Removing qcoeff buffers from the decoder.

We only need qcoeff buffers in the encoder. Reducing TileWorkerData struct
and VP9Decompressor struct sizes by 24K.

Change-Id: Id148868461f7ffa3d3dd634b371503ae9c57e207

7 files changed, 18 insertions, 14 deletions

diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index 444597067..71f7e7a52 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -69,6 +69,7 @@ typedef struct {
 
 struct macroblock_plane {
   DECLARE_ALIGNED(16, int16_t, src_diff[64 * 64]);
+  int16_t *qcoeff;
   int16_t *coeff;
   struct buf_2d src;
 
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 18cfddd55..38c753849 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -439,14 +439,14 @@ static void update_state(VP9_COMP *cpi, PICK_MODE_CONTEXT *ctx,
   max_plane = is_inter_block(mbmi) ? MAX_MB_PLANE : 1;
   for (i = 0; i < max_plane; ++i) {
     p[i].coeff = ctx->coeff_pbuf[i][1];
-    pd[i].qcoeff = ctx->qcoeff_pbuf[i][1];
+    p[i].qcoeff = ctx->qcoeff_pbuf[i][1];
     pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1];
     pd[i].eobs = ctx->eobs_pbuf[i][1];
   }
 
   for (i = max_plane; i < MAX_MB_PLANE; ++i) {
     p[i].coeff = ctx->coeff_pbuf[i][2];
-    pd[i].qcoeff = ctx->qcoeff_pbuf[i][2];
+    p[i].qcoeff = ctx->qcoeff_pbuf[i][2];
     pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][2];
     pd[i].eobs = ctx->eobs_pbuf[i][2];
   }
@@ -675,7 +675,7 @@ static void pick_sb_modes(VP9_COMP *cpi, const TileInfo *const tile,
 
   for (i = 0; i < MAX_MB_PLANE; ++i) {
     p[i].coeff = ctx->coeff_pbuf[i][0];
-    pd[i].qcoeff = ctx->qcoeff_pbuf[i][0];
+    p[i].qcoeff = ctx->qcoeff_pbuf[i][0];
     pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][0];
     pd[i].eobs = ctx->eobs_pbuf[i][0];
   }
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index bd9678afb..83f87b066 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -136,6 +136,7 @@ static void optimize_b(MACROBLOCK *mb,
                        ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
                        TX_SIZE tx_size) {
   MACROBLOCKD *const xd = &mb->e_mbd;
+  struct macroblock_plane *p = &mb->plane[plane];
   struct macroblockd_plane *pd = &xd->plane[plane];
   const int ref = is_inter_block(&xd->mi_8x8[0]->mbmi);
   vp9_token_state tokens[1025][2];
@@ -163,7 +164,7 @@ static void optimize_b(MACROBLOCK *mb,
 
   assert((!type && !plane) || (type && plane));
   dqcoeff_ptr = BLOCK_OFFSET(pd->dqcoeff, block);
-  qcoeff_ptr = BLOCK_OFFSET(pd->qcoeff, block);
+  qcoeff_ptr = BLOCK_OFFSET(p->qcoeff, block);
   assert(eob <= default_eob);
 
   /* Now set up a Viterbi trellis to evaluate alternative roundings. */
@@ -368,7 +369,7 @@ void vp9_xform_quant(int plane, int block, BLOCK_SIZE plane_bsize,
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
   int16_t *coeff = BLOCK_OFFSET(p->coeff, block);
-  int16_t *qcoeff = BLOCK_OFFSET(pd->qcoeff, block);
+  int16_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   int16_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   const scan_order *so;
   uint16_t *eob = &pd->eobs[block];
@@ -530,7 +531,7 @@ void vp9_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
   struct macroblock_plane *const p = &x->plane[plane];
   struct macroblockd_plane *const pd = &xd->plane[plane];
   int16_t *coeff = BLOCK_OFFSET(p->coeff, block);
-  int16_t *qcoeff = BLOCK_OFFSET(pd->qcoeff, block);
+  int16_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
   int16_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
   const scan_order *so;
   TX_TYPE tx_type;
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index 924f9f324..930e2f82b 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -530,7 +530,7 @@ void vp9_first_pass(VP9_COMP *cpi) {
 
