kept_as_ref =
     h->fdec->b_kept_as_ref = i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE;
@@ -1117,23 +1309,21 @@ do_encode:
 
 
     /* ------------------- Init                ----------------------------- */
+    /* build ref list 0/1 */
+    x264_reference_build_list( h, h->fdec->i_poc, i_slice_type );
+
     /* Init the rate control */
-    x264_ratecontrol_start( h, i_slice_type );
+    x264_ratecontrol_start( h, i_slice_type, h->fenc->i_qpplus1 );
     i_global_qp = x264_ratecontrol_qp( h );
+
     pic_out->i_qpplus1 =
     h->fdec->i_qpplus1 = i_global_qp + 1;
 
-    /* build ref list 0/1 */
-    x264_reference_build_list( h, h->fdec->i_poc, i_slice_type );
-
     if( i_slice_type == SLICE_TYPE_B )
         x264_macroblock_bipred_init( h );
 
-    /* increase frame num but only once for B frame */
-    if( i_slice_type != SLICE_TYPE_B || h->sh.i_type != SLICE_TYPE_B )
-    {
+    if( h->fenc->b_kept_as_ref )
         h->i_frame_num++;
-    }
 
     /* ------------------------ Create slice header  ----------------------- */
     x264_slice_init( h, i_nal_type, i_slice_type, i_global_qp );
@@ -1143,14 +1333,35 @@ do_encode:
     h->out.i_nal = 0;
     bs_init( &h->out.bs, h->out.p_bitstream, h->out.i_bitstream );
 
+    if(h->param.b_aud){
+        int pic_type;
+
+        if(i_slice_type == SLICE_TYPE_I)
+            pic_type = 0;
+        else if(i_slice_type == SLICE_TYPE_P)
+            pic_type = 1;
+        else if(i_slice_type == SLICE_TYPE_B)
+            pic_type = 2;
+        else
+            pic_type = 7;
+
+        x264_nal_start(h, NAL_AUD, NAL_PRIORITY_DISPOSABLE);
+        bs_write(&h->out.bs, 3, pic_type);
+        bs_rbsp_trailing(&h->out.bs);
+        x264_nal_end(h);
+    }
+
+    h->i_nal_type = i_nal_type;
+    h->i_nal_ref_idc = i_nal_ref_idc;
+
     /* Write SPS and PPS */
-    if( i_nal_type == NAL_SLICE_IDR )
+    if( i_nal_type == NAL_SLICE_IDR && h->param.b_repeat_headers )
     {
         if( h->fenc->i_frame == 0 )
         {
             /* identify ourself */
             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
-            x264_sei_version_write( &h->out.bs );
+            x264_sei_version_write( h, &h->out.bs );
             x264_nal_end( h );
         }
 
@@ -1165,8 +1376,8 @@ do_encode:
         x264_nal_end( h );
     }
 
-    /* Write the slice */
-    x264_slice_write( h, i_nal_type, i_nal_ref_idc );
+    /* Write frame */
+    i_frame_size = x264_slices_write( h );
 
