Fix backprop and matmul cpu schedules

Whoops.

juno_samples/matmul/src/matmul.sch

@@ -25,6 +25,10 @@ macro forkify!(X) {
   }
 }
 
+macro fork-chunk![n](X) {
+  fork-tile[n, 0, false, false](X);
+}
+
 macro fork-tile![n](X) {
   fork-tile[n, 0, false, true](X);
 }
@@ -66,8 +70,8 @@ if feature("cuda") {
 
   // Parallelize by computing output array as 16 chunks
   let par = matmul@outer \ matmul@inner;
-  fork-tile![4](par);
-  let (outer, inner, _) = fork-reshape[[1, 3], [0], [2]](par);
+  fork-chunk![4](par);
+  let (outer, inner, _) = fork-reshape[[0, 2], [1], [3]](par);
   parallelize!(outer \ inner);
 
   let body = outline(inner);