More optimizations

• A lot of misc. optimization work, notably parallelization of Rodinia CPU schedules.
• Support >32 tiles in GPU backend.
• Tweak some transformations.

juno_samples/rodinia/backprop/src/backprop.jn

@@ -69,8 +69,8 @@ fn adjust_weights<n, m: usize>(
   weights: f32[n + 1, m + 1],
   prev_weights: f32[n + 1, m + 1]
 ) -> f32[n + 1, m + 1], f32[n + 1, m + 1] {
-  for j in 1..=m {
-    for k in 0..=n {
+  @outer_loop for j in 1..=m {
+    @inner_loop for k in 0..=n {
       let new_dw = ETA * delta[j] * vals[k] + MOMENTUM * prev_weights[k, j];
       weights[k, j] += new_dw;
       prev_weights[k, j] = new_dw;