   for (i = 0; i < MAX_MB_PLANE; ++i) {
     p[i].coeff = ctx->coeff_pbuf[i][1];
-    pd[i].qcoeff = ctx->qcoeff_pbuf[i][1];
+    p[i].qcoeff = ctx->qcoeff_pbuf[i][1];
     pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1];
     pd[i].eobs = ctx->eobs_pbuf[i][1];
   }
diff --git a/vp9/encoder/vp9_quantize.c b/vp9/encoder/vp9_quantize.c
index d24be96f6..698130aed 100644
--- a/vp9/encoder/vp9_quantize.c
+++ b/vp9/encoder/vp9_quantize.c
@@ -173,7 +173,7 @@ void vp9_regular_quantize_b_4x4(MACROBLOCK *x, int y_blocks, int b_idx,
   vp9_quantize_b(BLOCK_OFFSET(p->coeff, pb_idx.block),
            16, x->skip_block,
            p->zbin, p->round, p->quant, p->quant_shift,
-           BLOCK_OFFSET(pd->qcoeff, pb_idx.block),
+           BLOCK_OFFSET(p->qcoeff, pb_idx.block),
            BLOCK_OFFSET(pd->dqcoeff, pb_idx.block),
            pd->dequant, p->zbin_extra, &pd->eobs[pb_idx.block], scan, iscan);
 }
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 43b7d6b7c..0c61b0d9f 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -514,11 +514,12 @@ static INLINE int cost_coeffs(MACROBLOCK *x,
                               const int16_t *scan, const int16_t *nb) {
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *mbmi = &xd->mi_8x8[0]->mbmi;
+  struct macroblock_plane *p = &x->plane[plane];
   struct macroblockd_plane *pd = &xd->plane[plane];
   const PLANE_TYPE type = pd->plane_type;
   const int16_t *band_count = &band_counts[tx_size][1];
   const int eob = pd->eobs[block];
-  const int16_t *const qcoeff_ptr = BLOCK_OFFSET(pd->qcoeff, block);
+  const int16_t *const qcoeff_ptr = BLOCK_OFFSET(p->qcoeff, block);
   const int ref = mbmi->ref_frame[0] != INTRA_FRAME;
   unsigned int (*token_costs)[2][PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS] =
                    x->token_costs[tx_size][type][ref];
@@ -1380,7 +1381,7 @@ static int64_t rd_pick_intra_sbuv_mode(VP9_COMP *cpi, MACROBLOCK *x,
         struct macroblockd_plane *const pd = x->e_mbd.plane;
         for (i = 1; i < MAX_MB_PLANE; ++i) {
           p[i].coeff    = ctx->coeff_pbuf[i][2];
-          pd[i].qcoeff  = ctx->qcoeff_pbuf[i][2];
+          p[i].qcoeff  = ctx->qcoeff_pbuf[i][2];
           pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][2];
           pd[i].eobs    = ctx->eobs_pbuf[i][2];
 
@@ -1390,7 +1391,7 @@ static int64_t rd_pick_intra_sbuv_mode(VP9_COMP *cpi, MACROBLOCK *x,
           ctx->eobs_pbuf[i][2]    = ctx->eobs_pbuf[i][0];
 
           ctx->coeff_pbuf[i][0]   = p[i].coeff;
-          ctx->qcoeff_pbuf[i][0]  = pd[i].qcoeff;
+          ctx->qcoeff_pbuf[i][0]  = p[i].qcoeff;
           ctx->dqcoeff_pbuf[i][0] = pd[i].dqcoeff;
           ctx->eobs_pbuf[i][0]    = pd[i].eobs;
         }
@@ -3057,7 +3058,7 @@ static void swap_block_ptr(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx,
 
   for (i = 0; i < max_plane; ++i) {
     p[i].coeff    = ctx->coeff_pbuf[i][1];
-    pd[i].qcoeff  = ctx->qcoeff_pbuf[i][1];
+    p[i].qcoeff  = ctx->qcoeff_pbuf[i][1];
     pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1];
     pd[i].eobs    = ctx->eobs_pbuf[i][1];
 
@@ -3067,7 +3068,7 @@ static void swap_block_ptr(MACROBLOCK *x, PICK_MODE_CONTEXT *ctx,
     ctx->eobs_pbuf[i][1]    = ctx->eobs_pbuf[i][0];
 
     ctx->coeff_pbuf[i][0]   = p[i].coeff;
-    ctx->qcoeff_pbuf[i][0]  = pd[i].qcoeff;
+    ctx->qcoeff_pbuf[i][0]  = p[i].qcoeff;
     ctx->dqcoeff_pbuf[i][0] = pd[i].dqcoeff;
     ctx->eobs_pbuf[i][0]    = pd[i].eobs;
   }
diff --git a/vp9/encoder/vp9_tokenize.c b/vp9/encoder/vp9_tokenize.c
index 3f1cc6fe8..1aaf72e40 100644
--- a/vp9/encoder/vp9_tokenize.c
+++ b/vp9/encoder/vp9_tokenize.c
@@ -108,7 +108,8 @@ static void tokenize_b(int plane, int block, BLOCK_SIZE plane_bsize,
   TOKENEXTRA *t = *tp;        /* store tokens starting here */
   const int eob = pd->eobs[block];
   const PLANE_TYPE type = pd->plane_type;
-  const int16_t *qcoeff_ptr = BLOCK_OFFSET(pd->qcoeff, block);
+  struct macroblock_plane *p = &cpi->mb.plane[plane];
+  const int16_t *qcoeff_ptr = BLOCK_OFFSET(p->qcoeff, block);
 
   const int segment_id = mbmi->segment_id;
   const int16_t *scan, *nb;