perf: fast display (row runs, bulk fills, pixel scroll), terminal draws changed cells only + scrolls pixels, print! batched (was IPC per fmt fragment), snake sleeps on timer

crates/libkos/src/window.rs

@@ -118,15 +118,36 @@ impl Window {
         }
     }
 
-    /// `count` u32s from `offset` set to `value`.
+    /// `count` u32s from `offset` set to `value`: one check, then a run of
+    /// stores (drawing fills thousands at a time).
     pub fn fill_u32(&self, offset: ByteOffset, count: usize, value: u32) {
         let bytes = count.checked_mul(4).expect("window fill too large");
-        self.at(offset, bytes, 4);
+        let addr = self.at(offset, bytes, 4) as *mut u32;
         for i in 0..count {
-            self.write(offset + i * 4, value);
+            // SAFETY: the whole run was checked in range and aligned.
+            unsafe { core::ptr::write_volatile(addr.add(i), value) }
         }
     }
 
+    /// `values` stored from `offset` on, in one checked run.
+    pub fn write_u32s(&self, offset: ByteOffset, values: &[u32]) {
+        let addr = self.at(offset, values.len() * 4, 4) as *mut u32;
+        for (i, v) in values.iter().enumerate() {
+            // SAFETY: the whole run was checked in range and aligned.
+            unsafe { core::ptr::write_volatile(addr.add(i), *v) }
+        }
+    }
+
+    /// Moves `len` bytes from `from` to `to` inside the window; the two may
+    /// overlap.
+    pub fn copy_within(&self, from: ByteOffset, to: ByteOffset, len: usize) {
+        let src = self.at(from, len, 1) as *const u8;
+        let dst = self.at(to, len, 1) as *mut u8;
+        // SAFETY: both ranges are inside the mapped window; `copy` allows
+        // overlap. Plain memory moves, not device registers.
+        unsafe { core::ptr::copy(src, dst, len) }
+    }
+
     /// `count` words from `offset` as atomics, for memory shared with
     /// threads or processes that only ever touch it atomically.
     pub fn atomics(&self, offset: ByteOffset, count: usize) -> &'static [AtomicUsize] {