Merge pull request #23 from Nicoshev/rapidhashV2.2

Upgrade to V2.2

Improves speed by about 10% for sizes less than 4.

Introduces RAPIDHASH_SUPER_COMPACT. This macro will further reduce code size, at the expense of reducing speed for sizes >= 96.

Author: Nicoshev

rapidhash.h

@@ -1,5 +1,5 @@
 /*
- * rapidhash V2.1 - Very fast, high quality, platform-independent hashing algorithm.
+ * rapidhash V2.2 - Very fast, high quality, platform-independent hashing algorithm.
  * Copyright (C) 2025 Nicolas De Carli
  *
  * Based on 'wyhash', by Wang Yi <godspeed_china@yeah.net>
@@ -57,7 +57,6 @@
  # define RAPIDHASH_ALWAYS_INLINE inline
  #endif
 
-  
  #ifdef __cplusplus
  # define RAPIDHASH_NOEXCEPT noexcept
  # define RAPIDHASH_CONSTEXPR constexpr
@@ -78,15 +77,24 @@
   *  Small input speed macro, improves small input speed at the expense of sacrificing large input speed.
   *
   *  RAPIDHASH_UNROLLED: Normal behavior, max large input speed.
-  *  RAPIDHASH_COMPACT: Improved small input speed.
+  *  RAPIDHASH_COMPACT: Improved small input speed and reduced code size.
+  *
+  *  RAPIDHASH_SUPER_COMPACT: Almost always decreases speed for sizes > 95, but will further reduce code size.
   */
+  #ifdef RAPIDHASH_SUPER_COMPACT
+  # ifndef RAPIDHASH_COMPACT
+  #  define RAPIDHASH_COMPACT
+  # endif
+  #endif
   #ifndef RAPIDHASH_COMPACT
   # define RAPIDHASH_UNROLLED
   # define RAPIDHASH_UNLIKELY_BULK(x) _unlikely_(x)
+  # define RAPIDHASH_LIKELY_BULK(x) _likely_(x)
   #elif defined(RAPIDHASH_UNROLLED)
   # error "cannot define RAPIDHASH_COMPACT and RAPIDHASH_UNROLLED simultaneously."
   #else
   # define RAPIDHASH_UNLIKELY_BULK(x) (x)
+  # define RAPIDHASH_LIKELY_BULK(x) (x)
   #endif
 
  /*
@@ -230,18 +238,6 @@
  }
  #endif
 
- /*
-  *  Reads and combines 3 bytes of input.
-  *
-  *  @param p  Buffer to read from.
-  *  @param k  Length of @p, in bytes.
-  *
-  *  Always reads and combines 3 bytes from memory.
-  *  Guarantees to read each buffer position at least once.
-  *
-  *  Returns a 64-bit value containing all three bytes read.
-  */
-  RAPIDHASH_INLINE_CONSTEXPR uint64_t rapid_readSmall(const uint8_t *p, size_t k) RAPIDHASH_NOEXCEPT { return (((uint64_t)p[0])<<56)|(((uint64_t)p[k>>1])<<32)|p[k-1];}
  /*
   *  rapidhash main function.
   *
@@ -268,11 +264,11 @@ RAPIDHASH_INLINE_CONSTEXPR uint64_t rapidhash_internal(const void *key, size_t l
         b = rapid_read32(plast);
       }
     } else if (_likely_(len > 0)) {
-      a = rapid_readSmall(p, len);
-      b = 0;
+      a = (((uint64_t)p[0])<<56)|p[len-1];
+      b = p[len>>1];
     } else
       a = b = 0;
-  } else if (_likely_(len <= 64)) {
+  } else if (len <= 64) {
     seed = rapid_mix(rapid_read64(p) ^ secret[0], rapid_read64(p + 8) ^ seed);
     if (len > 32) {
       seed = rapid_mix(rapid_read64(p + 16) ^ secret[1], rapid_read64(p + 24) ^ seed);
@@ -329,6 +325,7 @@ RAPIDHASH_INLINE_CONSTEXPR uint64_t rapidhash_internal(const void *key, size_t l
       i -= 112;
     }
 #endif
+#ifndef RAPIDHASH_SUPER_COMPACT
     if (_likely_(i >= 48)) {
       seed = rapid_mix(rapid_read64(p) ^ secret[0], rapid_read64(p + 8) ^ seed);
       see1 = rapid_mix(rapid_read64(p + 16) ^ secret[1], rapid_read64(p + 24) ^ see1);
@@ -343,13 +340,22 @@ RAPIDHASH_INLINE_CONSTEXPR uint64_t rapidhash_internal(const void *key, size_t l
         i -= 48;
       }
     }
+#else
+    while (_likely_(i >= 48)) {
+      seed = rapid_mix(rapid_read64(p) ^ secret[0], rapid_read64(p + 8) ^ seed);
+      see1 = rapid_mix(rapid_read64(p + 16) ^ secret[1], rapid_read64(p + 24) ^ see1);
+      see2 = rapid_mix(rapid_read64(p + 32) ^ secret[2], rapid_read64(p + 40) ^ see2);
+      p += 48;
+      i -= 48;
+    }
+#endif
     see3 ^= see4;
     see5 ^= see6;
     seed ^= see1;
     see3 ^= see2;
     seed ^= see5;
     seed ^= see3;
-    if ((i > 16)) {
+    if (i > 16) {
       seed = rapid_mix(rapid_read64(p) ^ secret[2], rapid_read64(p + 8) ^ seed);
       if (RAPIDHASH_UNLIKELY_BULK(i > 32))
         seed = rapid_mix(rapid_read64(p + 16) ^ secret[2], rapid_read64(p + 24) ^ seed);
@@ -390,4 +396,3 @@ RAPIDHASH_INLINE_CONSTEXPR uint64_t rapidhash_withSeed(const void *key, size_t l
 RAPIDHASH_INLINE_CONSTEXPR uint64_t rapidhash(const void *key, size_t len) RAPIDHASH_NOEXCEPT {
   return rapidhash_withSeed(key, len, 0);
 }
-