Added vp8_update_zbin_extra

vp8cx_mb_init_quantizer was being called for every mode checked
in vp8_rd_pick_inter_mode.  zbin_extra is the only value that
really needs to be recalculated.  This calculation is disabled
when using the fast quantizer for mode selection.
This gave a small performance boost (~.5% to 1%).
Note: This needs to be verified with segmentation_enabled.

Change-Id: I62716a870b3c82b4a998bdf95130ff0b02106f1e

diff --git a/vp8/encoder/block.h b/vp8/encoder/block.h
index bf94e508b5136b557ed770cf7b9777a2f9726a98..3ad40efc0284dd60d0b76f30accc0477ecb962f5 100644 (file)
--- a/vp8/encoder/block.h
+++ b/vp8/encoder/block.h
@@ -112,6 +112,7 @@ typedef struct
 
     unsigned int token_costs[BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens];
     int optimize;
+    int q_index;
 
     void (*vp8_short_fdct4x4)(short *input, short *output, int pitch);
     void (*vp8_short_fdct8x4)(short *input, short *output, int pitch);

diff --git a/vp8/encoder/encodeframe.c b/vp8/encoder/encodeframe.c
index 4c95f28d64ca54504eab688a836d51bbe856fc1d..1689b43d12ee97f894a8e309030549899eb10ca6 100644 (file)
--- a/vp8/encoder/encodeframe.c
+++ b/vp8/encoder/encodeframe.c
@@ -365,6 +365,33 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x)
     x->e_mbd.block[24].dequant = cpi->common.Y2dequant[QIndex];
     x->block[24].zrun_zbin_boost = cpi->zrun_zbin_boost_y2[QIndex];
     x->block[24].zbin_extra = (short)zbin_extra;
+
+    /* save this macroblock QIndex for vp8_update_zbin_extra() */
+    x->q_index = QIndex;
+}
+void vp8_update_zbin_extra(VP8_COMP *cpi, MACROBLOCK *x)
+{
+    int i;
+    int QIndex = x->q_index;
+    int zbin_extra;
+
+    // Y
+    zbin_extra = (cpi->common.Y1dequant[QIndex][1] * (cpi->zbin_over_quant + cpi->zbin_mode_boost)) >> 7;
+    for (i = 0; i < 16; i++)
+    {
+        x->block[i].zbin_extra = (short)zbin_extra;
+    }
+
+    // UV
+    zbin_extra = (cpi->common.UVdequant[QIndex][1] * (cpi->zbin_over_quant + cpi->zbin_mode_boost)) >> 7;
+    for (i = 16; i < 24; i++)
+    {
+        x->block[i].zbin_extra = (short)zbin_extra;
+    }
+
+    // Y2
+    zbin_extra = (cpi->common.Y2dequant[QIndex][1] * ((cpi->zbin_over_quant / 2) + cpi->zbin_mode_boost)) >> 7;
+    x->block[24].zbin_extra = (short)zbin_extra;
 }
 
 void vp8cx_frame_init_quantizer(VP8_COMP *cpi)
@@ -1261,10 +1288,17 @@ int vp8cx_encode_inter_macroblock
 
     if (cpi->sf.RD)
     {
+        int zbin_mode_boost_enabled = cpi->zbin_mode_boost_enabled;
+
         /* Are we using the fast quantizer for the mode selection? */
         if(cpi->sf.use_fastquant_for_pick)
+        {
             cpi->mb.quantize_b      = QUANTIZE_INVOKE(&cpi->rtcd.quantize, fastquantb);
 
+            /* the fast quantizer does not use zbin_extra, so
+             * do not recalculate */
+            cpi->zbin_mode_boost_enabled = 0;
+        }
         inter_error = vp8_rd_pick_inter_mode(cpi, x, recon_yoffset, recon_uvoffset, &rate, &distortion, &intra_error);
 
         /* switch back to the regular quantizer for the encode */
@@ -1273,6 +1307,9 @@ int vp8cx_encode_inter_macroblock
             cpi->mb.quantize_b    = QUANTIZE_INVOKE(&cpi->rtcd.quantize, quantb);
         }
 
+        /* restore cpi->zbin_mode_boost_enabled */
+        cpi->zbin_mode_boost_enabled = zbin_mode_boost_enabled;
+
     }
     else
 #endif
@@ -1289,7 +1326,7 @@ int vp8cx_encode_inter_macroblock
 #endif
 
     // MB level adjutment to quantizer setup
-    if (xd->segmentation_enabled || cpi->zbin_mode_boost_enabled)
+    if (xd->segmentation_enabled)
     {
         // If cyclic update enabled
         if (cpi->cyclic_refresh_mode_enabled)
@@ -1299,9 +1336,14 @@ int vp8cx_encode_inter_macroblock
                 ((xd->mode_info_context->mbmi.ref_frame != LAST_FRAME) || (xd->mode_info_context->mbmi.mode != ZEROMV)))
             {
                 xd->mode_info_context->mbmi.segment_id = 0;
+
+                /* segment_id changed, so update */
+                vp8cx_mb_init_quantizer(cpi, x);
             }
         }
+    }
 
+    {
         // Experimental code. Special case for gf and arf zeromv modes. Increase zbin size to supress noise
         if (cpi->zbin_mode_boost_enabled)
         {
@@ -1325,7 +1367,7 @@ int vp8cx_encode_inter_macroblock
         else
             cpi->zbin_mode_boost = 0;
 
-        vp8cx_mb_init_quantizer(cpi,  x);
+        vp8_update_zbin_extra(cpi, x);
     }
 
     cpi->count_mb_ref_frame_usage[xd->mode_info_context->mbmi.ref_frame] ++;

diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index b2a3e117f832a6421fe5e1aab2e165ad9cc573cc..3b898f1b273e533242246a3c42bfe589d5e55e6a 100644 (file)
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -43,7 +43,9 @@
 #endif
 
 
-void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x);
+extern void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x);
+extern void vp8_update_zbin_extra(VP8_COMP *cpi, MACROBLOCK *x);
+
 
 #define RDCOST(RM,DM,R,D) ( ((128+(R)*(RM)) >> 8) + (DM)*(D) )
 
@@ -1990,7 +1992,7 @@ int vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
                     cpi->zbin_mode_boost = MV_ZBIN_BOOST;
             }
 
-            vp8cx_mb_init_quantizer(cpi, x);
+            vp8_update_zbin_extra(cpi, x);
         }
 
         switch (this_mode)