     /* restore CPU state (before using float again) */
     x264_cpu_restore( h->param.cpu );
@@ -1174,9 +1385,10 @@ do_encode:
     if( i_slice_type == SLICE_TYPE_P && !h->param.rc.b_stat_read 
         && h->param.i_scenecut_threshold >= 0 )
     {
-        int i_mb_i = h->stat.frame.i_mb_count[I_4x4] + h->stat.frame.i_mb_count[I_16x16];
-        int i_mb_p = h->stat.frame.i_mb_count[P_L0] + h->stat.frame.i_mb_count[P_8x8];
-        int i_mb_s = h->stat.frame.i_mb_count[P_SKIP];
+        const int *mbs = h->stat.frame.i_mb_count;
+        int i_mb_i = mbs[I_16x16] + mbs[I_8x8] + mbs[I_4x4];
+        int i_mb_p = mbs[P_L0] + mbs[P_8x8];
+        int i_mb_s = mbs[P_SKIP];
         int i_mb   = h->sps->i_mb_width * h->sps->i_mb_height;
         int64_t i_inter_cost = h->stat.frame.i_inter_cost;
         int64_t i_intra_cost = h->stat.frame.i_intra_cost;
@@ -1184,7 +1396,7 @@ do_encode:
         float f_bias;
         int i_gop_size = h->fenc->i_frame - h->frames.i_last_idr;
         float f_thresh_max = h->param.i_scenecut_threshold / 100.0;
-        /* ratio of 10 pulled out of thin air */
+        /* magic numbers pulled out of thin air */
         float f_thresh_min = f_thresh_max * h->param.i_keyint_min
                              / ( h->param.i_keyint_max * 4 );
         if( h->param.i_keyint_min == h->param.i_keyint_max )
@@ -1211,19 +1423,11 @@ do_encode:
         /* Bad P will be reencoded as I */
         if( i_mb_s < i_mb &&
             i_inter_cost >= (1.0 - f_bias) * i_intra_cost )
-            /* i_mb_i >= (1.0 - f_bias) * i_mb ) */
-            /*
-            h->out.nal[h->out.i_nal-1].i_payload > h->i_last_intra_size +
-            h->i_last_intra_size * (3+h->i_last_intra_qp - i_global_qp) / 16 &&
-            i_mb_count[I_4x4] + i_mb_count[I_16x16] > i_mb_count[P_SKIP] + i_mb_count[P_L0]/2 &&
-            h->out.nal[h->out.i_nal-1].i_payload > 2 * h->i_last_inter_size &&
-            h->frames.i_last_i > 4)*/
         {
             int b;
 
-            x264_log( h, X264_LOG_DEBUG, "scene cut at %d size=%d Icost:%.0f Pcost:%.0f ratio:%.3f bias=%.3f lastIDR:%d (I:%d P:%d Skip:%d)\n",
+            x264_log( h, X264_LOG_DEBUG, "scene cut at %d Icost:%.0f Pcost:%.0f ratio:%.3f bias=%.3f lastIDR:%d (I:%d P:%d S:%d)\n",
                       h->fenc->i_frame,
-                      h->out.nal[h->out.i_nal-1].i_payload,
                       (double)i_intra_cost, (double)i_inter_cost,
                       (double)i_inter_cost / i_intra_cost,
                       f_bias, i_gop_size,
@@ -1258,16 +1462,12 @@ do_encode:
                 x264_frame_t *tmp;
 
                 /* Reset */
-                h->i_poc       = 0;
                 h->i_frame_num = 0;
 
                 /* Reinit field of fenc */
                 h->fenc->i_type = X264_TYPE_IDR;
                 h->fenc->i_poc = 0;
 
-                /* Next Poc */
-                h->i_poc += 2;
-
                 /* Put enqueued frames back in the pool */
                 while( (tmp = x264_frame_get( h->frames.current ) ) != NULL )
                     x264_frame_put( h->frames.next, tmp );
@@ -1279,17 +1479,11 @@ do_encode:
             }
             goto do_encode;
         }
-        h->i_last_inter_size = h->out.nal[h->out.i_nal-1].i_payload;
-    }
-    else
-    {
-        h->i_last_intra_size = h->out.nal[h->out.i_nal-1].i_payload;
-        h->i_last_intra_qp = i_global_qp;
     }
 
     /* End bitstream, set output  */
     *pi_nal = h->out.i_nal;
-    *pp_nal = &h->out.nal[0];
+    *pp_nal = h->out.nal;
 
     /* Set output picture properties */
     if( i_slice_type == SLICE_TYPE_I )
@@ -1300,42 +1494,64 @@ do_encode:
         pic_out->i_type = X264_TYPE_B;
     pic_out->i_pts = h->fenc->i_pts;
 
-    /* ---------------------- Update encoder state ------------------------- */
-    /* update cabac */
-    if( h->param.b_cabac )
-    {
-        x264_cabac_model_update( &h->cabac, i_slice_type, h->sh.pps->i_pic_init_qp + h->sh.i_qp_delta );
+    pic_out->img.i_plane = h->fdec->i_plane;
+    for(i = 0; i < 4; i++){
+        pic_out->img.i_stride[i] = h->fdec->i_stride[i];
+        pic_out->img.plane[i] = h->fdec->plane[i];
     }
 
