diff --git a/src/ion/fast_alloc.rs b/src/ion/fast_alloc.rs
index d20aa9c..dea22a8 100644
--- a/src/ion/fast_alloc.rs
+++ b/src/ion/fast_alloc.rs
@@ -931,6 +931,9 @@ fn handle_out_block_params<'a, F: Function>(
                 in_slot,
                 in_vreg
             );
+            if out_slot == in_slot {
+                continue;
+            }
             state.edits.push((
                 ProgPoint::before(last_inst),
                 Edit::Move {
@@ -975,7 +978,6 @@ fn handle_out_block_params<'a, F: Function>(
                 state.vregs[in_vreg.vreg()].def_block = Some(succ);
 
                 // TODO: if out_vreg dies at this edge, we could reuse its stack slot
-                // TODO: we should also be able to reuse the slot if the successor only has one predecessor (us); check with AE
                 let mut no_alias = false;
                 if !vregs_passed.contains(&out_vreg) {
                     let mut alloced = false;
@@ -995,6 +997,14 @@ fn handle_out_block_params<'a, F: Function>(
                         continue;
                     }
                     vregs_passed.push(out_vreg);
+
+                    if !state.liveouts[block.index()].get(out_vreg.vreg()) {
+                        // we can reuse the stack slot since the variable dies
+                        state.vregs[in_vreg.vreg()].slot_idx =
+                            Some(state.vregs[out_vreg.vreg()].slot_idx.unwrap());
+                        continue;
+                    }
+
                     no_alias = true;
                 }
 
@@ -1126,6 +1136,7 @@ impl BlockBitmap {
 
 // currently, copy from liveranges.rs
 // don't inline for better perf stats
+// this is currently very expensive, takes 3% of 10.2% total time for the register allocator
 #[inline(never)]
 fn calc_live_bitmaps<'a, F: Function>(
     state: &mut FastAllocState<'a, F>,