Some Cava optimization

• Optimize Cava's CPU schedule.
• Add fork-fusion pass for fusing unrelated but adjacent fork-joins.
• Add intrinsic functions to math expressions.
• Add print pass for debugging schedules.
• Avoid some edges in dot output to help little xdot out.
• TODO: need more relaxed bounds for forkify to parallelize demosaic.

hercules_cg/src/rt.rs

@@ -533,7 +533,7 @@ impl<'a> RTContext<'a> {
                 {
                     write!(block, "backing_{}.byte_add(", device.name())?;
                     self.codegen_dynamic_constant(offset, block)?;
-                    write!(block, "), ")?
+                    write!(block, " as usize), ")?
                 }
                 for dc in dynamic_constants {
                     self.codegen_dynamic_constant(*dc, block)?;