Make vp9_scale_and_extend_frame_ssse3 work for hbd when bitdepth = 8.

Only works for bitdepth = 8 when compiled with high bitdepth flag.
4x speed ups for handling 1:2 down/upsampling.

Validated manually for:
1) Dynamic resize for a single layer encoding
2) SVC encoding with 3 spatial layers
Results are bitexact with the patch and the speed gain (~4x) in the
scaling was verified.

BUG=webm:1371

Change-Id: I1bdb5f4d4bd0df67763fc271b6aa355e60f34712

diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 4b4c119..d3659f7 100644 (file)
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -228,11 +228,8 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
 #
 # frame based scale
 #
-if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
-} else {
-  add_proto qw/void vp9_scale_and_extend_frame/, "const struct yv12_buffer_config *src, struct yv12_buffer_config *dst";
-  specialize qw/vp9_scale_and_extend_frame ssse3/;
-}
+add_proto qw/void vp9_scale_and_extend_frame/, "const struct yv12_buffer_config *src, struct yv12_buffer_config *dst";
+specialize qw/vp9_scale_and_extend_frame ssse3/;
 
 }
 # end encoder functions

diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index aa8a27c..5d64b0a 100644 (file)
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -2406,7 +2406,8 @@ static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
 
   vpx_extend_frame_borders(dst);
 }
-#else
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 void vp9_scale_and_extend_frame_c(const YV12_BUFFER_CONFIG *src,
                                   YV12_BUFFER_CONFIG *dst) {
   const int src_w = src->y_crop_width;
@@ -2444,7 +2445,6 @@ void vp9_scale_and_extend_frame_c(const YV12_BUFFER_CONFIG *src,
 
   vpx_extend_frame_borders(dst);
 }
-#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 static int scale_down(VP9_COMP *cpi, int q) {
   RATE_CONTROL *const rc = &cpi->rc;
@@ -3661,8 +3661,13 @@ YV12_BUFFER_CONFIG *vp9_svc_twostage_scale(VP9_COMMON *cm,
   if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
       cm->mi_rows * MI_SIZE != unscaled->y_height) {
 #if CONFIG_VP9_HIGHBITDEPTH
-    scale_and_extend_frame(unscaled, scaled_temp, (int)cm->bit_depth);
-    scale_and_extend_frame(scaled_temp, scaled, (int)cm->bit_depth);
+    if (cm->bit_depth == VPX_BITS_8) {
+      vp9_scale_and_extend_frame(unscaled, scaled_temp);
+      vp9_scale_and_extend_frame(scaled_temp, scaled);
+    } else {
+      scale_and_extend_frame(unscaled, scaled_temp, (int)cm->bit_depth);
+      scale_and_extend_frame(scaled_temp, scaled, (int)cm->bit_depth);
+    }
 #else
     vp9_scale_and_extend_frame(unscaled, scaled_temp);
     vp9_scale_and_extend_frame(scaled_temp, scaled);
@@ -3682,7 +3687,10 @@ YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
 #if CONFIG_VP9_HIGHBITDEPTH
     if (use_normative_scaler && unscaled->y_width <= (scaled->y_width << 1) &&
         unscaled->y_height <= (scaled->y_height << 1))
-      scale_and_extend_frame(unscaled, scaled, (int)cm->bit_depth);
+      if (cm->bit_depth == VPX_BITS_8)
+        vp9_scale_and_extend_frame(unscaled, scaled);
+      else
+        scale_and_extend_frame(unscaled, scaled, (int)cm->bit_depth);
     else
       scale_and_extend_frame_nonnormative(unscaled, scaled, (int)cm->bit_depth);
 #else

diff --git a/vp9/vp9cx.mk b/vp9/vp9cx.mk
index a54e99e..3accaf5 100644 (file)
--- a/vp9/vp9cx.mk
+++ b/vp9/vp9cx.mk
@@ -119,9 +119,7 @@ endif
 
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_dct_intrin_sse2.c
 VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_dct_ssse3.c
-ifneq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
 VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_frame_scale_ssse3.c
-endif
 
 ifeq ($(CONFIG_VP9_TEMPORAL_DENOISING),yes)
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_denoiser_sse2.c