+    /* ---------------------- Update encoder state ------------------------- */
+
+    /* update rc */
+    x264_cpu_restore( h->param.cpu );
+    x264_ratecontrol_end( h, i_frame_size * 8 );
+
     /* handle references */
     if( i_nal_ref_idc != NAL_PRIORITY_DISPOSABLE )
-    {
         x264_reference_update( h );
-    }
+    x264_frame_put( h->frames.unused, h->fenc );
 
     /* increase frame count */
     h->i_frame++;
 
     /* restore CPU state (before using float again) */
-    /* XXX: not needed? (done above) */
     x264_cpu_restore( h->param.cpu );
 
-    /* update rc */
-    x264_ratecontrol_end( h, h->out.nal[h->out.i_nal-1].i_payload * 8 );
-
-    x264_frame_put( h->frames.unused, h->fenc );
+    x264_noise_reduction_update( h );
 
     TIMER_STOP( i_mtime_encode_frame );
 
     /* ---------------------- Compute/Print statistics --------------------- */
     /* Slice stat */
     h->stat.i_slice_count[i_slice_type]++;
-    h->stat.i_slice_size[i_slice_type] += bs_pos( &h->out.bs ) / 8 + NALU_OVERHEAD;
+    h->stat.i_slice_size[i_slice_type] += i_frame_size + NALU_OVERHEAD;
     h->stat.i_slice_qp[i_slice_type] += i_global_qp;
 
-    for( i = 0; i < 18; i++ )
-    {
+    for( i = 0; i < 19; i++ )
         h->stat.i_mb_count[h->sh.i_type][i] += h->stat.frame.i_mb_count[i];
+    for( i = 0; i < 2; i++ )
+        h->stat.i_mb_count_8x8dct[i] += h->stat.frame.i_mb_count_8x8dct[i];
+    if( h->sh.i_type != SLICE_TYPE_I )
+    {
+        for( i = 0; i < 7; i++ )
+            h->stat.i_mb_count_size[h->sh.i_type][i] += h->stat.frame.i_mb_count_size[i];
+        for( i = 0; i < 16; i++ )
+            h->stat.i_mb_count_ref[h->sh.i_type][i] += h->stat.frame.i_mb_count_ref[i];
+    }
+    if( i_slice_type == SLICE_TYPE_B )
+    {
+        h->stat.i_direct_frames[ h->sh.b_direct_spatial_mv_pred ] ++;
+        if( h->mb.b_direct_auto_write )
+        {
+            //FIXME somewhat arbitrary time constants
+            if( h->stat.i_direct_score[0] + h->stat.i_direct_score[1] > h->mb.i_mb_count )
+            {
+                for( i = 0; i < 2; i++ )
+                    h->stat.i_direct_score[i] = h->stat.i_direct_score[i] * 9/10;
+            }
+            for( i = 0; i < 2; i++ )
+                h->stat.i_direct_score[i] += h->stat.frame.i_direct_score[i];
+        }
     }
 
     if( h->param.analyse.b_psnr )
@@ -1343,9 +1559,10 @@ do_encode:
         int64_t i_sqe_y, i_sqe_u, i_sqe_v;
 
