perf(rust): add nsieve 3.rs (Wheel 8-of-30 Sieve)

bench/algorithm/nsieve/3.rs

@@ -0,0 +1,115 @@
+// The Computer Language Shootout
+// http://shootout.alioth.debian.org/
+// Wheel 8-of-30 bit-sieve Rust implementation by ndt0208
+
+const STEPS: [usize; 8] = [3, 2, 1, 2, 1, 2, 3, 1];
+
+fn nsieve(max: usize, bits: &mut [u64]) {
+    if max <= 2 {
+        println!("Primes up to {:8} {:8}", max, 0);
+        return;
+    }
+
+    let maxintsh = max >> 1;
+    let num_words = (maxintsh + 63) >> 6;
+    bits[..num_words].fill(0);
+    let ptr = bits.as_mut_ptr();
+
+    let qh = ((max as f64).sqrt() as usize) >> 1;
+
+    let mut step = 1;
+    let mut factorh = 3; // 7 >> 1
+
+    unsafe {
+        while factorh <= qh {
+            let w_idx = factorh >> 6;
+            let b_idx = factorh & 63;
+
+            if (*ptr.add(w_idx) & (1u64 << b_idx)) == 0 {
+                let factor = (factorh << 1) + 1;
+                let mut istep = step;
+                let mut i = (factor * factor) >> 1;
+
+                let s0 = factor * *STEPS.get_unchecked(istep);
+                let s1 = factor * *STEPS.get_unchecked((istep + 1) & 7);
+                let s2 = factor * *STEPS.get_unchecked((istep + 2) & 7);
+                let s3 = factor * *STEPS.get_unchecked((istep + 3) & 7);
+                let s4 = factor * *STEPS.get_unchecked((istep + 4) & 7);
+                let s5 = factor * *STEPS.get_unchecked((istep + 5) & 7);
+                let s6 = factor * *STEPS.get_unchecked((istep + 6) & 7);
+                let s7 = factor * *STEPS.get_unchecked((istep + 7) & 7);
+                let cycle = factor * 15;
+
+                while i + cycle < maxintsh {
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63); i += s0;
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63); i += s1;
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63); i += s2;
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63); i += s3;
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63); i += s4;
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63); i += s5;
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63); i += s6;
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63); i += s7;
+                }
+                while i < maxintsh {
+                    *ptr.add(i >> 6) |= 1u64 << (i & 63);
+                    i += factor * *STEPS.get_unchecked(istep);
+                    istep = (istep + 1) & 7;
+                }
+            }
+
+            factorh += *STEPS.get_unchecked(step);
+            step = (step + 1) & 7;
+        }
+
+        let mut count = (max > 2) as usize + (max > 3) as usize + (max > 5) as usize;
+        let mut factor = 7;
+        let cycle = 30;
+
+        while factor + cycle < max {
+            let h0 = factor >> 1; factor += 4;
+            let h1 = factor >> 1; factor += 2;
+            let h2 = factor >> 1; factor += 4;
+            let h3 = factor >> 1; factor += 2;
+            let h4 = factor >> 1; factor += 4;
+            let h5 = factor >> 1; factor += 6;
+            let h6 = factor >> 1; factor += 2;
+            let h7 = factor >> 1; factor += 6;
+
+            count += (((*ptr.add(h0 >> 6) >> (h0 & 63)) ^ 1) & 1) as usize;
+            count += (((*ptr.add(h1 >> 6) >> (h1 & 63)) ^ 1) & 1) as usize;
+            count += (((*ptr.add(h2 >> 6) >> (h2 & 63)) ^ 1) & 1) as usize;
+            count += (((*ptr.add(h3 >> 6) >> (h3 & 63)) ^ 1) & 1) as usize;
+            count += (((*ptr.add(h4 >> 6) >> (h4 & 63)) ^ 1) & 1) as usize;
+            count += (((*ptr.add(h5 >> 6) >> (h5 & 63)) ^ 1) & 1) as usize;
+            count += (((*ptr.add(h6 >> 6) >> (h6 & 63)) ^ 1) & 1) as usize;
+            count += (((*ptr.add(h7 >> 6) >> (h7 & 63)) ^ 1) & 1) as usize;
+        }
+
+        let mut cstep = 1;
+        while factor < max {
+            let half = factor >> 1;
+            count += (((*ptr.add(half >> 6) >> (half & 63)) ^ 1) & 1) as usize;
+            factor += *STEPS.get_unchecked(cstep) << 1;
+            cstep = (cstep + 1) & 7;
+        }
+
+        println!("Primes up to {:8} {:8}", max, count);
+    }
+}
+
+fn main() {
+    let n = std::env::args_os()
+        .nth(1)
+        .and_then(|s| s.into_string().ok())
+        .and_then(|n| n.parse().ok())
+        .unwrap_or(4);
+
+    let max_n = 10000 << n;
+    let max_words = ((max_n >> 1) + 63) >> 6;
+    let mut buffer = vec![0u64; max_words];
+
+    for i in 0..3 {
+        let m = 10000 << (n - i);
+        nsieve(m, &mut buffer);
+    }
+}

bench/bench_rust.yaml

@@ -81,6 +81,7 @@ problems:
     source:
       - 1.rs
       - 2.rs
+      - 3.rs
   - name: lru
     source:
       - 0-unsafe.rs