Merge "Reduced the size of Y1Dequant and friends to [128][2]"

3 files changed, 46 insertions, 12 deletions

diff --git a/vp8/encoder/encodeframe.c b/vp8/encoder/encodeframe.c
index 6a9ba291d..b5c5c7445 100644
--- a/vp8/encoder/encodeframe.c
+++ b/vp8/encoder/encodeframe.c
@@ -1120,7 +1120,7 @@ int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t,
         vp8_inverse_transform_mby(xd, IF_RTCD(&cpi->common.rtcd));
 
     DEQUANT_INVOKE (&cpi->common.rtcd.dequant, idct_add_uv_block)
-                    (xd->qcoeff+16*16, xd->block[16].dequant,
+                    (xd->qcoeff+16*16, xd->dequant_uv,
                      xd->dst.u_buffer, xd->dst.v_buffer,
                      xd->dst.uv_stride, xd->eobs+16);
     return rate;
@@ -1305,7 +1305,7 @@ int vp8cx_encode_inter_macroblock
             vp8_inverse_transform_mby(xd, IF_RTCD(&cpi->common.rtcd));
 
         DEQUANT_INVOKE (&cpi->common.rtcd.dequant, idct_add_uv_block)
-                        (xd->qcoeff+16*16, xd->block[16].dequant,
+                        (xd->qcoeff+16*16, xd->dequant_uv,
                          xd->dst.u_buffer, xd->dst.v_buffer,
                          xd->dst.uv_stride, xd->eobs+16);
     }
diff --git a/vp8/encoder/ethreading.c b/vp8/encoder/ethreading.c
index 69655989d..24339a5e0 100644
--- a/vp8/encoder/ethreading.c
+++ b/vp8/encoder/ethreading.c
@@ -384,10 +384,22 @@ static void setup_mbby_copy(MACROBLOCK *mbdst, MACROBLOCK *mbsrc)
         zd->mb_segement_abs_delta      = xd->mb_segement_abs_delta;
         vpx_memcpy(zd->segment_feature_data, xd->segment_feature_data, sizeof(xd->segment_feature_data));
 
-        for (i = 0; i < 25; i++)
-        {
-            zd->block[i].dequant = xd->block[i].dequant;
-        }
+        vpx_memcpy(zd->dequant_y1_dc, xd->dequant_y1_dc, sizeof(xd->dequant_y1_dc));
+        vpx_memcpy(zd->dequant_y1, xd->dequant_y1, sizeof(xd->dequant_y1));
+        vpx_memcpy(zd->dequant_y2, xd->dequant_y2, sizeof(xd->dequant_y2));
+        vpx_memcpy(zd->dequant_uv, xd->dequant_uv, sizeof(xd->dequant_uv));
+
+#if 1
+        /*TODO:  Remove dequant from BLOCKD.  This is a temporary solution until
+         * the quantizer code uses a passed in pointer to the dequant constants.
+         * This will also require modifications to the x86 and neon assembly.
+         * */
+        for (i = 0; i < 16; i++)
+            zd->block[i].dequant = zd->dequant_y1;
+        for (i = 16; i < 24; i++)
+            zd->block[i].dequant = zd->dequant_uv;
+        zd->block[24].dequant = zd->dequant_y2;
+#endif
     }
 }
 
diff --git a/vp8/encoder/quantize.c b/vp8/encoder/quantize.c
index ce04212e6..d2aa7fe72 100644
--- a/vp8/encoder/quantize.c
+++ b/vp8/encoder/quantize.c
@@ -504,7 +504,6 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
             cpi->Y1quant_shift[Q][i] = cpi->Y1quant_shift[Q][1];
             cpi->Y1zbin[Q][i] = cpi->Y1zbin[Q][1];
             cpi->Y1round[Q][i] = cpi->Y1round[Q][1];
-            cpi->common.Y1dequant[Q][i] = cpi->common.Y1dequant[Q][1];
             cpi->zrun_zbin_boost_y1[Q][i] = (cpi->common.Y1dequant[Q][1] *
                                              zbin_boost[i]) >> 7;
 