         /* PSNR */
-        i_sqe_y = x264_sqe( frame_psnr->plane[0], frame_psnr->i_stride[0], h->fenc->plane[0], h->fenc->i_stride[0], h->param.i_width, h->param.i_height );
-        i_sqe_u = x264_sqe( frame_psnr->plane[1], frame_psnr->i_stride[1], h->fenc->plane[1], h->fenc->i_stride[1], h->param.i_width/2, h->param.i_height/2);
-        i_sqe_v = x264_sqe( frame_psnr->plane[2], frame_psnr->i_stride[2], h->fenc->plane[2], h->fenc->i_stride[2], h->param.i_width/2, h->param.i_height/2);
+        i_sqe_y = x264_pixel_ssd_wxh( &h->pixf, frame_psnr->plane[0], frame_psnr->i_stride[0], h->fenc->plane[0], h->fenc->i_stride[0], h->param.i_width, h->param.i_height );
+        i_sqe_u = x264_pixel_ssd_wxh( &h->pixf, frame_psnr->plane[1], frame_psnr->i_stride[1], h->fenc->plane[1], h->fenc->i_stride[1], h->param.i_width/2, h->param.i_height/2);
+        i_sqe_v = x264_pixel_ssd_wxh( &h->pixf, frame_psnr->plane[2], frame_psnr->i_stride[2], h->fenc->plane[2], h->fenc->i_stride[2], h->param.i_width/2, h->param.i_height/2);
+        x264_cpu_restore( h->param.cpu );
 
         h->stat.i_sqe_global[i_slice_type] += i_sqe_y + i_sqe_u + i_sqe_v;
         h->stat.f_psnr_average[i_slice_type] += x264_psnr( i_sqe_y + i_sqe_u + i_sqe_v, 3 * h->param.i_width * h->param.i_height / 2 );
@@ -1365,28 +1582,27 @@ do_encode:
     }
     
     x264_log( h, X264_LOG_DEBUG,
-                  "frame=%4d QP=%i NAL=%d Slice:%c Poc:%-3d I4x4:%-4d I16x16:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
+                  "frame=%4d QP=%i NAL=%d Slice:%c Poc:%-3d I:%-4d P:%-4d SKIP:%-4d size=%d bytes%s\n",
               h->i_frame - 1,
               i_global_qp,
               i_nal_ref_idc,
               i_slice_type == SLICE_TYPE_I ? 'I' : (i_slice_type == SLICE_TYPE_P ? 'P' : 'B' ),
               frame_psnr->i_poc,
-              h->stat.frame.i_mb_count[I_4x4],
-              h->stat.frame.i_mb_count[I_16x16],
+              h->stat.frame.i_mb_count_i,
               h->stat.frame.i_mb_count_p,
               h->stat.frame.i_mb_count_skip,
-              h->out.nal[h->out.i_nal-1].i_payload,
+              i_frame_size,
               psz_message );
 
