M2 perf: skip internal deblocking edges of uniform-motion macroblocks

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

crates/h264/README.md

@@ -143,8 +143,8 @@ Pro (14 cores) while the machine was heavily shared (load average 40-180), so th
 
 | threads | fps |
 |---|---|
-| 1 | ~100-115 (≈ 36 Mcycles per frame) |
-| 14 | ~570-600 |
+| 1 | ~105-122 (≈ 33 Mcycles per frame) |
+| 14 | ~500-600 |
 
 The example `cargo run --release -p filmcraft-h264 --example h264dec -- in.h264 out.yuv` decodes a
 file; `H264_BENCH_ITERS=n` (and `H264_THREADS=t`) turns it into a benchmark, `H264_STATS=1` prints

crates/h264/src/deblock.rs

@@ -231,6 +231,10 @@ pub fn deblock_mb(pic: &mut PicState, addr: usize, mb_w: usize) {
         }
     } else {
         let qids = ref_ids8(q, sq);
+        let q_uniform = (0..2).all(|l| {
+            let r = q.ref_idx[l];
+            r[1] == r[0] && r[2] == r[0] && r[3] == r[0] && q.mv[l].iter().all(|m| *m == q.mv[l][0])
+        });
         let qm: [Motion; 16] = std::array::from_fn(|r| block_motion(q, &qids, r));
         for dir in 0..2 {
             let neighbor = if dir == 0 { left } else { top };
@@ -247,6 +251,10 @@ pub fn deblock_mb(pic: &mut PicState, addr: usize, mb_w: usize) {
                     }
                 }
             }
+            if q_uniform && q.nz_mask == 0 {
+                // one motion for the whole MB and no coefficients: all internal edges have bS 0
+                continue;
+            }
             for e in 1..4 {
                 if t8 && e % 2 == 1 {
                     continue;

crates/h264/src/slicedec.rs

@@ -64,7 +64,10 @@ impl PicState {
     /// Reuse the buffers of a finished picture of the same size for a new frame.
     pub fn reset(&mut self, frame: FrameRef) {
         debug_assert_eq!((self.mb_w, self.mb_h), (frame.mb_w, frame.mb_h()));
-        self.mbs.fill(MbState::default());
+        // start_mb() re-initialises each macroblock before use; only availability must be reset
+        for m in &mut self.mbs {
+            m.slice_num = u32::MAX;
+        }
         self.slices.clear();
         self.poc = frame.poc;
         self.decoded_mbs = 0;