More rodinia optimization

• Add option to const-inline to not inline collections.
• Add clean-up pass to predication to rewrite boolean selects to and/or ops.
• Misc. scheduling work on rodinia.

juno_samples/rodinia/srad/src/gpu.sch

-gvn(*);
-dce(*);
+macro simpl!(X) {
+  ccp(X);
+  simplify-cfg(X);
+  lift-dc-math(X);
+  gvn(X);
+  phi-elim(X);
+  dce(X);
+  infer-schedules(X);
+}
+
 phi-elim(*);
-dce(*);
+let sum_loop = outline(srad@loop1);
+let main_loops = outline(srad@loop2 | srad@loop3);
+gpu(main_loops, extract, compress);
+simpl!(*);
+const-inline[true](*);
 crc(*);
-dce(*);
 slf(*);
-dce(*);
-
-let auto = auto-outline(srad);
-gpu(auto.srad);
-
-inline(auto.srad);
-inline(auto.srad);
-delete-uncalled(*);
+write-predication(*);
+simpl!(*);
+predication(*);
+simpl!(*);
+predication(*);
+simpl!(*);
+fixpoint {
+  forkify(*);
+  fork-guard-elim(*);
+  fork-coalesce(*);
+}
+simpl!(*);
+reduce-slf(*);
+simpl!(*);
+array-slf(*);
+simpl!(*);
+slf(*);
+simpl!(*);
 
-sroa[false](auto.srad);
-dce(*);
-float-collections(*);
-dce(*);
+fork-dim-merge(sum_loop);
+simpl!(sum_loop);
+fork-tile[32, 0, false, true](sum_loop);
+let out = fork-split(sum_loop);
+clean-monoid-reduces(sum_loop);
+simpl!(sum_loop);
+let fission = fork-fission[out.srad_0.fj0](sum_loop);
+simpl!(sum_loop);
+fork-tile[32, 0, false, true](fission.srad_0.fj_bottom);
+let out = fork-split(fission.srad_0.fj_bottom);
+clean-monoid-reduces(sum_loop);
+simpl!(sum_loop);
+let top = outline(fission.srad_0.fj_top);
+let bottom = outline(out.srad_0.fj0);
+gpu(top, bottom);
+ip-sroa(*);
+sroa(*);
+simpl!(*);
 
 gcm(*);
-