feat: cache for bit_reverse

starky/src/fft_p.rs

@@ -96,9 +96,8 @@ pub fn bit_reverse<F: FieldExtension>(
     let n = 1 << nbits;
     let ris = BRs(0, n, nbits); // move it outside the loop. obtain it from cache.
 
-    // todo parallel. Does the buffdst
     let len = n * n_pols;
-    assert_eq!(len, buffdst.len()); // is ok?
+    assert_eq!(len, buffdst.len());
     for j in 0..len {
         let i = j / n_pols;
         let k = j % n_pols;
@@ -113,9 +112,10 @@ pub fn interpolate_bit_reverse<F: FieldExtension>(
     nbits: usize,
 ) {
     let n = 1 << nbits;
+    let ris = BRs(0, n, nbits); // move it outside the loop. obtain it from cache.
+
     for i in 0..n {
-        let ri = BR(i, nbits);
-        let rii = (n - ri) % n;
+        let rii = (n - ris[i]) % n;
         for k in 0..n_pols {
             buffdst[i * n_pols + k] = buffsrc[rii * n_pols + k];
         }
@@ -130,12 +130,15 @@ pub fn inv_bit_reverse<F: FieldExtension>(
 ) {
     let n = 1 << nbits;
     let n_inv = F::inv(&F::from(n));
-    for i in 0..n {
-        let ri = BR(i, nbits);
-        let rii = (n - ri) % n;
-        for p in 0..n_pols {
-            buffdst[i * n_pols + p] = buffsrc[rii * n_pols + p] * n_inv;
-        }
+    let ris = BRs(0, n, nbits); // move it outside the loop. obtain it from cache.
+
+    let len = n * n_pols;
+    assert_eq!(len, buffdst.len());
+    for j in 0..len {
+        let i = j / n_pols;
+        let k = j % n_pols;
+        let rii = (n - ris[i]) % n;
+        buffdst[j] = buffsrc[rii * n_pols + k] * n_inv;
     }
 }