Obe

Makefile

@@ -22,6 +22,7 @@ SRCS = common/mc.c common/predict.c common/pixel.c common/macroblock.c \
        common/mvpred.c common/bitstream.c \
        encoder/analyse.c encoder/me.c encoder/ratecontrol.c \
        encoder/set.c encoder/macroblock.c encoder/cabac.c \
+       encoder/speed.c \
        encoder/cavlc.c encoder/encoder.c encoder/lookahead.c
 
 SRCCLI = x264.c input/input.c input/timecode.c input/raw.c input/y4m.c \

common/common.c

@@ -150,6 +150,12 @@ void x264_param_default( x264_param_t *param )
     param->rc.i_zones = 0;
     param->rc.b_mb_tree = 1;
 
+    // speedcontrol
+    param->sc.f_speed = 0;
+    param->sc.i_buffer_size = 12;
+    param->sc.f_buffer_init = 0.75;
+    param->sc.max_preset = SC_PRESETS;
+
     /* Log */
     param->pf_log = x264_log_default;
     param->p_log_private = NULL;
@@ -1145,6 +1151,14 @@ int x264_param_parse( x264_param_t *p, const char *name, const char *value )
         p->rc.f_complexity_blur = atof(value);
     OPT("zones")
         p->rc.psz_zones = strdup(value);
+    OPT("speed")
+        p->sc.f_speed = atof(value);
+    OPT("speed-bufsize")
+        p->sc.i_buffer_size = atoi(value);
+    OPT("speed-init")
+        p->sc.f_buffer_init = atof(value);
+    OPT("speed-alt-timer")
+        p->sc.b_alt_timer = atobool(value);
     OPT("crop-rect")
         b_error |= sscanf( value, "%u,%u,%u,%u", &p->crop_rect.i_left, &p->crop_rect.i_top,
                                                  &p->crop_rect.i_right, &p->crop_rect.i_bottom ) != 4;
@@ -1474,6 +1488,8 @@ char *x264_param2string( x264_param_t *p, int b_res )
         s += sprintf( s, "bitdepth=%d ", BIT_DEPTH );
     }
 
+    // FIXME speedcontrol stuff
+
     if( p->b_opencl )
         s += sprintf( s, "opencl=%d ", p->b_opencl );
     s += sprintf( s, "cabac=%d", p->b_cabac );

common/common.h

@@ -118,6 +118,7 @@ do {\
 
 #define LOG2_16(x) (31 - x264_clz((x)|1))
 
+#define SC_PRESETS 13
 /****************************************************************************
  * Includes
  ****************************************************************************/
@@ -514,6 +515,7 @@ typedef struct x264_lookahead_t
 } x264_lookahead_t;
 
 typedef struct x264_ratecontrol_t   x264_ratecontrol_t;
+typedef struct x264_speedcontrol_t  x264_speedcontrol_t;
 
 typedef struct x264_left_table_t
 {
@@ -969,6 +971,7 @@ struct x264_t
 
     /* rate control encoding only */
     x264_ratecontrol_t *rc;
+    x264_speedcontrol_t *sc;
 
     /* stats */
     struct

common/osdep.c

@@ -47,10 +47,14 @@ extern int ptw32_processInitialized;
 
 int64_t x264_mdate( void )
 {
-#if SYS_WINDOWS
+#ifdef __MINGW32__
     struct timeb tb;
     ftime( &tb );
     return ((int64_t)tb.time * 1000 + (int64_t)tb.millitm) * 1000;
+#elif SYS_LINUX
+    struct timespec ts_current;
+    clock_gettime( CLOCK_MONOTONIC, &ts_current );
+    return (int64_t)ts_current.tv_sec * 1000000 + (int64_t)ts_current.tv_nsec / 1000;
 #else
     struct timeval tv_date;
     gettimeofday( &tv_date, NULL );

common/quant.c

@@ -730,7 +730,7 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf )
     {
         pf->quant_4x4 = x264_quant_4x4_avx2;
         pf->quant_4x4_dc = x264_quant_4x4_dc_avx2;
-        pf->quant_8x8 = x264_quant_8x8_avx2;
+        //pf->quant_8x8 = x264_quant_8x8_avx2;
         pf->quant_4x4x4 = x264_quant_4x4x4_avx2;
         pf->dequant_4x4 = x264_dequant_4x4_avx2;
         pf->dequant_8x8 = x264_dequant_8x8_avx2;

common/x86/mc-a.asm

@@ -1276,6 +1276,10 @@ AVG16_CACHELINE_LOOP_SSSE3 j, k
 %endrep
 %endif ; !HIGH_BIT_DEPTH
 
+;=============================================================================
+; hpel_filter_c
+;=============================================================================
+
 ;=============================================================================
 ; pixel copy
 ;=============================================================================

configure

@@ -499,6 +499,7 @@ case $host_os in
         SYS="LINUX"
         define HAVE_MALLOC_H
         libm="-lm"
+        LDFLAGS="$LDFLAGS -lrt"
         ;;
     gnu*)
         SYS="HURD"

encoder/encoder.c

@@ -1328,24 +1328,31 @@ static int x264_validate_parameters( x264_t *h, int b_open )
         h->param.rc.f_qblur = 0;
     if( h->param.rc.f_complexity_blur < 0 )
         h->param.rc.f_complexity_blur = 0;
+    if( h->param.sc.i_buffer_size < 0 || h->param.sc.f_speed <= 0 )
+        h->param.sc.i_buffer_size = 0;
 
     h->param.i_sps_id &= 31;
 
