shithub: libvpx

Download patch

ref: a6bc3dfb0f8e46017d7823985f8383a8dc6cc68b
parent: 009116cb6f7a75645c7c3092f97a5fa4cd42d15b
parent: e6446b4b60c7dd3cdeb6cfc3694f9a81df596764
author: Debargha Mukherjee <debargha@google.com>
date: Thu Sep 1 12:46:32 EDT 2016 

Merge "Refactor uv tx size with lookup arrays"


--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -228,20 +228,13 @@
 
 void vp9_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y);
 
-static INLINE TX_SIZE get_uv_tx_size_impl(TX_SIZE y_tx_size, BLOCK_SIZE bsize,
-                                          int xss, int yss) {
-  if (bsize < BLOCK_8X8) {
-    return TX_4X4;
-  } else {
-    const BLOCK_SIZE plane_bsize = ss_size_lookup[bsize][xss][yss];
-    return VPXMIN(y_tx_size, max_txsize_lookup[plane_bsize]);
-  }
-}
-
 static INLINE TX_SIZE get_uv_tx_size(const MODE_INFO *mi,
                                      const struct macroblockd_plane *pd) {
-  return get_uv_tx_size_impl(mi->tx_size, mi->sb_type, pd->subsampling_x,
-                             pd->subsampling_y);
+  assert(mi->sb_type < BLOCK_8X8 ||
+         ss_size_lookup[mi->sb_type][pd->subsampling_x][pd->subsampling_y] !=
+             BLOCK_INVALID);
+  return uv_txsize_lookup[mi->sb_type][mi->tx_size][pd->subsampling_x]
+                         [pd->subsampling_y];
 }
 
 static INLINE BLOCK_SIZE
--- a/vp9/common/vp9_common_data.c
+++ b/vp9/common/vp9_common_data.c
@@ -125,6 +125,102 @@
   { { BLOCK_64X64, BLOCK_64X32 }, { BLOCK_32X64, BLOCK_32X32 } },
 };
 
+const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES][2][2] = {
+  //  ss_x == 0    ss_x == 0        ss_x == 1      ss_x == 1
+  //  ss_y == 0    ss_y == 1        ss_y == 0      ss_y == 1
+  {
+      // BLOCK_4X4
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+  },
+  {
+      // BLOCK_4X8
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+  },
+  {
+      // BLOCK_8X4
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+  },
+  {
+      // BLOCK_8X8
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_4X4 }, { TX_4X4, TX_4X4 } },
+  },
+  {
+      // BLOCK_8X16
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_4X4, TX_4X4 } },
+  },
+  {
+      // BLOCK_16X8
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_4X4 }, { TX_8X8, TX_4X4 } },
+      { { TX_8X8, TX_4X4 }, { TX_8X8, TX_8X8 } },
+      { { TX_8X8, TX_4X4 }, { TX_8X8, TX_8X8 } },
+  },
+  {
+      // BLOCK_16X16
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_8X8 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_8X8 }, { TX_8X8, TX_8X8 } },
+  },
+  {
+      // BLOCK_16X32
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_16X16 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_16X16 }, { TX_8X8, TX_8X8 } },
+  },
+  {
+      // BLOCK_32X16
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_8X8 }, { TX_16X16, TX_8X8 } },
+      { { TX_16X16, TX_8X8 }, { TX_16X16, TX_8X8 } },
+  },
+  {
+      // BLOCK_32X32
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_16X16 }, { TX_16X16, TX_16X16 } },
+      { { TX_32X32, TX_16X16 }, { TX_16X16, TX_16X16 } },
+  },
+  {
+      // BLOCK_32X64
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_16X16 }, { TX_16X16, TX_16X16 } },
+      { { TX_32X32, TX_32X32 }, { TX_16X16, TX_16X16 } },
+  },
+  {
+      // BLOCK_64X32
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_16X16 }, { TX_16X16, TX_16X16 } },
+      { { TX_32X32, TX_16X16 }, { TX_32X32, TX_16X16 } },
+  },
+  {
+      // BLOCK_64X64
+      { { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
+      { { TX_8X8, TX_8X8 }, { TX_8X8, TX_8X8 } },
+      { { TX_16X16, TX_16X16 }, { TX_16X16, TX_16X16 } },
+      { { TX_32X32, TX_32X32 }, { TX_32X32, TX_32X32 } },
+  },
+};
+
 // Generates 4 bit field in which each bit set to 1 represents
 // a blocksize partition  1111 means we split 64x64, 32x32, 16x16
 // and 8x8.  1000 means we just split the 64x64 to 32x32
--- a/vp9/common/vp9_common_data.h
+++ b/vp9/common/vp9_common_data.h
@@ -33,6 +33,7 @@
 extern const BLOCK_SIZE txsize_to_bsize[TX_SIZES];
 extern const TX_SIZE tx_mode_to_biggest_tx_size[TX_MODES];
 extern const BLOCK_SIZE ss_size_lookup[BLOCK_SIZES][2][2];
+extern const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES][2][2];
 #if CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH
 extern const uint8_t need_top_left[INTRA_MODES];
 #endif  // CONFIG_BETTER_HW_COMPATIBILITY && CONFIG_VP9_HIGHBITDEPTH
--- a/vp9/common/vp9_loopfilter.c
+++ b/vp9/common/vp9_loopfilter.c
@@ -655,7 +655,7 @@
                         const int shift_uv, LOOP_FILTER_MASK *lfm) {
   const BLOCK_SIZE block_size = mi->sb_type;
   const TX_SIZE tx_size_y = mi->tx_size;
-  const TX_SIZE tx_size_uv = get_uv_tx_size_impl(tx_size_y, block_size, 1, 1);
+  const TX_SIZE tx_size_uv = uv_txsize_lookup[block_size][tx_size_y][1][1];
   const int filter_level = get_filter_level(lfi_n, mi);
   uint64_t *const left_y = &lfm->left_y[tx_size_y];
   uint64_t *const above_y = &lfm->above_y[tx_size_y];
@@ -1524,7 +1524,7 @@
   const TX_SIZE tx_size_y = mi->tx_size;
   const loop_filter_info_n *const lfi_n = &cm->lf_info;
   const int filter_level = get_filter_level(lfi_n, mi);
-  const TX_SIZE tx_size_uv = get_uv_tx_size_impl(tx_size_y, block_size, 1, 1);
+  const TX_SIZE tx_size_uv = uv_txsize_lookup[block_size][tx_size_y][1][1];
   LOOP_FILTER_MASK *const lfm = get_lfm(&cm->lf, mi_row, mi_col);
   uint64_t *const left_y = &lfm->left_y[tx_size_y];
   uint64_t *const above_y = &lfm->above_y[tx_size_y];
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -712,13 +712,6 @@
   }
 }
 
