Fork fission bufferize

• Refactor fork fission, add fork fission bufferized version.
• Add test that uses fork fission to create extra forks.

hercules_opt/src/forkify.rs

@@ -10,8 +10,8 @@ use hercules_ir::*;
 
 use crate::*;
 
-/* 
- * TODO: Forkify currently makes a bunch of small edits - this needs to be 
+/*
+ * TODO: Forkify currently makes a bunch of small edits - this needs to be
  * changed so that every loop that gets forkified corresponds to a single edit
  * + sub-edits. This would allow us to run forkify on a subset of a function.
  */
@@ -350,6 +350,27 @@ pub fn forkify_loop(
             };
             let reduce_id = edit.add_node(reduce);
 
+            if (!edit.get_node(init).is_reduce()
+                && edit.get_schedule(init).contains(&Schedule::ParallelReduce))
+                || (!edit.get_node(continue_latch).is_reduce()
+                    && edit
+                        .get_schedule(continue_latch)
+                        .contains(&Schedule::ParallelReduce))
+            {
+                edit = edit.add_schedule(reduce_id, Schedule::ParallelReduce)?;
+            }
+            if (!edit.get_node(init).is_reduce()
+                && edit
+                    .get_schedule(init)
+                    .contains(&Schedule::TightAssociative))
+                || (!edit.get_node(continue_latch).is_reduce()
+                    && edit
+                        .get_schedule(continue_latch)
+                        .contains(&Schedule::TightAssociative))
+            {
+                edit = edit.add_schedule(reduce_id, Schedule::TightAssociative)?;
+            }
+
             edit = edit.replace_all_uses_where(phi, reduce_id, |usee| *usee != reduce_id)?;
             edit = edit.replace_all_uses_where(continue_latch, reduce_id, |usee| {
                 !loop_nodes.contains(usee) && *usee != reduce_id
@@ -417,12 +438,8 @@ pub fn analyze_phis<'a>(
     phis: &'a [NodeID],
     loop_nodes: &'a HashSet<NodeID>,
 ) -> impl Iterator<Item = LoopPHI> + 'a {
-
-    // We are also moving the phi from the top of the loop (the header),
-    // to the very end (the join). If there are uses of the phi somewhere in the loop,
-    // then they may try to use the phi (now a reduce) before it hits the join. 
-    // Find data cycles within the loop of this phi, 
-    // Start from the phis loop_continue_latch, and walk its uses until we find the original phi. 
+    // Find data cycles within the loop of this phi,
+    // Start from the phis loop_continue_latch, and walk its uses until we find the original phi.
 
     phis.into_iter().map(move |phi| {
         let stop_on: HashSet<NodeID> = editor
@@ -471,7 +488,7 @@ pub fn analyze_phis<'a>(
             .unwrap();
 
         let loop_continue_latch = editor.node(phi).try_phi().unwrap().1[continue_idx];
-        
+
         let uses = walk_all_uses_stop_on(loop_continue_latch, editor, stop_on.clone());
         let users = walk_all_users_stop_on(*phi, editor, stop_on.clone());
 
@@ -524,7 +541,7 @@ pub fn analyze_phis<'a>(
             // If some other node in the cycle is used, there is not a valid node to assign it after making the cycle a reduce.
             if intersection
                 .iter()
-                .filter(|node| **node != loop_continue_latch )
+                .filter(|node| **node != loop_continue_latch)
                 .any(|data_node| {
                     editor
                         .get_users(*data_node)