From f95e869848e0b2a66cb0facc75062778c9881cb9 Mon Sep 17 00:00:00 2001
From: Nathan Vegdahl <cessen@cessen.com>
Date: Sat, 23 Jul 2022 13:24:24 -0700
Subject: [PATCH] Give Owen scramble functions their own hash.

This lets us move the seeding overhead outside the main loop,
which in turn lets us avoid taking it every round.
---
 src/scramble.rs | 44 ++++++++++++++++++++++++++++++++++++++------
 1 file changed, 38 insertions(+), 6 deletions(-)

diff --git a/src/scramble.rs b/src/scramble.rs
index 881a96f..78df767 100644
--- a/src/scramble.rs
+++ b/src/scramble.rs
@@ -1,12 +1,16 @@
-use crate::hash::hash_u32;
-
 /// Performs a base-2 Owen scramble on an integer.
 pub fn owen2(n: u32, seed: u32) -> u32 {
+    // Multiply by a large random prime and xor by a random number.
+    // This is to ensure that the seed doesn't behave poorly with
+    // e.g. incrementing parameters, and also that zero doesn't
+    // map to zero in the hash function.
+    let seed = seed.wrapping_mul(0xe8559dcb) ^ 0x372fcdb9;
+
     let mut result = n;
 
     for i in 0..32 {
-        let mask = (!0 << 1) << i;
-        result ^= hash_u32(n & mask, seed) & (1 << i);
+        let mask = (!0 << 1) << i; // Two shifts to avoid undefined overflow.
+        result ^= hash((n & mask) ^ seed) & (1 << i);
     }
 
     result
@@ -25,6 +29,7 @@ pub fn owen2(n: u32, seed: u32) -> u32 {
 /// The implementation below is a full proper base-4 Owen scramble,
 /// requiring only a 24-byte lookup table.
 pub fn owen4(n: u32, seed: u32) -> u32 {
+    // Bit-packed permutation table.
     const PERMUTATION_TABLE: [u8; 24] = [
         0 | (1 << 2) | (2 << 4) | (3 << 6), // [0, 1, 2, 3],
         0 | (1 << 2) | (3 << 4) | (2 << 6), // [0, 1, 3, 2],
@@ -52,11 +57,23 @@ pub fn owen4(n: u32, seed: u32) -> u32 {
         3 | (2 << 2) | (1 << 4) | (0 << 6), // [3, 2, 1, 0],
     ];
 
+    // Multiply by a large random prime and xor by a random number.
+    // This is to ensure that the seed doesn't behave poorly with
+    // e.g. incrementing parameters, and also that zero doesn't
+    // map to zero in the hash function.
+    let seed = seed.wrapping_mul(0xe8559dcb) ^ 0x372fcdb9;
+
     let mut result = 0;
 
     for i in 0..16 {
-        let mask = (!0 << 2) << (i * 2);
-        let perm_entry = PERMUTATION_TABLE[(hash_u32(n & mask, seed) % 24) as usize];
+        let mask = (!0 << 2) << (i * 2); // Two shifts to avoid undefined overflow.
+        let perm_entry = PERMUTATION_TABLE[
+            // The xor with `i` is to ensure runs of zeros in `n` still
+            // result in different shuffles on each iteration.  `i` is
+            // shifted to avoid interacting poorly with an incrementing
+            // `n`.
+            (hash((n & mask) ^ seed ^ (i << 16)) % 24) as usize
+        ];
         let perm_cell_idx = ((n >> (i * 2)) & 0b11) as usize;
 
         result |= (((perm_entry >> (perm_cell_idx * 2)) & 0b11) as u32) << (i * 2);
@@ -64,3 +81,18 @@ pub fn owen4(n: u32, seed: u32) -> u32 {
 
     result
 }
+
+//-------------------------------------------------------------
+
+/// Fast bit-mixing hash for use in the functions above.
+#[inline(always)]
+pub fn hash(mut n: u32) -> u32 {
+    // From https://github.com/skeeto/hash-prospector
+    n ^= n >> 16;
+    n = n.wrapping_mul(0x21f0aaad);
+    n ^= n >> 15;
+    n = n.wrapping_mul(0xd35a2d97);
+    n ^= n >> 15;
+
+    n
+}