Multithreaded encoder, late sync loopfilter

Second shot at this...

Sync with loopfilter thread as late as possible, usually just at the
beginning of next frame encoding. This returns control to application
faster and allows a better multicore scaling.

When PSNR packets are generated the final filtered frame is needed
imediatly so we cannot delay the sync. Same has to be done when
internal frame is previewed.

Change-Id: I64e110c8b224dd967faefffd9c93dd8dbad4a5b5

vp8/encoder/ethreading.c

@@ -24,9 +24,9 @@ extern void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip
 extern void vp8_build_block_offsets(MACROBLOCK *x);
 extern void vp8_setup_block_ptrs(MACROBLOCK *x);
 
-extern void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm);
+extern void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm);
 
-static THREAD_FUNCTION loopfilter_thread(void *p_data)
+static THREAD_FUNCTION thread_loopfilter(void *p_data)
 {
     VP8_COMP *cpi = (VP8_COMP *)(((LPFTHREAD_DATA *)p_data)->ptr1);
     VP8_COMMON *cm = &cpi->common;
@@ -41,7 +41,7 @@ static THREAD_FUNCTION loopfilter_thread(void *p_data)
             if (cpi->b_multi_threaded == 0) // we're shutting down
                 break;
 
-            loopfilter_frame(cpi, cm);
+            vp8_loopfilter_frame(cpi, cm);
 
             sem_post(&cpi->h_event_end_lpf);
         }
@@ -468,6 +468,7 @@ void vp8cx_create_encoder_threads(VP8_COMP *cpi)
 
     cpi->b_multi_threaded = 0;
     cpi->encoding_thread_count = 0;
+    cpi->b_lpf_running = 0;
 
     if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
     {
@@ -526,7 +527,7 @@ void vp8cx_create_encoder_threads(VP8_COMP *cpi)
             sem_init(&cpi->h_event_end_lpf, 0, 0);
 
             lpfthd->ptr1 = (void *)cpi;
-            pthread_create(&cpi->h_filter_thread, 0, loopfilter_thread, lpfthd);
+            pthread_create(&cpi->h_filter_thread, 0, thread_loopfilter, lpfthd);
         }
     }
 

vp8/encoder/onyx_if.c

@@ -3086,7 +3086,7 @@ void update_reference_frames(VP8_COMMON *cm)
     }
 }
 
-void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
+void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
 {
     if (cm->no_lpf)
     {
@@ -3716,6 +3716,15 @@ static void encode_frame_to_data_rate
             vp8_setup_key_frame(cpi);
         }
 
+#if CONFIG_MULTITHREAD
+        /*  wait for the last picture loopfilter thread done */
+        if (cpi->b_lpf_running)
+        {
+            sem_wait(&cpi->h_event_end_lpf);
+            cpi->b_lpf_running = 0;
+        }
+#endif
+
         // transform / motion compensation build reconstruction frame
         vp8_encode_frame(cpi);
 
@@ -4074,11 +4083,12 @@ static void encode_frame_to_data_rate
     if (cpi->b_multi_threaded)
     {
         sem_post(&cpi->h_event_start_lpf); /* start loopfilter in separate thread */
+        cpi->b_lpf_running = 1;
     }
     else
 #endif
     {
-        loopfilter_frame(cpi, cm);
+        vp8_loopfilter_frame(cpi, cm);
     }
 
     update_reference_frames(cm);
@@ -4098,10 +4108,11 @@ static void encode_frame_to_data_rate
     vp8_pack_bitstream(cpi, dest, dest_end, size);
 
 #if CONFIG_MULTITHREAD
-    /* wait for loopfilter thread done */
-    if (cpi->b_multi_threaded)
+    /* if PSNR packets are generated we have to wait for the lpf */
+    if (cpi->b_lpf_running && cpi->b_calculate_psnr)
     {
         sem_wait(&cpi->h_event_end_lpf);
+        cpi->b_lpf_running = 0;
     }
 #endif
 
@@ -5107,6 +5118,15 @@ int vp8_get_preview_raw_frame(VP8_COMP *cpi, YV12_BUFFER_CONFIG *dest, vp8_ppfla
     else
     {
         int ret;
+
+#if CONFIG_MULTITHREAD
+        if(cpi->b_lpf_running)
+        {
+            sem_wait(&cpi->h_event_end_lpf);
+            cpi->b_lpf_running = 0;
+        }
+#endif
+
 #if CONFIG_POSTPROC
         ret = vp8_post_proc_frame(&cpi->common, dest, flags);
 #else

vp8/encoder/onyx_int.h

@@ -526,6 +526,7 @@ typedef struct VP8_COMP
     int mt_sync_range;
     int b_multi_threaded;
     int encoding_thread_count;
+    int b_lpf_running;
 
     pthread_t *h_encoding_thread;
     pthread_t h_filter_thread;