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.

juno_samples/cava/src/cava.jn

@@ -35,20 +35,6 @@ fn scale<row : usize, col : usize>(input : u8[CHAN, row, col]) -> f32[CHAN, row,
   return res;
 }
 
-fn descale<row : usize, col : usize>(input : f32[CHAN, row, col]) -> u8[CHAN, row, col] {
-  let res : u8[CHAN, row, col];
-
-  for chan = 0 to CHAN {
-    for r = 0 to row {
-      for c = 0 to col {
-        res[chan, r, c] = min!::<f32>(max!::<f32>(input[chan, r, c] * 255, 0), 255) as u8;
-      }
-    }
-  }
-
-  return res;
-}
-
 fn demosaic<row : usize, col : usize>(input : f32[CHAN, row, col]) -> f32[CHAN, row, col] {
   @res2 let res : f32[CHAN, row, col];
 
@@ -184,7 +170,7 @@ fn gamut<row : usize, col : usize, num_ctrl_pts : usize>(
 
 fn tone_map<row : usize, col:usize>
   (input : f32[CHAN, row, col], tone_map : f32[256, CHAN]) -> f32[CHAN, row, col] {
-  let result : f32[CHAN, row, col];
+  @res1 let result : f32[CHAN, row, col];
 
   for chan = 0 to CHAN {
     for r = 0 to row {
@@ -198,6 +184,20 @@ fn tone_map<row : usize, col:usize>
   return result;
 }
 
+fn descale<row : usize, col : usize>(input : f32[CHAN, row, col]) -> u8[CHAN, row, col] {
+  @res2 let res : u8[CHAN, row, col];
+
+  for chan = 0 to CHAN {
+    for r = 0 to row {
+      for c = 0 to col {
+        res[chan, r, c] = min!::<f32>(max!::<f32>(input[chan, r, c] * 255, 0), 255) as u8;
+      }
+    }
+  }
+
+  return res;
+}
+
 #[entry]
 fn cava<r, c, num_ctrl_pts : usize>(
   input : u8[CHAN, r, c],
@@ -212,7 +212,7 @@ fn cava<r, c, num_ctrl_pts : usize>(
   @fuse2 let denosd = denoise::<r, c>(demosc);
   @fuse3 let transf = transform::<r, c>(denosd, TsTw);
   @fuse4 let gamutd = gamut::<r, c, num_ctrl_pts>(transf, ctrl_pts, weights, coefs);
-  let tonemd = tone_map::<r, c>(gamutd, tonemap);
-  let dscald = descale::<r, c>(tonemd);
+  @fuse5 let tonemd = tone_map::<r, c>(gamutd, tonemap);
+  @fuse5 let dscald = descale::<r, c>(tonemd);
   return dscald;
 }