*/
#include "hb.h"
-#include "ffmpeg/avcodec.h"
+#include "hbffmpeg.h"
#include "mpeg2dec/mpeg2.h"
#define SUPPRESS_AV_LOG
#define MIN3(a,b,c) MIN(MIN(a,b),c)
#define MAX3(a,b,c) MAX(MAX(a,b),c)
+typedef struct yadif_arguments_s {
+ uint8_t **dst;
+ int parity;
+ int tff;
+ int stop;
+} yadif_arguments_t;
+
struct hb_filter_private_s
{
int pix_fmt;
uint8_t * yadif_ref[4][3];
int yadif_ref_stride[3];
+ int cpu_count;
+
+ hb_thread_t ** yadif_threads; // Threads for Yadif - one per CPU
+ hb_lock_t ** yadif_begin_lock; // Thread has work
+ hb_lock_t ** yadif_complete_lock; // Thread has completed work
+ yadif_arguments_t *yadif_arguments; // Arguments to thread for work
+
int mcdeint_mode;
int mcdeint_qp;
hb_deinterlace_close,
};
+
static void yadif_store_ref( const uint8_t ** pic,
hb_filter_private_t * pv )
{
uint8_t * ref = pv->yadif_ref[2][i];
int w = pv->width[i];
- int h = pv->height[i];
int ref_stride = pv->yadif_ref_stride[i];
int y;
}
}
-static void yadif_filter( uint8_t ** dst,
- int parity,
- int tff,
- hb_filter_private_t * pv )
+typedef struct yadif_thread_arg_s {
+ hb_filter_private_t *pv;
+ int segment;
+} yadif_thread_arg_t;
+
+/*
+ * deinterlace this segment of all three planes in a single thread.
+ */
+void yadif_filter_thread( void *thread_args_v )
{
- int i;
- for( i = 0; i < 3; i++ )
+ yadif_arguments_t *yadif_work = NULL;
+ hb_filter_private_t * pv;
+ int run = 1;
+ int plane;
+ int segment, segment_start, segment_stop;
+ yadif_thread_arg_t *thread_args = thread_args_v;
+ uint8_t **dst;
+ int parity, tff, y, w, h, ref_stride, penultimate, ultimate;
+
+
+ pv = thread_args->pv;
+ segment = thread_args->segment;
+
+ hb_log("Yadif Deinterlace thread started for segment %d", segment);
+
+ while( run )
{
- int w = pv->width[i];
- int h = pv->height[i];
- int ref_stride = pv->yadif_ref_stride[i];
+ /*
+ * Wait here until there is work to do. hb_lock() blocks until
+ * render releases it to say that there is more work to do.
+ */
+ hb_lock( pv->yadif_begin_lock[segment] );
- int y;
- for( y = 0; y < h; y++ )
+ yadif_work = &pv->yadif_arguments[segment];
+
+ if( yadif_work->stop )
+ {
+ /*
+ * No more work to do, exit this thread.
+ */
+ run = 0;
+ continue;
+ }
+
+ if( yadif_work->dst == NULL )
+ {
+ hb_error( "Thread started when no work available" );
+ hb_snooze(500);
+ continue;
+ }
+
+ /*
+ * Process all three planes, but only this segment of it.
+ */
+ for( plane = 0; plane < 3; plane++)
{
- if( (y ^ parity) & 1 )
- {
- uint8_t *prev = &pv->yadif_ref[0][i][y*ref_stride];
- uint8_t *cur = &pv->yadif_ref[1][i][y*ref_stride];
- uint8_t *next = &pv->yadif_ref[2][i][y*ref_stride];
- uint8_t *dst2 = &dst[i][y*w];
- yadif_filter_line( dst2, prev, cur, next, i, parity ^ tff, pv );
+ dst = yadif_work->dst;
+ parity = yadif_work->parity;
+ tff = yadif_work->tff;
+ w = pv->width[plane];
+ h = pv->height[plane];
+ penultimate = h -2;
+ ultimate = h - 1;
+ ref_stride = pv->yadif_ref_stride[plane];
+ segment_start = ( h / pv->cpu_count ) * segment;
+ if( segment == pv->cpu_count - 1 )
+ {
+ /*
+ * Final segment
+ */
+ segment_stop = h;
+ } else {
+ segment_stop = ( h / pv->cpu_count ) * ( segment + 1 );
}
- else
+
+ for( y = segment_start; y < segment_stop; y++ )
{
- memcpy( &dst[i][y*w],
- &pv->yadif_ref[1][i][y*ref_stride],
- w * sizeof(uint8_t) );
+ if( ( ( y ^ parity ) & 1 ) )
+ {
+ /* This is the bottom field when TFF and vice-versa.