@@ -513,7 +512,6 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
             cpi->Y2quant_shift[Q][i] = cpi->Y2quant_shift[Q][1];
             cpi->Y2zbin[Q][i] = cpi->Y2zbin[Q][1];
             cpi->Y2round[Q][i] = cpi->Y2round[Q][1];
-            cpi->common.Y2dequant[Q][i] = cpi->common.Y2dequant[Q][1];
             cpi->zrun_zbin_boost_y2[Q][i] = (cpi->common.Y2dequant[Q][1] *
                                              zbin_boost[i]) >> 7;
 
@@ -522,7 +520,6 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
             cpi->UVquant_shift[Q][i] = cpi->UVquant_shift[Q][1];
             cpi->UVzbin[Q][i] = cpi->UVzbin[Q][1];
             cpi->UVround[Q][i] = cpi->UVround[Q][1];
-            cpi->common.UVdequant[Q][i] = cpi->common.UVdequant[Q][1];
             cpi->zrun_zbin_boost_uv[Q][i] = (cpi->common.UVdequant[Q][1] *
                                              zbin_boost[i]) >> 7;
         }
@@ -641,6 +638,31 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip)
      */
     if (!ok_to_skip || QIndex != x->q_index)
     {
+
+        xd->dequant_y1_dc[0] = 1;
+        xd->dequant_y1[0] = cpi->common.Y1dequant[QIndex][0];
+        xd->dequant_y2[0] = cpi->common.Y2dequant[QIndex][0];
+        xd->dequant_uv[0] = cpi->common.UVdequant[QIndex][0];
+
+        for (i = 1; i < 16; i++)
+        {
+            xd->dequant_y1_dc[i] =
+            xd->dequant_y1[i] = cpi->common.Y1dequant[QIndex][1];
+            xd->dequant_y2[i] = cpi->common.Y2dequant[QIndex][1];
+            xd->dequant_uv[i] = cpi->common.UVdequant[QIndex][1];
+        }
+#if 1
+        /*TODO:  Remove dequant from BLOCKD.  This is a temporary solution until
+         * the quantizer code uses a passed in pointer to the dequant constants.
+         * This will also require modifications to the x86 and neon assembly.
+         * */
+        for (i = 0; i < 16; i++)
+            x->e_mbd.block[i].dequant = xd->dequant_y1; //cpi->common.Y1dequant[QIndex];
+        for (i = 16; i < 24; i++)
+            x->e_mbd.block[i].dequant = xd->dequant_uv; //cpi->common.UVdequant[QIndex];
+        x->e_mbd.block[24].dequant = xd->dequant_y2; //cpi->common.Y2dequant[QIndex];
+#endif
+
         // Y
         zbin_extra = ZBIN_EXTRA_Y;
 
@@ -651,7 +673,6 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip)
             x->block[i].quant_shift = cpi->Y1quant_shift[QIndex];
             x->block[i].zbin = cpi->Y1zbin[QIndex];
             x->block[i].round = cpi->Y1round[QIndex];
-            x->e_mbd.block[i].dequant = cpi->common.Y1dequant[QIndex];
             x->block[i].zrun_zbin_boost = cpi->zrun_zbin_boost_y1[QIndex];
             x->block[i].zbin_extra = (short)zbin_extra;
         }
@@ -666,7 +687,6 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip)
             x->block[i].quant_shift = cpi->UVquant_shift[QIndex];
             x->block[i].zbin = cpi->UVzbin[QIndex];
             x->block[i].round = cpi->UVround[QIndex];
-            x->e_mbd.block[i].dequant = cpi->common.UVdequant[QIndex];
             x->block[i].zrun_zbin_boost = cpi->zrun_zbin_boost_uv[QIndex];
             x->block[i].zbin_extra = (short)zbin_extra;
         }
@@ -679,7 +699,6 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip)
         x->block[24].quant_shift = cpi->Y2quant_shift[QIndex];
         x->block[24].zbin = cpi->Y2zbin[QIndex];
         x->block[24].round = cpi->Y2round[QIndex];
-        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;
 
@@ -689,6 +708,9 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x, int ok_to_skip)
         cpi->last_zbin_over_quant = cpi->zbin_over_quant;
         cpi->last_zbin_mode_boost = cpi->zbin_mode_boost;
         x->last_act_zbin_adj = x->act_zbin_adj;
+
+
+
     }
     else if(cpi->last_zbin_over_quant != cpi->zbin_over_quant
             || cpi->last_zbin_mode_boost != cpi->zbin_mode_boost