-static INLINE TX_SIZE dec_get_uv_tx_size(const MODE_INFO *mi, int n4_wl,
-                                         int n4_hl) {
-  // get minimum log2 num4x4s dimension
-  const int x = VPXMIN(n4_wl, n4_hl);
-  return VPXMIN(mi->tx_size, x);
-}
-
 static INLINE void dec_reset_skip_context(MACROBLOCKD *xd) {
   int i;
   for (i = 0; i < MAX_MB_PLANE; i++) {
@@ -799,8 +792,7 @@
     int plane;
     for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
       const struct macroblockd_plane *const pd = &xd->plane[plane];
-      const TX_SIZE tx_size =
-          plane ? dec_get_uv_tx_size(mi, pd->n4_wl, pd->n4_hl) : mi->tx_size;
+      const TX_SIZE tx_size = plane ? get_uv_tx_size(mi, pd) : mi->tx_size;
       const int num_4x4_w = pd->n4_w;
       const int num_4x4_h = pd->n4_h;
       const int step = (1 << tx_size);
@@ -833,8 +825,7 @@
 
       for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
         const struct macroblockd_plane *const pd = &xd->plane[plane];
-        const TX_SIZE tx_size =
-            plane ? dec_get_uv_tx_size(mi, pd->n4_wl, pd->n4_hl) : mi->tx_size;
+        const TX_SIZE tx_size = plane ? get_uv_tx_size(mi, pd) : mi->tx_size;
         const int num_4x4_w = pd->n4_w;
         const int num_4x4_h = pd->n4_h;
         const int step = (1 << tx_size);
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -2848,8 +2848,8 @@
       return;
     }
   }
-  max_uv_tx_size = get_uv_tx_size_impl(
-      xd->mi[0]->tx_size, bsize, pd[1].subsampling_x, pd[1].subsampling_y);
+  max_uv_tx_size = uv_txsize_lookup[bsize][xd->mi[0]->tx_size]
+                                   [pd[1].subsampling_x][pd[1].subsampling_y];
   rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv, &rate_uv_tokenonly, &dist_uv,
                           &uv_skip, VPXMAX(BLOCK_8X8, bsize), max_uv_tx_size);
 
@@ -3344,8 +3344,8 @@
                       best_rd);
       if (rate_y == INT_MAX) continue;
 
-      uv_tx = get_uv_tx_size_impl(mi->tx_size, bsize, pd->subsampling_x,
-                                  pd->subsampling_y);
+      uv_tx = uv_txsize_lookup[bsize][mi->tx_size][pd->subsampling_x]
+                              [pd->subsampling_y];
       if (rate_uv_intra[uv_tx] == INT_MAX) {
         choose_intra_uv_mode(cpi, x, ctx, bsize, uv_tx, &rate_uv_intra[uv_tx],
                              &rate_uv_tokenonly[uv_tx], &dist_uv[uv_tx],