vp9: Enable row multithreading for SVC in real-time mode.

Enable row-mt for SVC for real-time mode, speed >=5.

Add the controls to the sample encoders, but keep it off for now.
Add the control and enable it for the 1 pass CBR unittests.

For speed 7, 3 layer SVC, 2 threads, row-mt enabled gives about ~5% speedup.

Change-Id: Ie8e77323c17263e3e7a7b9858aec12a3a93ec0c1

diff --git a/examples/vp9_spatial_svc_encoder.c b/examples/vp9_spatial_svc_encoder.c
index 0e40938..1f5078a 100644 (file)
--- a/examples/vp9_spatial_svc_encoder.c
+++ b/examples/vp9_spatial_svc_encoder.c
@@ -697,8 +697,10 @@ int main(int argc, const char **argv) {
 
   if (svc_ctx.speed != -1)
     vpx_codec_control(&codec, VP8E_SET_CPUUSED, svc_ctx.speed);
-  if (svc_ctx.threads)
+  if (svc_ctx.threads) {
     vpx_codec_control(&codec, VP9E_SET_TILE_COLUMNS, (svc_ctx.threads >> 1));
+    vpx_codec_control(&codec, VP9E_SET_ROW_MT, 0);
+  }
   if (svc_ctx.speed >= 5 && svc_ctx.aqmode == 1)
     vpx_codec_control(&codec, VP9E_SET_AQ_MODE, 3);
   if (svc_ctx.speed >= 5)

diff --git a/test/datarate_test.cc b/test/datarate_test.cc
index e66761f..e339030 100644 (file)
--- a/test/datarate_test.cc
+++ b/test/datarate_test.cc
@@ -1096,6 +1096,7 @@ class DatarateOnePassCbrSvc
       encoder->Control(VP9E_SET_TILE_COLUMNS, 0);
       encoder->Control(VP8E_SET_MAX_INTRA_BITRATE_PCT, 300);
       encoder->Control(VP9E_SET_TILE_COLUMNS, (cfg_.g_threads >> 1));
+      encoder->Control(VP9E_SET_ROW_MT, 1);
       encoder->Control(VP8E_SET_STATIC_THRESHOLD, 1);
     }
     const vpx_rational_t tb = video->timebase();

diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 03bf385..9ff97ea 100644 (file)
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -5201,8 +5201,7 @@ void vp9_set_row_mt(VP9_COMP *cpi) {
 
   // In realtime mode, enable row based multi-threading for all the speed levels
   // where non-rd path is used.
-  if (cpi->oxcf.mode == REALTIME && cpi->oxcf.speed >= 5 && cpi->oxcf.row_mt &&
-      !cpi->use_svc) {
+  if (cpi->oxcf.mode == REALTIME && cpi->oxcf.speed >= 5 && cpi->oxcf.row_mt) {
     cpi->row_mt = 1;
   }
 }

diff --git a/vp9/encoder/vp9_ethread.c b/vp9/encoder/vp9_ethread.c
index b618b42..2b2a556 100644 (file)
--- a/vp9/encoder/vp9_ethread.c
+++ b/vp9/encoder/vp9_ethread.c
@@ -77,8 +77,9 @@ static void create_enc_workers(VP9_COMP *cpi, int num_workers) {
     int allocated_workers = num_workers;
 
     // While using SVC, we need to allocate threads according to the highest
-    // resolution.
-    if (cpi->use_svc) {
+    // resolution. When row based multithreading is enabled, it is OK to
+    // allocate more threads than the number of max tile columns.
+    if (cpi->use_svc && !cpi->row_mt) {
       int max_tile_cols = get_max_tile_cols(cpi);
       allocated_workers = VPXMIN(cpi->oxcf.max_threads, max_tile_cols);
     }
@@ -615,7 +616,6 @@ void vp9_encode_tiles_row_mt(VP9_COMP *cpi) {
   for (i = 0; i < num_workers; i++) {
     EncWorkerData *thread_data;
     thread_data = &cpi->tile_thr_data[i];
-
     // Before encoding a frame, copy the thread data from cpi.
     if (thread_data->td != &cpi->td) {
       thread_data->td->mb = cpi->td.mb;