 
 #ifdef DEBUG_MB_TYPE
 {
-    static const char mb_chars[] = { 'i', 'I', 'C', 'P', '8', 'S',
+    static const char mb_chars[] = { 'i', 'i', 'I', 'C', 'P', '8', 'S',
         'D', '<', 'X', 'B', 'X', '>', 'B', 'B', 'B', 'B', '8', 'S' };
     int mb_xy;
     for( mb_xy = 0; mb_xy < h->sps->i_mb_width * h->sps->i_mb_height; mb_xy++ )
     {
-        if( h->mb.type[mb_xy] < 18 && h->mb.type[mb_xy] >= 0 )
+        if( h->mb.type[mb_xy] < 19 && h->mb.type[mb_xy] >= 0 )
             fprintf( stderr, "%c ", mb_chars[ h->mb.type[mb_xy] ] );
         else
             fprintf( stderr, "? " );
@@ -1400,16 +1616,6 @@ do_encode:
 #ifdef DEBUG_DUMP_FRAME
     /* Dump reconstructed frame */
     x264_frame_dump( h, frame_psnr, "fdec.yuv" );
-#endif
-#if 0
-    if( h->i_ref0 > 0 )
-    {
-        x264_frame_dump( h, h->fref0[0], "ref0.yuv" );
-    }
-    if( h->i_ref1 > 0 )
-    {
-        x264_frame_dump( h, h->fref1[0], "ref1.yuv" );
-    }
 #endif
     return 0;
 }
@@ -1447,20 +1653,19 @@ void    x264_encoder_close  ( x264_t *h )
             if( h->param.analyse.b_psnr )
             {
                 x264_log( h, X264_LOG_INFO,
-                          "slice %s:%-4d Avg QP:%5.2f Avg size:%6.0f PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f MSE*Size:%5.3f\n",
+                          "slice %s:%-5d Avg QP:%5.2f  size:%6.0f  PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f\n",
                           slice_name[i_slice],
                           i_count,
                           (double)h->stat.i_slice_qp[i_slice] / i_count,
                           (double)h->stat.i_slice_size[i_slice] / i_count,
                           h->stat.f_psnr_mean_y[i_slice] / i_count, h->stat.f_psnr_mean_u[i_slice] / i_count, h->stat.f_psnr_mean_v[i_slice] / i_count,
                           h->stat.f_psnr_average[i_slice] / i_count,
-                          x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ),
-                          x264_mse( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) * h->stat.i_slice_size[i_slice] / i_count );
+                          x264_psnr( h->stat.i_sqe_global[i_slice], i_count * i_yuv_size ) );
             }
             else
             {
                 x264_log( h, X264_LOG_INFO,
-                          "slice %s:%-4d Avg QP:%5.2f Avg size:%6.0f\n",
+                          "slice %s:%-5d Avg QP:%5.2f  size:%6.0f\n",
                           slice_name[i_slice],
                           i_count,
                           (double)h->stat.i_slice_qp[i_slice] / i_count,
@@ -1475,64 +1680,113 @@ void    x264_encoder_close  ( x264_t *h )
         const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_I];
         const double i_count = h->stat.i_slice_count[SLICE_TYPE_I] * h->mb.i_mb_count / 100.0;
         x264_log( h, X264_LOG_INFO,
-                  "slice I   Avg I4x4:%.1f%%  I16x16:%.1f%%\n",
-                  i_mb_count[I_4x4]  / i_count,
-                  i_mb_count[I_16x16]/ i_count );
+                  "mb I  I16..4: %4.1f%% %4.1f%% %4.1f%%\n",
+                  i_mb_count[I_16x16]/ i_count,
+                  i_mb_count[I_8x8]  / i_count,
+                  i_mb_count[I_4x4]  / i_count );
     }
     if( h->stat.i_slice_count[SLICE_TYPE_P] > 0 )
     {
         const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_P];
+        const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_P];
         const double i_count = h->stat.i_slice_count[SLICE_TYPE_P] * h->mb.i_mb_count / 100.0;
         x264_log( h, X264_LOG_INFO,
-                  "slice P   Avg I4x4:%.1f%%  I16x16:%.1f%%  P:%.1f%%  P8x8:%.1f%%  PSKIP:%.1f%%\n",
-                  i_mb_count[I_4x4]  / i_count,
+                  "mb P  I16..4: %4.1f%% %4.1f%% %4.1f%%  P16..4: %4.1f%% %4.1f%% %4.1f%% %4.1f%% %4.1f%%    skip:%4.1f%%\n",
                   i_mb_count[I_16x16]/ i_count,
-                  i_mb_count[P_L0]   / i_count,
-                  i_mb_count[P_8x8]  / i_count,
+                  i_mb_count[I_8x8]  / i_count,
+                  i_mb_count[I_4x4]  / i_count,
+                  i_mb_size[PIXEL_16x16] / (i_count*4),
+                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
+                  i_mb_size[PIXEL_8x8] / (i_count*4),
+                  (i_mb_size[PIXEL_8x4] + i_mb_size[PIXEL_4x8]) / (i_count*4),
+                  i_mb_size[PIXEL_4x4] / (i_count*4),
                   i_mb_count[P_SKIP] / i_count );
     }
     if( h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
     {
         const int64_t *i_mb_count = h->stat.i_mb_count[SLICE_TYPE_B];
+        const int64_t *i_mb_size = h->stat.i_mb_count_size[SLICE_TYPE_B];
         const double i_count = h->stat.i_slice_count[SLICE_TYPE_B] * h->mb.i_mb_count / 100.0;
         x264_log( h, X264_LOG_INFO,
-                  "slice B   Avg I4x4:%.1f%%  I16x16:%.1f%%  P:%.1f%%  B:%.1f%%  B8x8:%.1f%%  DIRECT:%.1f%%  BSKIP:%.1f%%\n",
-                  i_mb_count[I_4x4]    / i_count,
+                  "mb B  I16..4: %4.1f%% %4.1f%% %4.1f%%  B16..8: %4.1f%% %4.1f%% %4.1f%%  direct:%4.1f%%  skip:%4.1f%%\n",
                   i_mb_count[I_16x16]  / i_count,
-                  (i_mb_count[B_L0_L0] + i_mb_count[B_L1_L1] + i_mb_count[B_L1_L0] + i_mb_count[B_L0_L1]) / i_count,
-                  (i_mb_count[B_BI_BI] + i_mb_count[B_L0_BI] + i_mb_count[B_L1_BI] + i_mb_count[B_BI_L0] + i_mb_count[B_BI_L1]) / i_count,
-                  i_mb_count[B_8x8]    / i_count,
+                  i_mb_count[I_8x8]    / i_count,
+                  i_mb_count[I_4x4]    / i_count,
+                  i_mb_size[PIXEL_16x16] / (i_count*4),
+                  (i_mb_size[PIXEL_16x8] + i_mb_size[PIXEL_8x16]) / (i_count*4),
+                  i_mb_size[PIXEL_8x8] / (i_count*4),
                   i_mb_count[B_DIRECT] / i_count,
                   i_mb_count[B_SKIP]   / i_count );
     }
 
