diff --git a/src/Makefile b/src/Makefile
index 5f7536e8..cb7afa02 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -204,7 +204,7 @@ $(REDIS_CHECK_AOF_NAME): $(REDIS_CHECK_AOF_OBJ)
 	$(REDIS_LD) -o $@ $^ $(FINAL_LIBS)
 
 dict-benchmark: dict.c zmalloc.c sds.c
-	$(REDIS_CC) $(FINAL_CFLAGS) dict.c zmalloc.c sds.c -D DICT_BENCHMARK_MAIN -o dict-benchmark
+	$(REDIS_CC) $(FINAL_CFLAGS) dict.c zmalloc.c sds.c siphash.c -D DICT_BENCHMARK_MAIN -o dict-benchmark
 
 # Because the jemalloc.h header is generated as a part of the jemalloc build,
 # building it should complete before building any other object. Instead of
diff --git a/src/dict.c b/src/dict.c
index 8ce73596..69fb3b8f 100644
--- a/src/dict.c
+++ b/src/dict.c
@@ -1109,7 +1109,7 @@ void dictGetStats(char *buf, size_t bufsize, dict *d) {
 
 #include "sds.h"
 
-unsigned int hashCallback(const void *key) {
+uint64_t hashCallback(const void *key) {
     return dictGenHashFunction((unsigned char*)key, sdslen((char*)key));
 }
 
diff --git a/src/siphash.c b/src/siphash.c
index 3bcada72..7219d4b8 100644
--- a/src/siphash.c
+++ b/src/siphash.c
@@ -40,6 +40,16 @@
 #include <string.h>
 #include <ctype.h>
 
+/* Fast tolower() alike function that does not care about locale
+ * but just returns a-z insetad of A-Z. */
+int siptlw(int c) {
+    if (c >= 'A' && c <= 'Z') {
+        return c+('a'-'A');
+    } else {
+        return c;
+    }
+}
+
 /* Test of the CPU is Little Endian and supports not aligned accesses.
  * Two interesting conditions to speedup the function that happen to be
  * in most of x86 servers. */
@@ -70,14 +80,14 @@
 #endif
 
 #define U8TO64_LE_NOCASE(p)                                                    \
-    (((uint64_t)(tolower((p)[0]))) |                                           \
-     ((uint64_t)(tolower((p)[1])) << 8) |                                      \
-     ((uint64_t)(tolower((p)[2])) << 16) |                                     \
-     ((uint64_t)(tolower((p)[3])) << 24) |                                     \
-     ((uint64_t)(tolower((p)[4])) << 32) |                                              \
-     ((uint64_t)(tolower((p)[5])) << 40) |                                              \
-     ((uint64_t)(tolower((p)[6])) << 48) |                                              \
-     ((uint64_t)(tolower((p)[7])) << 56))
+    (((uint64_t)(siptlw((p)[0]))) |                                           \
+     ((uint64_t)(siptlw((p)[1])) << 8) |                                      \
+     ((uint64_t)(siptlw((p)[2])) << 16) |                                     \
+     ((uint64_t)(siptlw((p)[3])) << 24) |                                     \
+     ((uint64_t)(siptlw((p)[4])) << 32) |                                              \
+     ((uint64_t)(siptlw((p)[5])) << 40) |                                              \
+     ((uint64_t)(siptlw((p)[6])) << 48) |                                              \
+     ((uint64_t)(siptlw((p)[7])) << 56))
 
 #define SIPROUND                                                               \
     do {                                                                       \
@@ -192,13 +202,13 @@ uint64_t siphash_nocase(const uint8_t *in, const size_t inlen, const uint8_t *k)
     }
 
     switch (left) {
-    case 7: b |= ((uint64_t)tolower(in[6])) << 48;
-    case 6: b |= ((uint64_t)tolower(in[5])) << 40;
-    case 5: b |= ((uint64_t)tolower(in[4])) << 32;
-    case 4: b |= ((uint64_t)tolower(in[3])) << 24;
-    case 3: b |= ((uint64_t)tolower(in[2])) << 16;
-    case 2: b |= ((uint64_t)tolower(in[1])) << 8;
-    case 1: b |= ((uint64_t)tolower(in[0])); break;
+    case 7: b |= ((uint64_t)siptlw(in[6])) << 48;
+    case 6: b |= ((uint64_t)siptlw(in[5])) << 40;
+    case 5: b |= ((uint64_t)siptlw(in[4])) << 32;
+    case 4: b |= ((uint64_t)siptlw(in[3])) << 24;
+    case 3: b |= ((uint64_t)siptlw(in[2])) << 16;
+    case 2: b |= ((uint64_t)siptlw(in[1])) << 8;
+    case 1: b |= ((uint64_t)siptlw(in[0])); break;
     case 0: break;
     }