Moving txfm_size bits before prediction mode bits.

Hardware implementation needs to load coeff probs based on the
transform size. For selectable transform size, moving these bits
earlier in the bitstream adds some delay giving time to preload
the probs and speeds up the decoding process.

Change-Id: I3bfc1f662ae6f219c9286fe9ae6310c7d8a63ea7

1 files changed, 36 insertions, 37 deletions

diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index 2bf308669..9f6a0f2cc 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -118,6 +118,24 @@ static void kfread_modes(VP9D_COMP *pbi, MODE_INFO *m,
   if (!m->mbmi.mb_skip_coeff)
     m->mbmi.mb_skip_coeff = vp9_read(r, vp9_get_pred_prob(cm, xd, PRED_MBSKIP));
 
+  if (cm->txfm_mode == TX_MODE_SELECT &&
+      m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+    const int allow_16x16 = m->mbmi.sb_type >= BLOCK_SIZE_MB16X16;
+    const int allow_32x32 = m->mbmi.sb_type >= BLOCK_SIZE_SB32X32;
+    m->mbmi.txfm_size = select_txfm_size(cm, r, allow_16x16, allow_32x32);
+  } else if (cm->txfm_mode >= ALLOW_32X32 &&
+             m->mbmi.sb_type >= BLOCK_SIZE_SB32X32) {
+    m->mbmi.txfm_size = TX_32X32;
+  } else if (cm->txfm_mode >= ALLOW_16X16 &&
+             m->mbmi.sb_type >= BLOCK_SIZE_MB16X16) {
+    m->mbmi.txfm_size = TX_16X16;
+  } else if (cm->txfm_mode >= ALLOW_8X8 &&
+             m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
+    m->mbmi.txfm_size = TX_8X8;
+  } else {
+    m->mbmi.txfm_size = TX_4X4;
+  }
+
   // luma mode
   if (m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
     const MB_PREDICTION_MODE A = above_block_mode(m, 0, mis);
@@ -153,25 +171,6 @@ static void kfread_modes(VP9D_COMP *pbi, MODE_INFO *m,
   }
 
   m->mbmi.uv_mode = read_uv_mode(r, cm->kf_uv_mode_prob[m->mbmi.mode]);
-
-  if (cm->txfm_mode == TX_MODE_SELECT &&
-      m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
-    const int allow_16x16 = m->mbmi.sb_type >= BLOCK_SIZE_MB16X16;
-    const int allow_32x32 = m->mbmi.sb_type >= BLOCK_SIZE_SB32X32;
-    m->mbmi.txfm_size = select_txfm_size(cm, r, allow_16x16, allow_32x32);
-  } else if (cm->txfm_mode >= ALLOW_32X32 &&
-             m->mbmi.sb_type >= BLOCK_SIZE_SB32X32) {
-    m->mbmi.txfm_size = TX_32X32;
-  } else if (cm->txfm_mode >= ALLOW_16X16 &&
-             m->mbmi.sb_type >= BLOCK_SIZE_MB16X16 &&
-             m->mbmi.mode <= TM_PRED) {
-    m->mbmi.txfm_size = TX_16X16;
-  } else if (cm->txfm_mode >= ALLOW_8X8 &&
-             m->mbmi.sb_type >= BLOCK_SIZE_SB8X8) {
-    m->mbmi.txfm_size = TX_8X8;
-  } else {
-    m->mbmi.txfm_size = TX_4X4;
-  }
 }
 
 static int read_mv_component(vp9_reader *r,
@@ -569,6 +568,24 @@ static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
   // Read the reference frame
   mbmi->ref_frame = read_ref_frame(pbi, r, mbmi->segment_id);
 
+  if (cm->txfm_mode == TX_MODE_SELECT &&
+      (mbmi->mb_skip_coeff == 0 || mbmi->ref_frame == INTRA_FRAME) &&
+      bsize >= BLOCK_SIZE_SB8X8) {
+    const int allow_16x16 = bsize >= BLOCK_SIZE_MB16X16;
+    const int allow_32x32 = bsize >= BLOCK_SIZE_SB32X32;
+    mbmi->txfm_size = select_txfm_size(cm, r, allow_16x16, allow_32x32);
+  } else if (bsize >= BLOCK_SIZE_SB32X32 &&
+             cm->txfm_mode >= ALLOW_32X32) {
+    mbmi->txfm_size = TX_32X32;
+  } else if (cm->txfm_mode >= ALLOW_16X16 &&
+             bsize >= BLOCK_SIZE_MB16X16) {
+    mbmi->txfm_size = TX_16X16;
+  } else if (cm->txfm_mode >= ALLOW_8X8 && (bsize >= BLOCK_SIZE_SB8X8)) {
+    mbmi->txfm_size = TX_8X8;
+  } else {
+    mbmi->txfm_size = TX_4X4;
+  }
+
   // If reference frame is an Inter frame
   if (mbmi->ref_frame) {
     int_mv nearest, nearby, best_mv;
@@ -833,24 +850,6 @@ static void read_mb_modes_mv(VP9D_COMP *pbi, MODE_INFO *mi, MB_MODE_INFO *mbmi,
     mbmi->uv_mode = read_uv_mode(r, cm->fc.uv_mode_prob[mbmi->mode]);
     cm->fc.uv_mode_counts[mbmi->mode][mbmi->uv_mode]++;
   }
-
-  if (cm->txfm_mode == TX_MODE_SELECT &&
-      (mbmi->mb_skip_coeff == 0 || mbmi->ref_frame == INTRA_FRAME) &&
-      bsize >= BLOCK_SIZE_SB8X8) {
-    const int allow_16x16 = bsize >= BLOCK_SIZE_MB16X16;
-    const int allow_32x32 = bsize >= BLOCK_SIZE_SB32X32;
-    mbmi->txfm_size = select_txfm_size(cm, r, allow_16x16, allow_32x32);
-  } else if (bsize >= BLOCK_SIZE_SB32X32 &&
-             cm->txfm_mode >= ALLOW_32X32) {
-    mbmi->txfm_size = TX_32X32;
-  } else if (cm->txfm_mode >= ALLOW_16X16 &&
-             bsize >= BLOCK_SIZE_MB16X16) {
-    mbmi->txfm_size = TX_16X16;
-  } else if (cm->txfm_mode >= ALLOW_8X8 && (bsize >= BLOCK_SIZE_SB8X8)) {
-    mbmi->txfm_size = TX_8X8;
-  } else {
-    mbmi->txfm_size = TX_4X4;
-  }
 }
 
 void vp9_decode_mode_mvs_init(VP9D_COMP* const pbi, vp9_reader *r) {