added optional 16-bit look-up table

include/pearson.h

@@ -21,3 +21,5 @@ void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len);
 void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len);
 
 uint16_t pearson_hash_16 (const uint8_t *in, size_t len);
+
+void pearson_hash_init();

src/edge_utils.c

@@ -233,6 +233,8 @@ n2n_edge_t* edge_init(const tuntap_dev *dev, const n2n_edge_conf_t *conf, int *r
   eee->pending_peers  = NULL;
   eee->sup_attempts = N2N_EDGE_SUP_ATTEMPTS;
 
+  pearson_hash_init();
+
   if(eee->conf.compression == N2N_COMPRESSION_ID_LZO)
     if(lzo_init() != LZO_E_OK) {
       traceEvent(TRACE_ERROR, "LZO compression error");

src/pearson.c

@@ -17,7 +17,6 @@
  */
 
 // taken from https://github.com/Logan007/pearson
-
 // This is free and unencumbered software released into the public domain.
 
 #include <stddef.h>
@@ -25,6 +24,10 @@
 
 #include "pearson.h"
 
+// compile with 'LOW_MEM_FOOTPRINT' defined to make use of 256 byte look-up tabe only
+// otherwise, a 16-bit look-up table is used which allows considerably faster hashing
+// however, it needs to be generated by once calling pearson_hash_init() upfront
+// #define LOW_MEM_FOOTPRINT
 
 // table as in original paper "Fast Hashing of Variable-Length Text Strings" by Peter K. Pearson
 // as published in The Communications of the ACM  Vol.33, No.  6 (June 1990), pp. 677-680.
@@ -67,11 +70,16 @@ static const uint8_t t[256] ={
 0xd8, 0x83, 0x59, 0x15, 0x1c, 0x85, 0x25, 0x99, 0x95, 0x50, 0xaa, 0x44, 0x06, 0xa9, 0xea, 0x97 }; */
 
 
+#ifndef LOW_MEM_FOOTPRINT
+static uint16_t t16[65536]; // 16-bit look-up table
+#endif
+
 #define ROR64(x,r) (((x)>>(r))|((x)<<(64-(r))))
 
 
 void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) {
 
+  size_t i;
   /* initial values -  astonishingly, assembling using SHIFTs and ORs (in register)
    * works faster on well pipelined CPUs than loading the 64-bit value from memory.
    * however, there is one advantage to loading from memory: as we also store back to
@@ -79,10 +87,8 @@ void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) {
   uint8_t upper[8] = { 0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
   uint8_t lower[8] = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00 };
 
-  uint64_t *upper_hash_mask_ptr = (uint64_t*)&upper;
+  uint64_t upper_hash_mask = *(uint64_t*)&upper;
-  uint64_t *lower_hash_mask_ptr = (uint64_t*)&lower;
+  uint64_t lower_hash_mask = *(uint64_t*)&lower;
-  uint64_t upper_hash_mask = *upper_hash_mask_ptr;
-  uint64_t lower_hash_mask = *lower_hash_mask_ptr;
   uint64_t high_upper_hash_mask = upper_hash_mask + 0x1010101010101010;
   uint64_t high_lower_hash_mask = lower_hash_mask + 0x1010101010101010;
 
@@ -90,7 +96,6 @@ void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) {
   uint64_t lower_hash = 0;
   uint64_t high_upper_hash = 0;
   uint64_t high_lower_hash = 0;
-  size_t i;
 
   for (i = 0; i < len; i++) {
     // broadcast the character, xor into hash, make them different permutations
@@ -104,8 +109,9 @@ void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) {
     high_lower_hash ^= c ^ high_lower_hash_mask;
 
     // table lookup
-    uint8_t x;
     uint64_t h = 0;
+#ifdef LOW_MEM_FOOTPRINT // 256 byte look-up table ----------
+    uint8_t x;
     x = upper_hash; x = t[x]; upper_hash >>= 8; h |= x; h=ROR64(h,8);
     x = upper_hash; x = t[x]; upper_hash >>= 8; h |= x; h=ROR64(h,8);
     x = upper_hash; x = t[x]; upper_hash >>= 8; h |= x; h=ROR64(h,8);
@@ -148,6 +154,35 @@ void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) {
     x = high_lower_hash; x = t[x]; high_lower_hash >>= 8; h |= x; h=ROR64(h,8);
     x = high_lower_hash; x = t[x]; high_lower_hash >>= 8; h |= x; h=ROR64(h,8);
     high_lower_hash = h;
+#else // 16-bit look-up table -------------------------------
+    uint16_t x;
+    x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    upper_hash = h;
+
+    h = 0;
+    x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    lower_hash = h;
+
+    h = 0;
+    x = high_upper_hash; x = t16[x]; high_upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = high_upper_hash; x = t16[x]; high_upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = high_upper_hash; x = t16[x]; high_upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = high_upper_hash; x = t16[x]; high_upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    high_upper_hash = h;
+
+    h = 0;
+    x = high_lower_hash; x = t16[x]; high_lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = high_lower_hash; x = t16[x]; high_lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = high_lower_hash; x = t16[x]; high_lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = high_lower_hash; x = t16[x]; high_lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    high_lower_hash = h;
+#endif // LOW_MEM_FOOTPRINT ------
   }
   // store output
   uint64_t *o;
@@ -164,6 +199,7 @@ void pearson_hash_256 (uint8_t *out, const uint8_t *in, size_t len) {
 
 void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len) {
 
+  size_t i;
   /* initial values -  astonishingly, assembling using SHIFTs and ORs (in register)
    * works faster on well pipelined CPUs than loading the 64-bit value from memory.
    * however, there is one advantage to loading from memory: as we also store back to
@@ -171,14 +207,11 @@ void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len) {
   uint8_t upper[8] = { 0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
   uint8_t lower[8] = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00 };
 
