fix(optimizer): preserve cross-block LoadImm uses in const_fold DCE

`const_fold_block`'s per-block dead-code pass was collecting temp
usage from only the block it was folding, so a `LoadImm` whose
destination is consumed by a *sibling* block (for example via the
merge block's branch terminator) was incorrectly treated as dead
and dropped. The `and` / `or` short-circuit lowering emits exactly
that shape: the false path writes `LoadImm(result, 0)` and joins
with the right path at an `and_end` / `or_end` block whose branch
terminator reads `result`. After the DCE the false path's store
was gone, leaving the zero-page result slot to carry whatever value
the *previous* `and` / `or` evaluation had written there — stale
data that bled into subsequent conditional branches.

I found this while instrumenting `examples/platformer.ne` through a
puppeteer-driven jsnes harness, stepping one frame at a time and
snapshotting the full zero-page trace of each scenario (title-skip,
hold-right, hold-left, jump-spam, coin-drift, enemy-stomp, long-run).
In a clean idle run the enemy-1 stomp bounce (`rise_count = 6`,
`fall_vy = 0`) fired at emulator frames 83 and 96 with `camera_x`
= 61 and 74, i.e. with `e1_sx` = 39 and 26, nowhere near the
intended `[72, 96)` pickup window. The trigger turned out to be
the slot alias: every time `c2_sx` landed in its pickup window
(so the coin-2 `and` stored 1 into ZP(130)) and the player was
mid-fall at or past `player_y = 152`, the enemy-1 stomp `and`
short-circuited to its false path, left ZP(130) at 1, and the
stomp `if` fired on stale data.

The fix is to compute function-wide source-operand usage once before
folding each function's blocks and OR it into the per-block liveness
check, so a LoadImm is only dropped if nobody — neither its own
block nor any other block in the function — reads the temp. Added a
regression test (`const_fold_preserves_loadimm_used_by_sibling_branch`)
that builds the exact CFG shape the `and` lowering emits and
verifies the false-path `LoadImm(result, 0)` survives optimization.

Impact on the example ROMs:

- `examples/platformer.nes`: enemy-1 stomp now fires only when
  `e1_sx ∈ [72, 96)`, as the source intends. The pixel golden is
  unchanged (`player_y` converges back to the ground line before
  frame 180), but the audio hash flips because the spurious
  `play hit` sfx calls during coin-2 passage are gone. Committing
  the new `tests/emulator/goldens/platformer.audio.hash`.

- `examples/logic_ops.nes`, `examples/bitwise_ops.nes`,
  `examples/match_demo.nes`, `examples/mmc3_per_state_split.nes`,
  `examples/two_player.nes`: byte-different but observably
  unchanged — their pixel + audio goldens still match to the byte.
  They exercise `and` / `or` in the source and now compile through
  the corrected DCE.

All other example ROMs are byte-identical to pre-fix. `cargo fmt`,
`cargo clippy --all-targets`, `cargo test --release` (498 tests),
and `tests/emulator/run_examples.mjs` (22/22 goldens) are clean.

https://claude.ai/code/session_013Bi4H4YQ5or5HtMB4doUFi

src/optimizer/mod.rs

@@ -218,13 +218,22 @@ fn strength_reduce_block(block: &mut IrBasicBlock) {
 /// destination temps are no longer referenced anywhere in the block.
 fn const_fold(program: &mut IrProgram) {
     for func in &mut program.functions {
+        // Pre-compute function-wide source-operand usage: a LoadImm's
+        // destination may be read by an op in a sibling block or
+        // consumed by a branch/jump terminator in another block, so
+        // the per-block DCE below can't decide liveness by looking
+        // at its own block alone. Cf. the `and` / `or` short-circuit
+        // lowering: the false path writes `LoadImm(result, 0)` but
+        // `result` is read by the merge block's branch, not in the
+        // false block itself.
+        let func_used = collect_used_temps(func);
         for block in &mut func.blocks {
-            const_fold_block(block);
+            const_fold_block(block, &func_used);
         }
     }
 }
 
-fn const_fold_block(block: &mut IrBasicBlock) {
+fn const_fold_block(block: &mut IrBasicBlock, func_used: &HashSet<IrTemp>) {
     let mut constants: HashMap<IrTemp, u8> = HashMap::new();
 
     // First pass: fold arithmetic / comparison ops into LoadImm where possible.
@@ -349,12 +358,21 @@ fn const_fold_block(block: &mut IrBasicBlock) {
         }
     }
 