+    x264_ratecontrol_summary( h );
+
     if( h->stat.i_slice_count[SLICE_TYPE_I] + h->stat.i_slice_count[SLICE_TYPE_P] + h->stat.i_slice_count[SLICE_TYPE_B] > 0 )
     {
         const int i_count = h->stat.i_slice_count[SLICE_TYPE_I] +
                             h->stat.i_slice_count[SLICE_TYPE_P] +
                             h->stat.i_slice_count[SLICE_TYPE_B];
         float fps = (float) h->param.i_fps_num / h->param.i_fps_den;
+#define SUM3(p) (p[SLICE_TYPE_I] + p[SLICE_TYPE_P] + p[SLICE_TYPE_B])
+#define SUM3b(p,o) (p[SLICE_TYPE_I][o] + p[SLICE_TYPE_P][o] + p[SLICE_TYPE_B][o])
+        float f_bitrate = fps * SUM3(h->stat.i_slice_size) / i_count / 125;
 
-        if( h->param.analyse.b_psnr )
-            x264_log( h, X264_LOG_INFO,
-                      "PSNR Mean Y:%5.2f U:%5.2f V:%5.2f Avg:%5.2f Global:%5.2f kb/s:%.1f\n",
-                      (h->stat.f_psnr_mean_y[SLICE_TYPE_I] + h->stat.f_psnr_mean_y[SLICE_TYPE_P] + h->stat.f_psnr_mean_y[SLICE_TYPE_B]) / i_count,
-                      (h->stat.f_psnr_mean_u[SLICE_TYPE_I] + h->stat.f_psnr_mean_u[SLICE_TYPE_P] + h->stat.f_psnr_mean_u[SLICE_TYPE_B]) / i_count,
-                      (h->stat.f_psnr_mean_v[SLICE_TYPE_I] + h->stat.f_psnr_mean_v[SLICE_TYPE_P] + h->stat.f_psnr_mean_v[SLICE_TYPE_B]) / i_count,
+        if( h->param.analyse.b_transform_8x8 )
+        {
+            int64_t i_i8x8 = SUM3b( h->stat.i_mb_count, I_8x8 );
+            int64_t i_intra = i_i8x8 + SUM3b( h->stat.i_mb_count, I_4x4 )
+                                     + SUM3b( h->stat.i_mb_count, I_16x16 );
+            x264_log( h, X264_LOG_INFO, "8x8 transform  intra:%.1f%%  inter:%.1f%%\n",
+                      100. * i_i8x8 / i_intra,
+                      100. * h->stat.i_mb_count_8x8dct[1] / h->stat.i_mb_count_8x8dct[0] );
+        }
 