-  uint64_t *upper_hash_mask_ptr = (uint64_t*)&upper;
+  uint64_t upper_hash_mask = *(uint64_t*)&upper;
-  uint64_t *lower_hash_mask_ptr = (uint64_t*)&lower;
+  uint64_t lower_hash_mask = *(uint64_t*)&lower;
-  uint64_t upper_hash_mask = *upper_hash_mask_ptr;
-  uint64_t lower_hash_mask = *lower_hash_mask_ptr;
 
   uint64_t upper_hash = 0;
   uint64_t lower_hash = 0;
-  size_t i;
 
   for (i = 0; i < len; i++) {
     // broadcast the character, xor into hash, make them different permutations
@@ -189,8 +222,9 @@ void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len) {
     upper_hash ^= c ^ upper_hash_mask;
     lower_hash ^= c ^ lower_hash_mask;
     // table lookup
-    uint8_t x;
     uint64_t h = 0;
+#ifdef LOW_MEM_FOOTPRINT // 256 byte look-up table ----------
+    uint8_t x;
     x = upper_hash; x = t[x]; upper_hash >>= 8; h |= x; h=ROR64(h,8);
     x = upper_hash; x = t[x]; upper_hash >>= 8; h |= x; h=ROR64(h,8);
     x = upper_hash; x = t[x]; upper_hash >>= 8; h |= x; h=ROR64(h,8);
@@ -210,7 +244,22 @@ void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len) {
     x = lower_hash; x = t[x]; lower_hash >>= 8; h |= x; h=ROR64(h,8);
     x = lower_hash; x = t[x]; lower_hash >>= 8; h |= x; h=ROR64(h,8);
     x = lower_hash; x = t[x]; lower_hash >>= 8; h |= x; h=ROR64(h,8);
+    lower_hash= h;
+#else // 16-bit look-up table -------------------------------
+    uint16_t x;
+    x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = upper_hash; x = t16[x]; upper_hash >>= 16; h |= x; h=ROR64(h,16);
+    upper_hash = h;
+
+    h = 0;
+    x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
+    x = lower_hash; x = t16[x]; lower_hash >>= 16; h |= x; h=ROR64(h,16);
     lower_hash = h;
+#endif // LOW_MEM_FOOTPRINT ------
   }
   // store output
   uint64_t *o;
@@ -220,6 +269,38 @@ void pearson_hash_128 (uint8_t *out, const uint8_t *in, size_t len) {
   *o = lower_hash;
 }
 
+/* --- for later use ---
+// 32-bit hash: the return value has to be interpreted as uint32_t and
+// follows machine-specific endianess in memory
+uint32_t pearson_hash_32 (const uint8_t *in, size_t len) {
+
+  size_t i;
+  uint32_t hash = 0;
+  uint32_t hash_mask = 0x03020100;
+
+  for (i = 0; i < len; i++) {
+    // broadcast the character, xor into hash, make them different permutations
+    uint32_t c = (uint8_t)in[i];
+    c |= c <<  8;
+    c |= c << 16;
+    hash ^= c ^ hash_mask;
+    // table lookup
+#ifdef LOW_MEM_FOOTPRINT
+    uint32_t h = 0;
+    uint8_t x;
+    x = hash; x = t[x]; hash >>= 8; h |= x; h=ROR32(h,8);
+    x = hash; x = t[x]; hash >>= 8; h |= x; h=ROR32(h,8);
+    x = hash; x = t[x]; hash >>= 8; h |= x; h=ROR32(h,8);
+    x = hash; x = t[x]; hash >>= 8; h |= x; h=ROR32(h,8);
+    hash = h;
+#else
+    hash = (t16[hash >> 16] << 16) + t16[(uint16_t)hash];
+#endif
+  }
+  // output
+  return hash;
+} --- pearson_hash_32 for later use --- */
+
 
 // 16-bit hash: the return value has to be interpreted as uint16_t and
 // follows machine-specific endianess in memory
@@ -234,8 +315,23 @@ uint16_t pearson_hash_16 (const uint8_t *in, size_t len) {
     c |= c <<  8;
     hash ^= c ^ hash_mask;
     // table lookup
+#ifdef LOW_MEM_FOOTPRINT
     hash = t[(uint8_t)hash] + (t[hash >> 8] << 8);
+#else
+    hash = t16[hash];
+#endif
   }
   // output
   return hash;
 }
+
+
+void pearson_hash_init () {
+
+#ifndef LOW_MEM_FOOTPRINT
+  size_t i;
+
+  for (i = 0; i < 65536; i++)
+    t16[i] = (t[i >> 8] << 8) + t[(uint8_t)i];
+#endif
+}