Allow macroblock_plane to have its own rounding buffer

Add 8 bytes buffer to macroblock_plane to support rounding factor.

Change-Id: I3751689e4449c0caea28d3acf6cd17d7f39508ed

diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index c879a19..37a4605 100644 (file)
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -34,8 +34,8 @@ struct macroblock_plane {
   struct buf_2d src;
 
   // Quantizer setings
+  DECLARE_ALIGNED(16, int16_t, round_fp[8]);
   int16_t *quant_fp;
-  int16_t *round_fp;
   int16_t *quant;
   int16_t *quant_shift;
   int16_t *zbin;

diff --git a/vp9/encoder/vp9_quantize.c b/vp9/encoder/vp9_quantize.c
index 75f3a8b..26d1434 100644 (file)
--- a/vp9/encoder/vp9_quantize.c
+++ b/vp9/encoder/vp9_quantize.c
@@ -288,12 +288,12 @@ void vp9_init_plane_quantizers(VP9_COMP *cpi, MACROBLOCK *x) {
   // Y
   x->plane[0].quant = quants->y_quant[qindex];
   x->plane[0].quant_fp = quants->y_quant_fp[qindex];
-  x->plane[0].round_fp = quants->y_round_fp[qindex];
+  memcpy(x->plane[0].round_fp, quants->y_round_fp[qindex],
+         8 * sizeof(*(x->plane[0].round_fp)));
   x->plane[0].quant_shift = quants->y_quant_shift[qindex];
   x->plane[0].zbin = quants->y_zbin[qindex];
   x->plane[0].round = quants->y_round[qindex];
   xd->plane[0].dequant = cpi->y_dequant[qindex];
-
   x->plane[0].quant_thred[0] = x->plane[0].zbin[0] * x->plane[0].zbin[0];
   x->plane[0].quant_thred[1] = x->plane[0].zbin[1] * x->plane[0].zbin[1];
 
@@ -301,12 +301,12 @@ void vp9_init_plane_quantizers(VP9_COMP *cpi, MACROBLOCK *x) {
   for (i = 1; i < 3; i++) {
     x->plane[i].quant = quants->uv_quant[qindex];
     x->plane[i].quant_fp = quants->uv_quant_fp[qindex];
-    x->plane[i].round_fp = quants->uv_round_fp[qindex];
+    memcpy(x->plane[i].round_fp, quants->uv_round_fp[qindex],
+           8 * sizeof(*(x->plane[i].round_fp)));
     x->plane[i].quant_shift = quants->uv_quant_shift[qindex];
     x->plane[i].zbin = quants->uv_zbin[qindex];
     x->plane[i].round = quants->uv_round[qindex];
     xd->plane[i].dequant = cpi->uv_dequant[qindex];
-
     x->plane[i].quant_thred[0] = x->plane[i].zbin[0] * x->plane[i].zbin[0];
     x->plane[i].quant_thred[1] = x->plane[i].zbin[1] * x->plane[i].zbin[1];
   }