vp3: fix motion compensation code

diff --git a/nihav-duck/src/codecs/vp3.rs b/nihav-duck/src/codecs/vp3.rs
index 947377498cf698d95cbace4d5d33462a2296f0a2..e52c9400f1fddaef63542f6ec36f21d8930fba0e 100644 (file)
--- a/nihav-duck/src/codecs/vp3.rs
+++ b/nihav-duck/src/codecs/vp3.rs
@@ -552,6 +552,15 @@ fn vp31_loop_filter_h(frm: &mut NASimpleVideoFrame<u8>, x: usize, y: usize, plan
     vp31_loop_filter(frm.data, off, frm.stride[plane], 1, 8, loop_str);
 }
 
+fn vp3_mv_mode(mvx: i16, mvy: i16) -> usize {
+    let mode = ((mvx & 1) + (mvy & 1) * 2) as usize;
+    if (mode == 3) && (mvx ^ mvy < 0) {
+        4
+    } else {
+        mode
+    }
+}
+
 impl VP34Decoder {
     fn new(version: u8) -> Self {
         let vt = alloc_video_buffer(NAVideoInfo::new(24, 24, false, YUV420_FORMAT), 4).unwrap();
@@ -1550,7 +1559,7 @@ println!("intra, ver {} (self {})", version, self.version);
                     let xoff = (i &  1) * 8;
                     let yoff = (i >> 1) * 8;
                     
-                    let mode = ((mvs[i].x & 1) + (mvs[i].y & 1) * 2) as usize;
+                    let mode = vp3_mv_mode(mvs[i].x, mvs[i].y);
                     if self.version != 4 {
                         copy_block(frm, src.clone(), 0, bx * 8 + xoff, by * 8 + yoff,
                                    mvs[i].x >> 1, mvs[i].y >> 1, 8, 8, 0, 1, mode, VP3_INTERP_FUNCS);
@@ -1563,7 +1572,7 @@ println!("intra, ver {} (self {})", version, self.version);
 
                 let mx = (mv_sum.x >> 1) | (mv_sum.x & 1);
                 let my = (mv_sum.y >> 1) | (mv_sum.y & 1);
-                let mode = ((mx & 1) + (my & 1) * 2) as usize;
+                let mode = vp3_mv_mode(mx, my);
                 copy_block(frm, src.clone(), 1, bx * 4, by * 4, mx >> 1, my >> 1, 8, 8, 0, 1, mode, VP3_INTERP_FUNCS);
                 copy_block(frm, src.clone(), 2, bx * 4, by * 4, mx >> 1, my >> 1, 8, 8, 0, 1, mode, VP3_INTERP_FUNCS);
             }
@@ -1581,7 +1590,7 @@ println!("intra, ver {} (self {})", version, self.version);
                         } else {
                             self.shuf.get_golden().unwrap()
                         };
-                    let mode = ((blk.mv.x & 1) + (blk.mv.y & 1) * 2) as usize;
+                    let mode = vp3_mv_mode(blk.mv.x, blk.mv.y);
                     if self.version != 4 {
                         copy_block(frm, src.clone(), 0, bx * 8, by * 8,
                                    blk.mv.x >> 1, blk.mv.y >> 1, 16, 16, 0, 1, mode, VP3_INTERP_FUNCS);
@@ -1601,7 +1610,7 @@ println!("intra, ver {} (self {})", version, self.version);
                     }
                     let mx = (blk.mv.x >> 1) | (blk.mv.x & 1);
                     let my = (blk.mv.y >> 1) | (blk.mv.y & 1);
-                    let mode = ((mx & 1) + (my & 1) * 2) as usize;
+                    let mode = vp3_mv_mode(mx, my);
                     if self.version != 4 {
                         copy_block(frm, src.clone(), 1, bx * 4, by * 4,
                                    mx >> 1, my >> 1, 8, 8, 0, 1, mode, VP3_INTERP_FUNCS);

diff --git a/nihav-duck/src/codecs/vpcommon.rs b/nihav-duck/src/codecs/vpcommon.rs
index d1f6d446266041824673c6ff41de987c032d43a0..bd7d1cf49066df5e99ea8d61ac501686c378e3b3 100644 (file)
--- a/nihav-duck/src/codecs/vpcommon.rs
+++ b/nihav-duck/src/codecs/vpcommon.rs
@@ -432,21 +432,33 @@ fn vp3_interp10(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, bw:
     }
 }
 
-fn vp3_interp11(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, bw: usize, bh: usize)
+fn vp3_interp1x(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, bw: usize, bh: usize)
 {
     let mut didx = 0;
     let mut sidx = 0;
     for _ in 0..bh {
         for x in 0..bw {
             dst[didx + x] = (((src[sidx + x] as u16) +
-                              (src[sidx + x + 1] as u16) +
-                              (src[sidx + x + sstride] as u16) +
-                              (src[sidx + x + sstride + 1] as u16)) >> 2) as u8;
+                              (src[sidx + x + sstride + 1] as u16)) >> 1) as u8;
         }
         didx += dstride;
         sidx += sstride;
     }
 }
 
-pub const VP3_INTERP_FUNCS: &[blockdsp::BlkInterpFunc] = &[ vp3_interp00, vp3_interp01, vp3_interp10, vp3_interp11 ];
+fn vp3_interp1y(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, bw: usize, bh: usize)
+{
+    let mut didx = 0;
+    let mut sidx = 0;
+    for _ in 0..bh {
+        for x in 0..bw {
+            dst[didx + x] = (((src[sidx + x + 1] as u16) +
+                              (src[sidx + x + sstride] as u16)) >> 1) as u8;
+        }
+        didx += dstride;
+        sidx += sstride;
+    }
+}
+
+pub const VP3_INTERP_FUNCS: &[blockdsp::BlkInterpFunc] = &[ vp3_interp00, vp3_interp01, vp3_interp10, vp3_interp1x, vp3_interp1y ];