diff --git a/snappy-internal.h b/snappy-internal.h
index 720ccd8..1b8e4dd 100644
--- a/snappy-internal.h
+++ b/snappy-internal.h
@@ -103,7 +103,21 @@ static inline std::pair<size_t, bool> FindMatchLength(const char* s1,
                                                       uint64_t* data) {
   assert(s2_limit >= s2);
   size_t matched = 0;
-
+ 
+  #if defined(__has_builtin)
+  #if __has_builtin(__builtin_prefetch)
+    __builtin_prefetch(s1);
+    __builtin_prefetch(s2);
+  #endif
+  #elif defined(__SSE__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 1))
+  #include <xmmintrin.h>
+    _mm_prefetch(s1, _MM_HINT_T1);
+    _mm_prefetch(s2, _MM_HINT_T1);
+  #elif defined(_M_ARM64) || defined(_M_ARM)
+    __prefetch(s1);
+    __prefetch(s2);
+  #endif
+ 
   // This block isn't necessary for correctness; we could just start looping
   // immediately.  As an optimization though, it is useful.  It creates some not
   // uncommon code paths that determine, without extra effort, whether the match