Fix permutation quality issue (#48)

* Fix permutation quality issue

* Add more quality tests

---------

Co-authored-by: Olivier Giniaux <[email protected]>

Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "gxhash"
 authors = ["Olivier Giniaux"]
-version = "2.3.0"
+version = "2.3.1"
 edition = "2021"
 description = "GxHash non-cryptographic algorithm"
 license = "MIT"

benches/quality/main.rs

@@ -1,4 +1,4 @@
-use std::{hash::{Hash, Hasher, BuildHasher}, iter};
+use std::hash::{Hash, Hasher, BuildHasher};
 use rand::Rng;
 use criterion::black_box;
 
@@ -60,11 +60,14 @@ fn bench_hasher_quality<B>(name: &str)
     check!(collisions_flipped_n_bits::<B, 256>(2));
 
     check!(collisions_powerset_bytes::<B>(&[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]));
+    check!(collisions_powerset_bytes::<B>(&[0, 1, 2, 4, 8, 16, 32, 64, 128]));
 
     check!(collisions_permuted_hasher_values::<B, u8>(&[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]));
     check!(collisions_permuted_hasher_values::<B, u32>(&[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]));
+    check!(collisions_permuted_hasher_values::<B, u32>(&[0, 1, 2, 4, 8, 16, 32, 64, 128, 256]));
 
     check!(collisions_powerset_hasher_values::<B, u32>(&[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]));
+    check!(collisions_powerset_hasher_values::<B, u32>(&[0, 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384]));
 }
 
 fn collisions_permuted_hasher_values<B, D>(data: &[impl Hash]) -> f64

src/gxhash/platform/arm_128.rs

@@ -73,6 +73,16 @@ pub unsafe fn compress_fast(a: int8x16_t, b: int8x16_t) -> int8x16_t {
     vreinterpretq_s8_u8(aes_encrypt(vreinterpretq_u8_s8(a), vreinterpretq_u8_s8(b)))
 }
 
+#[inline(always)]
+pub unsafe fn compress_1(a: int8x16_t, b: int8x16_t) -> int8x16_t {
+    let keys_1 = vld1q_u32([0xFC3BC28E, 0x89C222E5, 0xB09D3E21, 0xF2784542].as_ptr());
+
+    let mut bs = vreinterpretq_u8_s8(b);
+    bs = aes_encrypt(bs, vreinterpretq_u8_u32(keys_1));
+
+    vreinterpretq_s8_u8(aes_encrypt_last(vreinterpretq_u8_s8(a), bs))
+}
+
 #[inline(always)]
 // See https://blog.michaelbrase.com/2018/05/08/emulating-x86-aes-intrinsics-on-armv8-a
 unsafe fn aes_encrypt(data: uint8x16_t, keys: uint8x16_t) -> uint8x16_t {

src/gxhash/platform/x86_128.rs

@@ -66,6 +66,15 @@ pub unsafe fn compress_fast(a: State, b: State) -> State {
     _mm_aesenc_si128(a, b)
 }
 
+#[inline(always)]
+#[allow(overflowing_literals)]
+pub unsafe fn compress_1(a: State, b: State) -> State {
+    let keys_1 = _mm_set_epi32(0xF2784542, 0xB09D3E21, 0x89C222E5, 0xFC3BC28E);
+
+    let mut b = _mm_aesenc_si128(b, keys_1);
+    _mm_aesenclast_si128(a, b)
+}
+
 #[inline(always)]
 #[allow(overflowing_literals)]
 pub unsafe fn finalize(hash: State) -> State {

src/gxhash/platform/x86_256.rs

@@ -66,6 +66,15 @@ pub unsafe fn compress_fast(a: State, b: State) -> State {
     return _mm256_aesenc_epi128(a, b);
 }
 
+#[inline(always)]
+#[allow(overflowing_literals)]
+pub unsafe fn compress_1(a: State, b: State) -> State {
+    let keys_1 = _mm256_set_epi32(0xFC3BC28E, 0x89C222E5, 0xB09D3E21, 0xF2784542, 0x4155EE07, 0xC897CCE2, 0x780AF2C3, 0x8A72B781);
+
+    let mut b = _mm256_aesenc_epi128(b, keys_1);
+    return _mm256_aesenclast_epi128(a, b);
+}
+
 #[inline(always)]
 #[allow(overflowing_literals)]
 pub unsafe fn finalize(hash: State) -> State {

src/hasher.rs

@@ -98,7 +98,7 @@ macro_rules! write {
         #[inline]
         fn $name(&mut self, value: $type) {
             self.state = unsafe {
-                compress_fast($load(value), self.state)
+                compress_1($load(value), self.state)
             };
         }
     }
@@ -115,7 +115,7 @@ impl Hasher for GxHasher {
 
     #[inline]
     fn write(&mut self, bytes: &[u8]) {
-        self.state = unsafe { compress_fast(compress_all(bytes), self.state) };
+        self.state = unsafe { compress_1(compress_all(bytes), self.state) };
     }
 
     write!(write_u8, u8, load_u8);