BitZ/single_threaded.patch at main · worldfnd/BitZ · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
diff --git a/crates/gkr/src/lib.rs b/crates/gkr/src/lib.rs
index 89a1bf0..bb364ff 100644
--- a/crates/gkr/src/lib.rs
+++ b/crates/gkr/src/lib.rs
@@ -87,11 +87,11 @@ fn prove_layer(
         let r = ps.verifier_message();
         next_point.push_back(r);

-        lo_l.par_iter_mut()
-            .zip(lo_r.par_iter_mut())
-            .zip(hi_l.par_iter())
-            .zip(hi_r.par_iter())
-            .with_min_len(PARALLEL_MIN_LANES)
+        lo_l.iter_mut()
+            .zip(lo_r.iter_mut())
+            .zip(hi_l.iter())
+            .zip(hi_r.iter())
+            // .with_min_len(PARALLEL_MIN_LANES)
             .for_each(|(((l_lo, r_lo), &l_hi), &r_hi)| {
                 // Recalculating delta rather than writing it in the top half of the array in the previous loop to save on writes.
                 // No benchmarking has been done to check the difference.
@@ -161,16 +161,16 @@ fn reduce_round<const SEND_ONE: bool>(
     hi_r: &[Field],
     eq: &[Field],
 ) -> (Wide256, Wide256) {
-    lo_l.par_chunks_exact(2)
-        .zip(lo_r.par_chunks_exact(2))
-        .zip(hi_l.par_chunks_exact(2))
-        .zip(hi_r.par_chunks_exact(2))
-        .zip(eq.par_chunks_exact(2))
+    lo_l.chunks_exact(2)
+        .zip(lo_r.chunks_exact(2))
+        .zip(hi_l.chunks_exact(2))
+        .zip(hi_r.chunks_exact(2))
+        .zip(eq.chunks_exact(2))
         // Chunked by 2, so the element-count threshold below is halved in
         // terms of chunks.
-        .with_min_len(PARALLEL_MIN_LANES / 2)
+        // .with_min_len(PARALLEL_MIN_LANES / 2)
         .fold(
-            || (Wide256::zero(), Wide256::zero()),
+            (Wide256::zero(), Wide256::zero()),
             |(mut sum_endpoint, mut sum_inf), ((((l_lo, r_lo), l_hi), r_hi), e)| {
                 let (d_l0, d_r0) = (l_hi[0] - l_lo[0], r_hi[0] - r_lo[0]);
                 let (d_l1, d_r1) = (l_hi[1] - l_lo[1], r_hi[1] - r_lo[1]);
@@ -197,10 +197,10 @@ fn reduce_round<const SEND_ONE: bool>(
                 (sum_endpoint, sum_inf)
             },
         )
-        .reduce(
-            || (Wide256::zero(), Wide256::zero()),
-            |(a0, ainf), (b0, binf)| (a0 + b0, ainf + binf),
-        )
+    // .reduce(
+    //     || (Wide256::zero(), Wide256::zero()),
+    //     |(a0, ainf), (b0, binf)| (a0 + b0, ainf + binf),
+    // )
 }

 /// `eq(r, z) = r*z + (1 - r)*(1 - z)`. Expanding gives