-                      (h->stat.f_psnr_average[SLICE_TYPE_I] + h->stat.f_psnr_average[SLICE_TYPE_P] + h->stat.f_psnr_average[SLICE_TYPE_B]) / i_count,
+        if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_AUTO
+            && h->stat.i_slice_count[SLICE_TYPE_B] )
+        {
+            x264_log( h, X264_LOG_INFO, "direct mvs  spatial:%.1f%%  temporal:%.1f%%\n",
+                      h->stat.i_direct_frames[1] * 100. / h->stat.i_slice_count[SLICE_TYPE_B],
+                      h->stat.i_direct_frames[0] * 100. / h->stat.i_slice_count[SLICE_TYPE_B] );
+        }
 
-                      x264_psnr( h->stat.i_sqe_global[SLICE_TYPE_I] + h->stat.i_sqe_global[SLICE_TYPE_P]+ h->stat.i_sqe_global[SLICE_TYPE_B],
-                                 i_count * i_yuv_size ),
-                      fps * 8*(h->stat.i_slice_size[SLICE_TYPE_I]+h->stat.i_slice_size[SLICE_TYPE_P]+h->stat.i_slice_size[SLICE_TYPE_B]) / i_count / 1000 );
-        else
+        if( h->param.i_frame_reference > 1 )
+        {
+            int i_slice;
+            for( i_slice = 0; i_slice < 2; i_slice++ )
+            {
+                char buf[200];
+                char *p = buf;
+                int64_t i_den = 0;
+                int i_max = 0;
+                for( i = 0; i < h->param.i_frame_reference; i++ )
+                    if( h->stat.i_mb_count_ref[i_slice][i] )
+                    {
+                        i_den += h->stat.i_mb_count_ref[i_slice][i];
+                        i_max = i;
+                    }
+                if( i_max == 0 )
+                    continue;
+                for( i = 0; i <= i_max; i++ )
+                    p += sprintf( p, " %4.1f%%", 100. * h->stat.i_mb_count_ref[i_slice][i] / i_den );
+                x264_log( h, X264_LOG_INFO, "ref %c %s\n", i_slice==SLICE_TYPE_P ? 'P' : 'B', buf );
+            }
+        }
+
+        if( h->param.analyse.b_psnr )
             x264_log( h, X264_LOG_INFO,
-                      "kb/s:%.1f\n",
-                      fps * 8*(h->stat.i_slice_size[SLICE_TYPE_I]+h->stat.i_slice_size[SLICE_TYPE_P]+h->stat.i_slice_size[SLICE_TYPE_B]) / i_count / 1000 );
+                      "PSNR Mean Y:%6.3f U:%6.3f V:%6.3f Avg:%6.3f Global:%6.3f kb/s:%.2f\n",
+                      SUM3( h->stat.f_psnr_mean_y ) / i_count,
+                      SUM3( h->stat.f_psnr_mean_u ) / i_count,
+                      SUM3( h->stat.f_psnr_mean_v ) / i_count,
+                      SUM3( h->stat.f_psnr_average ) / i_count,
+                      x264_psnr( SUM3( h->stat.i_sqe_global ), i_count * i_yuv_size ),
+                      f_bitrate );
+        else
+            x264_log( h, X264_LOG_INFO, "kb/s:%.1f\n", f_bitrate );
     }
 
     /* frames */
-    for( i = 0; i < X264_BFRAME_MAX + 1; i++ )
+    for( i = 0; i < X264_BFRAME_MAX + 3; i++ )
     {
         if( h->frames.current[i] ) x264_frame_delete( h->frames.current[i] );
         if( h->frames.next[i] )    x264_frame_delete( h->frames.next[i] );
@@ -1557,6 +1811,8 @@ void    x264_encoder_close  ( x264_t *h )
 
     x264_macroblock_cache_end( h );
     x264_free( h->out.p_bitstream );
+    for( i = 1; i < h->param.i_threads; i++ )
+        x264_free( h->thread[i] );
     x264_free( h );
 }