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

@@ -22,7 +22,7 @@ fn medianMatrix<a : number, rows, cols : usize>(m : a[rows, cols]) -> a {
 const CHAN : u64 = 3;
 
 fn scale<row : usize, col : usize>(input : u8[CHAN, row, col]) -> f32[CHAN, row, col] {
-  @const let res : f32[CHAN, row, col];
+  @res1 let res : f32[CHAN, row, col];
 
   for chan = 0 to CHAN {
     for r = 0 to row {
@@ -50,7 +50,7 @@ fn descale<row : usize, col : usize>(input : f32[CHAN, row, col]) -> u8[CHAN, ro
 }
 
 fn demosaic<row : usize, col : usize>(input : f32[CHAN, row, col]) -> f32[CHAN, row, col] {
-  let res : f32[CHAN, row, col];
+  @res2 let res : f32[CHAN, row, col];
 
   for r = 1 to row-1 {
     for c = 1 to col-1 {
@@ -129,7 +129,7 @@ fn denoise<row : usize, col : usize>(input : f32[CHAN, row, col]) -> f32[CHAN, r
 fn transform<row : usize, col : usize>
   (input : f32[CHAN, row, col], tstw_trans : f32[CHAN, CHAN])
     -> f32[CHAN, row, col] {
-  let result : f32[CHAN, row, col];
+  @res let result : f32[CHAN, row, col];
 
   for chan = 0 to CHAN {
     for r = 0 to row {
@@ -152,11 +152,11 @@ fn gamut<row : usize, col : usize, num_ctrl_pts : usize>(
   weights  : f32[num_ctrl_pts, CHAN],
   coefs : f32[4, CHAN]
 ) -> f32[CHAN, row, col] {
-  let result : f32[CHAN, row, col];
-  let l2_dist : f32[num_ctrl_pts];
+  @res let result : f32[CHAN, row, col];
 
   for r = 0 to row {
     for c = 0 to col {
+      @l2 let l2_dist : f32[num_ctrl_pts];
       for cp = 0 to num_ctrl_pts {
         let v1 = input[0, r, c] - ctrl_pts[cp, 0];
         let v2 = input[1, r, c] - ctrl_pts[cp, 1];
@@ -210,8 +210,8 @@ fn cava<r, c, num_ctrl_pts : usize>(
   @fuse1 let scaled = scale::<r, c>(input);
   @fuse1 let demosc = demosaic::<r, c>(scaled);
   @fuse2 let denosd = denoise::<r, c>(demosc);
-  let transf = transform::<r, c>(denosd, TsTw);
-  let gamutd = gamut::<r, c, num_ctrl_pts>(transf, ctrl_pts, weights, coefs);
+  @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);
   return dscald;