For 1 pass: always use the normative filter in vp9_scale_if_required()

The normative (convolve8) filter is optimized/faster than
the nonnormative one. Pass usage of scaler (normative/nonomorative)
to vp9_scale_if_required(), and always use normative one for 1 pass.

Change-Id: I2b71d9ff18b3c7499b058d1325a9554de993dd52

4 files changed, 16 insertions, 14 deletions

diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 7a4cf3cc6..39b4c294c 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -3200,7 +3200,8 @@ static void encode_without_recode_loop(VP9_COMP *cpi,
 
   cpi->Source = vp9_scale_if_required(cm,
                                       cpi->un_scaled_source,
-                                      &cpi->scaled_source);
+                                      &cpi->scaled_source,
+                                      (cpi->oxcf.pass == 0));
 
   // Avoid scaling last_source unless its needed.
   // Last source is currently only used for screen-content mode,
@@ -3210,7 +3211,8 @@ static void encode_without_recode_loop(VP9_COMP *cpi,
       cpi->sf.partition_search_type == SOURCE_VAR_BASED_PARTITION))
     cpi->Last_Source = vp9_scale_if_required(cm,
                                              cpi->unscaled_last_source,
-                                             &cpi->scaled_last_source);
+                                             &cpi->scaled_last_source,
+                                             (cpi->oxcf.pass == 0));
 
   if (cpi->oxcf.pass == 0 &&
       cpi->oxcf.rc_mode == VPX_CBR &&
@@ -3342,11 +3344,13 @@ static void encode_with_recode_loop(VP9_COMP *cpi,
     }
 
     cpi->Source = vp9_scale_if_required(cm, cpi->un_scaled_source,
-                                      &cpi->scaled_source);
+                                      &cpi->scaled_source,
+                                      (cpi->oxcf.pass == 0));
 
     if (cpi->unscaled_last_source != NULL)
       cpi->Last_Source = vp9_scale_if_required(cm, cpi->unscaled_last_source,
-                                               &cpi->scaled_last_source);
+                                               &cpi->scaled_last_source,
+                                               (cpi->oxcf.pass == 0));
 
     if (frame_is_intra_only(cm) == 0) {
       if (loop_count > 0) {
@@ -3597,20 +3601,17 @@ static void set_ext_overrides(VP9_COMP *cpi) {
 
 YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
                                           YV12_BUFFER_CONFIG *unscaled,
-                                          YV12_BUFFER_CONFIG *scaled) {
+                                          YV12_BUFFER_CONFIG *scaled,
+                                          int use_normative_scaler) {
   if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
       cm->mi_rows * MI_SIZE != unscaled->y_height) {
 #if CONFIG_VP9_HIGHBITDEPTH
-    if (unscaled->y_width == (scaled->y_width << 1) &&
-        unscaled->y_height == (scaled->y_height << 1))
+    if (use_normative_scaler)
       scale_and_extend_frame(unscaled, scaled, (int)cm->bit_depth);
     else
       scale_and_extend_frame_nonnormative(unscaled, scaled, (int)cm->bit_depth);
 #else
-    // Use the faster normative (convolve8) scaling filter: for now only for
-    // scaling factor of 2.
-    if (unscaled->y_width == (scaled->y_width << 1) &&
-        unscaled->y_height == (scaled->y_height << 1))
+    if (use_normative_scaler)
       scale_and_extend_frame(unscaled, scaled);
     else
       scale_and_extend_frame_nonnormative(unscaled, scaled);
diff --git a/vp9/encoder/vp9_encoder.h b/vp9/encoder/vp9_encoder.h
index 7c2101850..c3a983956 100644
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -613,7 +613,8 @@ void vp9_set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv);
 
 YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
                                           YV12_BUFFER_CONFIG *unscaled,
-                                          YV12_BUFFER_CONFIG *scaled);
+                                          YV12_BUFFER_CONFIG *scaled,
+                                          int use_normative_scaler);
 
 void vp9_apply_encoding_flags(VP9_COMP *cpi, vpx_enc_frame_flags_t flags);
 
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index 51cc40302..a6b5ebb01 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -597,7 +597,7 @@ void vp9_first_pass(VP9_COMP *cpi, const struct lookahead_entry *source) {
                  (cpi->ref_frame_flags & VP9_GOLD_FLAG) ? GOLDEN_FRAME : NONE);
 
     cpi->Source = vp9_scale_if_required(cm, cpi->un_scaled_source,
-                                        &cpi->scaled_source);
+                                        &cpi->scaled_source, 0);
   }
 
   vp9_setup_block_planes(&x->e_mbd, cm->subsampling_x, cm->subsampling_y);
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index 6ef3b49d4..16f9c8573 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -721,7 +721,7 @@ void vp9_temporal_filter(VP9_COMP *cpi, int distance) {
                                "Failed to reallocate alt_ref_buffer");
           }
           frames[frame] = vp9_scale_if_required(
-              cm, frames[frame], &cpi->svc.scaled_frames[frame_used]);
+              cm, frames[frame], &cpi->svc.scaled_frames[frame_used], 0);
           ++frame_used;
         }
       }