Fix DC_PRED and CFL_PRED at tile boundaries

A BlockRegion only knows its absolute offset, not its offset relative to
the tile.

In predict_intra_inner(), pass the whole region for the tile and the
block offset relative to the tile.

Signed-off-by: Romain Vimont <rom1v@videolabs.io>

Author: barrbrain

src/encoder.rs

@@ -1026,7 +1026,7 @@ pub fn encode_tx_block<T: Pixel>(
   if mode.is_intra() {
     let bit_depth = fi.sequence.bit_depth;
     let edge_buf = get_intra_edges(&rec.as_const(), po, tx_size, bit_depth, Some(mode));
-    mode.predict_intra(&mut rec.subregion_mut(area), tx_size, bit_depth, &ac, alpha, &edge_buf);
+    mode.predict_intra(rec, tile_bo, tx_size, bit_depth, &ac, alpha, &edge_buf);
   }
 
   if skip { return (false, -1); }

src/partition.rs

@@ -1009,65 +1009,68 @@ pub fn get_intra_edges<T: Pixel>(
 
 impl PredictionMode {
   pub fn predict_intra<T: Pixel>(
-    self, dst: &mut PlaneRegionMut<'_, T>, tx_size: TxSize, bit_depth: usize,
+    self, dst: &mut PlaneRegionMut<'_, T>, bo: BlockOffset, tx_size: TxSize, bit_depth: usize,
     ac: &[i16], alpha: i16, edge_buf: &AlignedArray<[T; 4 * MAX_TX_SIZE + 1]>
   ) {
     assert!(self.is_intra());
 
     match tx_size {
       TxSize::TX_4X4 =>
-        self.predict_intra_inner::<Block4x4, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block4x4, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_8X8 =>
-        self.predict_intra_inner::<Block8x8, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block8x8, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_16X16 =>
-        self.predict_intra_inner::<Block16x16, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block16x16, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_32X32 =>
-        self.predict_intra_inner::<Block32x32, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block32x32, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_64X64 =>
-        self.predict_intra_inner::<Block64x64, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block64x64, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
 
       TxSize::TX_4X8 =>
-        self.predict_intra_inner::<Block4x8, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block4x8, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_8X4 =>
-        self.predict_intra_inner::<Block8x4, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block8x4, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_8X16 =>
-        self.predict_intra_inner::<Block8x16, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block8x16, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_16X8 =>
-        self.predict_intra_inner::<Block16x8, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block16x8, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_16X32 =>
-        self.predict_intra_inner::<Block16x32, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block16x32, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_32X16 =>
-        self.predict_intra_inner::<Block32x16, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block32x16, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_32X64 =>
-        self.predict_intra_inner::<Block32x64, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block32x64, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_64X32 =>
-        self.predict_intra_inner::<Block64x32, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block64x32, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
 
       TxSize::TX_4X16 =>
-        self.predict_intra_inner::<Block4x16, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block4x16, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_16X4 =>
-        self.predict_intra_inner::<Block16x4, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block16x4, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_8X32 =>
-        self.predict_intra_inner::<Block8x32, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block8x32, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_32X8 =>
-        self.predict_intra_inner::<Block32x8, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block32x8, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_16X64 =>
-        self.predict_intra_inner::<Block16x64, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block16x64, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
       TxSize::TX_64X16 =>
-        self.predict_intra_inner::<Block64x16, _>(dst, bit_depth, ac, alpha, edge_buf),
+        self.predict_intra_inner::<Block64x16, _>(dst, bo, bit_depth, ac, alpha, edge_buf),
     }
   }
 
   #[inline(always)]
   fn predict_intra_inner<B: Intra<T>, T: Pixel>(
-    self, dst: &mut PlaneRegionMut<'_, T>, bit_depth: usize, ac: &[i16],
-    alpha: i16, edge_buf: &AlignedArray<[T; 4 * MAX_TX_SIZE + 1]>
+    self, dst: &mut PlaneRegionMut<'_, T>, tile_bo: BlockOffset,
+    bit_depth: usize, ac: &[i16], alpha: i16,
+    edge_buf: &AlignedArray<[T; 4 * MAX_TX_SIZE + 1]>
   ) {
     // left pixels are order from bottom to top and right-aligned
     let (left, not_left) = edge_buf.array.split_at(2*MAX_TX_SIZE);
     let (top_left, above) = not_left.split_at(1);
 
-    let &Rect { x, y, .. } = dst.rect();
+    let BlockOffset { x, y, .. } = tile_bo;
+    let area = Area::BlockStartingAt { bo: tile_bo };
+    let dst = &mut dst.subregion_mut(area);
 
     let mode: PredictionMode = match self {
       PredictionMode::PAETH_PRED => match (x, y) {

src/rdo.rs

@@ -734,9 +734,9 @@ pub fn rdo_mode_decision<T: Pixel>(
         .iter()
         .map(|&luma_mode| {
           let rec = &mut ts.rec.planes[0];
-          let mut rec_region = rec.subregion_mut(Area::BlockStartingAt { bo: tile_bo });
           luma_mode.predict_intra(
-            &mut rec_region,
+            rec,
+            tile_bo,
             tx_size,
             fi.sequence.bit_depth,
             &[0i16; 2],
@@ -745,7 +745,7 @@ pub fn rdo_mode_decision<T: Pixel>(
           );
 
           let plane_org = ts.input_tile.planes[0].subregion(Area::BlockStartingAt { bo: tile_bo });
-          let plane_ref = rec_region.as_const();
+          let plane_ref = rec.subregion(Area::BlockStartingAt { bo: tile_bo });
 
           (
             luma_mode,
@@ -914,9 +914,9 @@ pub fn rdo_cfl_alpha<T: Pixel>(
             Some(PredictionMode::UV_CFL_PRED)
           );
 
-          let mut rec_region = rec.subregion_mut(Area::BlockStartingAt { bo: tile_bo });
           PredictionMode::UV_CFL_PRED.predict_intra(
-            &mut rec_region,
+            rec,
+            tile_bo,
             uv_tx_size,
             bit_depth,
             &ac.array,
@@ -925,7 +925,7 @@ pub fn rdo_cfl_alpha<T: Pixel>(
           );
           sse_wxh(
             &input.subregion(Area::BlockStartingAt { bo: tile_bo }),
-            &rec_region.as_const(),
+            &rec.subregion(Area::BlockStartingAt { bo: tile_bo }),
             uv_tx_size.width(),
             uv_tx_size.height()
           )