Need LTE in loop-bound-canon for backprop

juno_samples/rodinia/backprop/src/backprop.jn

@@ -4,7 +4,7 @@ fn squash(x: f32) -> f32 {
 }
 
 fn layer_forward<n, m: usize>(vals: f32[n + 1], weights: f32[n + 1, m + 1]) -> f32[m + 1] {
-  let result : f32[m + 1];
+  @res let result : f32[m + 1];
   result[0] = 1.0;
 
   for j in 1..=m {

juno_samples/rodinia/backprop/src/cpu.sch

-gvn(*);
-dce(*);
-phi-elim(*);
-dce(*);
-crc(*);
-dce(*);
-slf(*);
-dce(*);
+macro simpl!(X) {
+  ccp(X);
+  simplify-cfg(X);
+  lift-dc-math(X);
+  gvn(X);
+  phi-elim(X);
+  dce(X);
+  infer-schedules(X);
+}
 
-let auto = auto-outline(backprop);
-cpu(auto.backprop);
-
-inline(auto.backprop);
-inline(auto.backprop);
+simpl!(*);
+inline(layer_forward);
 delete-uncalled(*);
 
-sroa[true](*);
-dce(*);
-float-collections(*);
-reuse-products(*);
-dce(*);
+no-memset(layer_forward@res);
+lift-dc-math(*);
+loop-bound-canon(*);
+fixpoint {
+  forkify(*);
+  fork-guard-elim(*);
+  fork-coalesce(*);
+}
 
 gcm(*);