-    h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_CBR );
+    if( PARAM_INTERLACED )
+        h->param.b_pic_struct = 1;
+
+    h->param.i_nal_hrd = x264_clip3( h->param.i_nal_hrd, X264_NAL_HRD_NONE, X264_NAL_HRD_FAKE_CBR );
 
     if( h->param.i_nal_hrd && !h->param.rc.i_vbv_buffer_size )
     {
         x264_log( h, X264_LOG_WARNING, "NAL HRD parameters require VBV parameters\n" );
         h->param.i_nal_hrd = X264_NAL_HRD_NONE;
     }
 
-    if( h->param.i_nal_hrd == X264_NAL_HRD_CBR &&
-       (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
+     if( (h->param.i_nal_hrd == X264_NAL_HRD_CBR || h->param.i_nal_hrd == X264_NAL_HRD_FAKE_CBR) &&
+         (h->param.rc.i_bitrate != h->param.rc.i_vbv_max_bitrate || !h->param.rc.i_vbv_max_bitrate) )
     {
         x264_log( h, X264_LOG_WARNING, "CBR HRD requires constant bitrate\n" );
-        h->param.i_nal_hrd = X264_NAL_HRD_VBR;
+        h->param.i_nal_hrd = h->param.i_nal_hrd == X264_NAL_HRD_CBR ? X264_NAL_HRD_VBR : X264_NAL_HRD_FAKE_CBR;
     }
 
+    h->param.sc.max_preset = x264_clip3( h->param.sc.max_preset, 1, SC_PRESETS );
+
     if( h->param.i_nal_hrd == X264_NAL_HRD_CBR )
         h->param.rc.b_filler = 1;
 
@@ -1580,6 +1587,7 @@ x264_t *x264_encoder_open( x264_param_t *param )
     x264_set_aspect_ratio( h, &h->param, 1 );
 
     x264_sps_init( h->sps, h->param.i_sps_id, &h->param );
+    h->param.i_profile = h->sps->i_profile_idc;
     x264_pps_init( h, h->pps, h->param.i_sps_id, &h->param, h->sps );
 
     x264_validate_levels( h, 1 );
@@ -1680,6 +1688,10 @@ x264_t *x264_encoder_open( x264_param_t *param )
     mbcmp_init( h );
     chroma_dsp_init( h );
 
+    if( h->param.sc.i_buffer_size )
+        x264_speedcontrol_new( h );
+
+
     p = buf + sprintf( buf, "using cpu capabilities:" );
     for( int i = 0; x264_cpu_names[i].flags; i++ )
     {
@@ -2674,7 +2686,7 @@ static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
     if( h->sps->i_poc_type == 0 )
     {
         h->sh.i_poc = h->fdec->i_poc;
-        if( PARAM_INTERLACED )
+        if( PARAM_INTERLACED && h->pps->b_pic_order )
         {
             h->sh.i_delta_poc_bottom = h->param.b_tff ? 1 : -1;
             h->sh.i_poc += h->sh.i_delta_poc_bottom == -1;
@@ -3853,7 +3865,7 @@ int     x264_encoder_encode( x264_t *h,
         }
 
         /* when frame threading is used, buffering period sei is written in x264_encoder_frame_end */
-        if( h->i_thread_frames == 1 && h->sps->vui.b_nal_hrd_parameters_present )
+        if( h->i_thread_frames == 1 && h->param.i_nal_hrd )
         {
             x264_hrd_fullness( h );
             x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
@@ -3946,6 +3958,10 @@ int     x264_encoder_encode( x264_t *h,
         overhead += h->out.nal[h->out.i_nal-1].i_payload + SEI_OVERHEAD;
     }
 
+    /* Init the speed control */
+    if( h->param.sc.i_buffer_size )
+        x264_speedcontrol_frame( h );
+
     if( h->fenc->b_keyframe && h->param.b_intra_refresh )
         h->i_cpb_delay_pir_offset_next = h->fenc->i_cpb_delay;
 
@@ -4061,7 +4077,7 @@ static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
     x264_emms();
 
     /* generate buffering period sei and insert it into place */
-    if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->sps->vui.b_nal_hrd_parameters_present )
+    if( h->i_thread_frames > 1 && h->fenc->b_keyframe && h->param.i_nal_hrd )
     {
         x264_hrd_fullness( h );
         x264_nal_start( h, NAL_SEI, NAL_PRIORITY_DISPOSABLE );
@@ -4085,7 +4101,6 @@ static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
 
     /* Set output picture properties */
     pic_out->i_type = h->fenc->i_type;
-
     pic_out->b_keyframe = h->fenc->b_keyframe;
     pic_out->i_pic_struct = h->fenc->i_pic_struct;
 
@@ -4162,6 +4177,9 @@ static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
         }
     }
 
+    if( h->fenc->b_keyframe && h->param.b_intra_refresh )
+        h->i_cpb_delay_pir_offset = h->fenc->i_cpb_delay;
+
     /* End bitstream, set output  */
     *pi_nal = h->out.i_nal;
     *pp_nal = h->out.nal;
@@ -4170,6 +4188,9 @@ static int x264_encoder_frame_end( x264_t *h, x264_t *thread_current,
 
     x264_noise_reduction_update( h );
 
+    if( h->param.sc.i_buffer_size )
+        x264_speedcontrol_frame_end( h );
+
     /* ---------------------- Compute/Print statistics --------------------- */
     x264_thread_sync_stat( h, h->thread[0] );
 
@@ -4622,6 +4643,7 @@ void    x264_encoder_close  ( x264_t *h )
 
     /* rc */
     x264_ratecontrol_delete( h );
+    x264_speedcontrol_delete( h );
 
     /* param */
     if( h->param.rc.psz_stat_out )