Set up cava benchmark

• Set up cava benchmark w/ criterion.
• GPU schedule for cava.
• Convert a bunch of benchmarks from using hercules refs to use hercules boxes.
• Define AlignedAlloc global allocator in hercules_rt that aligns everything to 32 bytes.
• Bump nom version.

hercules_samples/matmul/src/main.rs

@@ -10,11 +10,11 @@ juno_build::juno!("matmul");
 fn main() {
     async_std::task::block_on(async {
         const I: usize = 256;
-        const J: usize = 8; // hardcoded constant in matmul.hir
+        const J: usize = 64;
         const K: usize = 128;
-        let a: Box<[f32]> = (0..I * J).map(|_| random::<f32>()).collect();
-        let b: Box<[f32]> = (0..J * K).map(|_| random::<f32>()).collect();
-        let mut correct_c: Box<[f32]> = (0..I * K).map(|_| 0.0).collect();
+        let a: Box<[i32]> = (0..I * J).map(|_| random::<i32>() % 100).collect();
+        let b: Box<[i32]> = (0..J * K).map(|_| random::<i32>() % 100).collect();
+        let mut correct_c: Box<[i32]> = (0..I * K).map(|_| 0).collect();
         for i in 0..I {
             for k in 0..K {
                 for j in 0..J {
@@ -22,13 +22,11 @@ fn main() {
                 }
             }
         }
-        let a = HerculesImmBox::from(&a as &[f32]);
-        let b = HerculesImmBox::from(&b as &[f32]);
+        let a = HerculesImmBox::from(a.as_ref());
+        let b = HerculesImmBox::from(b.as_ref());
         let mut r = runner!(matmul);
-        let mut c = HerculesMutBox::from(r.run(I as u64, J as u64, K as u64, a.to(), b.to()).await);
-        for (calc, correct) in zip(c.as_slice().into_iter().map(|x: &mut f32| *x), correct_c) {
-            assert!((calc - correct).abs() < 0.0001, "{} != {}", calc, correct);
-        }
+        let mut c: HerculesMutBox<i32> = HerculesMutBox::from(r.run(I as u64, J as u64, K as u64, a.to(), b.to()).await);
+        assert_eq!(c.as_slice(), correct_c.as_ref());
     });
 }