-    // Second pass: remove LoadImm ops whose dest temps are no longer referenced
+    // Second pass: remove LoadImm ops whose dest temps are no longer
-    // as source operands by anything else in the block (ops + terminator).
+    // referenced locally AND aren't referenced function-wide. The
-    let used = collect_used_temps_in_block(block);
+    // function-wide check is what makes this pass correct in the
+    // presence of control-flow merges — a LoadImm written in one
+    // block and consumed in another (for example the `and`/`or`
+    // short-circuit false path, whose `LoadImm(result, 0)` is only
+    // read by the downstream merge block's branch terminator) must
+    // not be dropped. The previous implementation only consulted
+    // block-local usage and silently dropped these cross-block
+    // LoadImms, leaving the zero-page result slot to carry whatever
+    // value the *previous* AND/OR had written into it.
+    let used_local = collect_used_temps_in_block(block);
     block.ops.retain(|op| {
         if let IrOp::LoadImm(t, _) = op {
-            used.contains(t)
+            used_local.contains(t) || func_used.contains(t)
         } else {
             true
         }
@@ -397,7 +415,10 @@ fn collect_used_temps(func: &IrFunction) -> HashSet<IrTemp> {
     used
 }
 
-/// Collect temps used as source operands within a single block (ops + terminator).
+/// Collect temps used as source operands within a single block
+/// (ops + terminator). Used by the per-block `LoadImm` DCE so we
+/// can cheaply find local uses before falling back on the
+/// function-wide liveness set.
 fn collect_used_temps_in_block(block: &IrBasicBlock) -> HashSet<IrTemp> {
     let mut used = HashSet::new();
     collect_used_from_block(block, &mut used);

src/optimizer/tests.rs

@@ -436,3 +436,85 @@ fn inline_removes_trivial() {
         .any(|op| matches!(op, IrOp::Call(..)));
     assert!(!has_call, "call to trivial function should be removed");
 }
+
+// ---------------------------------------------------------------------------
+// Cross-block LoadImm liveness (regression for stale-and-result bug)
+// ---------------------------------------------------------------------------
+
+/// A `LoadImm(result, 0)` in a predecessor block whose only consumer
+/// is a branch terminator in a *successor* block must not be dropped
+/// by the const-folding pass. This is the shape the logical-`and`
+/// short-circuit false path lowers to, and dropping the `LoadImm`
+/// leaves the zero-page slot holding whichever value the previous
+/// `and` evaluation wrote — which was observable in the platformer
+/// example as spurious enemy-stomp bounces firing whenever coin 2
+/// happened to be in its pickup window.
+#[test]
+fn const_fold_preserves_loadimm_used_by_sibling_branch() {
+    let cond = IrTemp(0);
+    let result = IrTemp(1);
+    let zero = IrTemp(2);
+
+    // Shape mirrors `lower_logical_and`: an entry block that branches
+    // on `cond`, a right-side block that writes `result` via an
+    // `Or(result, <true-case>, zero)`, a false-side block that only
+    // writes `LoadImm(result, 0)`, and a merge block whose terminator
+    // reads `result`.
+    let entry = IrBasicBlock {
+        label: "entry".to_string(),
+        ops: vec![IrOp::LoadImm(cond, 1)],
+        terminator: IrTerminator::Branch(cond, "right".to_string(), "false_path".to_string()),
+    };
+    let right = IrBasicBlock {
+        label: "right".to_string(),
+        ops: vec![IrOp::LoadImm(zero, 0), IrOp::Or(result, cond, zero)],
+        terminator: IrTerminator::Jump("end".to_string()),
+    };
+    // This is the block that previously tripped the bug: its
+    // `LoadImm(result, 0)` is the only op, and `result` is not used
+    // as a source operand anywhere *within* this block — but it is
+    // read by the `end` block's branch terminator.
+    let false_path = IrBasicBlock {
+        label: "false_path".to_string(),
+        ops: vec![IrOp::LoadImm(result, 0)],
+        terminator: IrTerminator::Jump("end".to_string()),
+    };
+    let end = IrBasicBlock {
+        label: "end".to_string(),
+        ops: vec![],
+        terminator: IrTerminator::Return(Some(result)),
+    };
+
+    let mut prog = IrProgram {
+        functions: vec![IrFunction {
+            name: "and_like".to_string(),
+            blocks: vec![entry, right, false_path, end],
+            locals: vec![],
+            param_count: 0,
+            has_return: true,
+            source_span: Span::new(0, 0, 0),
+        }],
+        globals: vec![],
+        rom_data: vec![],
+        states: vec![],
+        start_state: String::new(),
+    };
+
+    optimize(&mut prog);
+
+    // Find the false_path block post-optimization and verify that
+    // its LoadImm(result, 0) is still there.
+    let func = &prog.functions[0];
+    let fp = func
+        .blocks
+        .iter()
+        .find(|b| b.label == "false_path")
+        .expect("false_path block should survive as reachable");
+    assert!(
+        fp.ops
+            .iter()
+            .any(|op| matches!(op, IrOp::LoadImm(t, 0) if *t == result)),
+        "false-path LoadImm(result, 0) must survive optimization; block ops were {:?}",
+        fp.ops
+    );
+}

tests/emulator/goldens/platformer.audio.hash

@@ -1 +1 @@
-7c8e60e4 132084
+a00949db 132084