Add bench_field_mul_interleaved benchmark

src/bin/bench_field_mul.rs

@@ -1,3 +1,5 @@
+//! Performs a single exponentiation.
+
 use std::time::Instant;
 
 use plonky2::field::crandall_field::CrandallField;
@@ -5,17 +7,19 @@ use plonky2::field::field::Field;
 
 type F = CrandallField;
 
+const EXPONENT: usize = 1000000000;
+
 fn main() {
-    let m = F::from_canonical_u64(12345678901234567890);
-    let mut x = F::ONE;
+    let base = F::rand();
+    let mut state = F::ONE;
+
     let start = Instant::now();
-    let num_muls = 2000000000;
-    for _ in 0..num_muls {
-        x *= m;
+    for _ in 0..EXPONENT {
+        state *= base;
     }
     let duration = start.elapsed();
 
-    println!("result {:?}", x);
-    println!("took {:?}", duration);
-    println!("avg {:?}ns", duration.as_secs_f64() * 1e9 / (num_muls as f64));
+    println!("Result: {:?}", state);
+    println!("Average field mul: {:?}ns",
+             duration.as_secs_f64() * 1e9 / EXPONENT as f64);
 }

src/bin/bench_field_mul_interleaved.rs

@@ -0,0 +1,33 @@
+//! Performs several exponentiations in an interleaved loop, to enable parallelism on the core.
+
+use std::time::Instant;
+
+use plonky2::field::crandall_field::CrandallField;
+use plonky2::field::field::Field;
+
+type F = CrandallField;
+
+/// The number of exponentiations to perform in parallel.
+const WIDTH: usize = 6;
+
+const EXPONENT: usize = 1000000000;
+
+fn main() {
+    let mut bases = [F::ZERO; WIDTH];
+    for i in 0..WIDTH {
+        bases[i] = F::rand();
+    }
+    let mut state = [F::ONE; WIDTH];
+
+    let start = Instant::now();
+    for _ in 0..EXPONENT {
+        for i in 0..WIDTH {
+            state[i] *= bases[i];
+        }
+    }
+    let duration = start.elapsed();
+
+    println!("Result: {:?}", state);
+    println!("Average field mul: {:?}ns",
+             duration.as_secs_f64() * 1e9 / (WIDTH * EXPONENT) as f64);
+}