+ It's the field that gets filtered. Because yadif
+ needs 2 lines above and below the one being filtered,
+ we need to mirror the edges. When TFF, this means
+ replacing the 2nd line with a copy of the 1st,
+ and the last with the second-to-last. */
+ if( y > 1 && y < ( h -2 ) )
+ {
+ /* This isn't the top or bottom, proceed as normal to yadif. */
+ uint8_t *prev = &pv->yadif_ref[0][plane][y*ref_stride];
+ uint8_t *cur = &pv->yadif_ref[1][plane][y*ref_stride];
+ uint8_t *next = &pv->yadif_ref[2][plane][y*ref_stride];
+ uint8_t *dst2 = &dst[plane][y*w];
+
+ yadif_filter_line( dst2,
+ prev,
+ cur,
+ next,
+ plane,
+ parity ^ tff,
+ pv );
+ }
+ else if( y == 0 )
+ {
+ /* BFF, so y0 = y1 */
+ memcpy( &dst[plane][y*w],
+ &pv->yadif_ref[1][plane][1*ref_stride],
+ w * sizeof(uint8_t) );
+ }
+ else if( y == 1 )
+ {
+ /* TFF, so y1 = y0 */
+ memcpy( &dst[plane][y*w],
+ &pv->yadif_ref[1][plane][0],
+ w * sizeof(uint8_t) );
+ }
+ else if( y == penultimate )
+ {
+ /* BFF, so penultimate y = ultimate y */
+ memcpy( &dst[plane][y*w],
+ &pv->yadif_ref[1][plane][ultimate*ref_stride],
+ w * sizeof(uint8_t) );
+ }
+ else if( y == ultimate )
+ {
+ /* TFF, so ultimate y = penultimate y */
+ memcpy( &dst[plane][y*w],
+ &pv->yadif_ref[1][plane][penultimate*ref_stride],
+ w * sizeof(uint8_t) );
+ }
+ }
+ else
+ {
+ /* Preserve this field unfiltered */
+ memcpy( &dst[plane][y*w],
+ &pv->yadif_ref[1][plane][y*ref_stride],
+ w * sizeof(uint8_t) );
+ }
}
}
+ /*
+ * Finished this segment, let everyone know.
+ */
+ hb_unlock( pv->yadif_complete_lock[segment] );
}
+ free( thread_args_v );
+}
+
+
+/*
+ * threaded yadif - each thread deinterlaces a single segment of all
+ * three planes. Where a segment is defined as the frame divided by
+ * the number of CPUs.
+ *
+ * This function blocks until the frame is deinterlaced.
+ */
+static void yadif_filter( uint8_t ** dst,
+ int parity,
+ int tff,
+ hb_filter_private_t * pv )
+{
+
+ int segment;
+
+ for( segment = 0; segment < pv->cpu_count; segment++ )
+ {
+ /*
+ * Setup the work for this plane.
+ */
+ pv->yadif_arguments[segment].parity = parity;
+ pv->yadif_arguments[segment].tff = tff;
+ pv->yadif_arguments[segment].dst = dst;
+
+ /*
+ * Let the thread for this plane know that we've setup work
+ * for it by releasing the begin lock (ensuring that the
+ * complete lock is already locked so that we block when
+ * we try to lock it again below).
+ */
+ hb_lock( pv->yadif_complete_lock[segment] );
+ hb_unlock( pv->yadif_begin_lock[segment] );
+ }
+
+ /*
+ * Wait until all three threads have completed by trying to get
+ * the complete lock that we locked earlier for each thread, which
+ * will block until that thread has completed the work on that
+ * plane.
+ */
+ for( segment = 0; segment < pv->cpu_count; segment++ )
+ {
+ hb_lock( pv->yadif_complete_lock[segment] );
+ hb_unlock( pv->yadif_complete_lock[segment] );
+ }
+
+ /*
+ * Entire frame is now deinterlaced.
+ */
}
static void mcdeint_filter( uint8_t ** dst,
pv->width[1] = pv->width[2] = width >> 1;
pv->height[1] = pv->height[2] = height >> 1;
- int buf_size = 3 * width * height / 2;
- pv->buf_out[0] = hb_buffer_init( buf_size );
- pv->buf_out[1] = hb_buffer_init( buf_size );
+ pv->buf_out[0] = hb_video_buffer_init( width, height );
+ pv->buf_out[1] = hb_video_buffer_init( width, height );
pv->buf_settings = hb_buffer_init( 0 );
pv->yadif_ready = 0;
&pv->mcdeint_qp );
}
+ pv->cpu_count = hb_get_cpu_count();
+
/* Allocate yadif specific buffers */
if( pv->yadif_mode >= 0 )
{
pv->yadif_ref[j][i] = malloc( w*h*sizeof(uint8_t) ) + 3*w;
}
}
+
+ /*
+ * Create yadif threads and locks.
+ */
+ pv->yadif_threads = malloc( sizeof( hb_thread_t* ) * pv->cpu_count );
+ pv->yadif_begin_lock = malloc( sizeof( hb_lock_t * ) * pv->cpu_count );
+ pv->yadif_complete_lock = malloc( sizeof( hb_lock_t * ) * pv->cpu_count );
+ pv->yadif_arguments = malloc( sizeof( yadif_arguments_t ) * pv->cpu_count );
+
+ for( i = 0; i < pv->cpu_count; i++ )
+ {
+ yadif_thread_arg_t *thread_args;
+
+ thread_args = malloc( sizeof( yadif_thread_arg_t ) );
+
+ if( thread_args ) {
+ thread_args->pv = pv;
+ thread_args->segment = i;
+
+ pv->yadif_begin_lock[i] = hb_lock_init();
+ pv->yadif_complete_lock[i] = hb_lock_init();
+
+ /*
+ * Important to start off with the threads locked waiting
+ * on input.
+ */
+ hb_lock( pv->yadif_begin_lock[i] );
+
+ pv->yadif_arguments[i].stop = 0;
+ pv->yadif_arguments[i].dst = NULL;
+
+ pv->yadif_threads[i] = hb_thread_init( "yadif_filter_segment",
+ yadif_filter_thread,
+ thread_args,
+ HB_NORMAL_PRIORITY );
+ } else {
+ hb_error( "Yadif could not create threads" );
+ }
+ }
}
/* Allocate mcdeint specific buffers */
avctx_enc->flags |= CODEC_FLAG_QPEL;
}
- avcodec_open(avctx_enc, enc);
+ hb_avcodec_open(avctx_enc, enc);
}
pv->mcdeint_frame = avcodec_alloc_frame();
*p = NULL;
}
}
+
+ for( i = 0; i < pv->cpu_count; i++)
+ {
+ /*
+ * Tell each yadif thread to stop, and then cleanup.
+ */
+ pv->yadif_arguments[i].stop = 1;
+ hb_unlock( pv->yadif_begin_lock[i] );
+
+ hb_thread_close( &pv->yadif_threads[i] );
+ hb_lock_close( &pv->yadif_begin_lock[i] );
+ hb_lock_close( &pv->yadif_complete_lock[i] );
+ }
+
+ /*
+ * free memory for yadif structs
+ */
+ free( pv->yadif_threads );
+ free( pv->yadif_begin_lock );
+ free( pv->yadif_complete_lock );
+ free( pv->yadif_arguments );
}
/* Cleanup mcdeint specific buffers */
{
if( pv->mcdeint_avctx_enc )
{
- avcodec_close( pv->mcdeint_avctx_enc );
+ hb_avcodec_close( pv->mcdeint_avctx_enc );
av_freep( &pv->mcdeint_avctx_enc );
}
if( pv->mcdeint_outbuf )
OSDN Git Service
