diff --git a/Cargo.toml b/Cargo.toml
index d3dee26..93d14bd 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -18,6 +18,7 @@ codecov = {repository = "sile/libflate"}
 adler32 = "1"
 byteorder = "1"
 crc32fast = "1"
+rle-decode-fast = "1.0.0"
 take_mut = "0.2.2"
 
 [dev-dependencies]
diff --git a/src/deflate/decode.rs b/src/deflate/decode.rs
index 862992e..3f0217c 100644
--- a/src/deflate/decode.rs
+++ b/src/deflate/decode.rs
@@ -3,11 +3,11 @@ use byteorder::ReadBytesExt;
 use std::cmp;
 use std::io;
 use std::io::Read;
+use rle_decode_fast::rle_decode;
 
 use super::symbol;
 use bit;
 use lz77;
-use util;
 
 /// DEFLATE decoder.
 #[derive(Debug)]
@@ -116,19 +116,7 @@ where
                             distance
                         ));
                     }
-                    let old_len = self.buffer.len();
-                    self.buffer.reserve(length as usize);
-                    unsafe {
-                        self.buffer.set_len(old_len + length as usize);
-                        let start = old_len - distance as usize;
-                        let ptr = self.buffer.as_mut_ptr();
-                        util::ptr_copy(
-                            ptr.add(start),
-                            ptr.add(old_len),
-                            length as usize,
-                            length > distance,
-                        );
-                    }
+                    rle_decode(&mut self.buffer, usize::from(distance), usize::from(length));
                 }
                 symbol::Symbol::EndOfBlock => {
                     break;
diff --git a/src/lib.rs b/src/lib.rs
index 3803e16..a414e02 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -3,6 +3,7 @@
 extern crate adler32;
 extern crate byteorder;
 extern crate crc32fast;
+extern crate rle_decode_fast;
 extern crate take_mut;
 
 pub use finish::Finish;
diff --git a/src/non_blocking/deflate/decode.rs b/src/non_blocking/deflate/decode.rs
index 438b72a..f672911 100644
--- a/src/non_blocking/deflate/decode.rs
+++ b/src/non_blocking/deflate/decode.rs
@@ -3,12 +3,11 @@ use byteorder::ReadBytesExt;
 use std::cmp;
 use std::io;
 use std::io::Read;
+use rle_decode_fast::rle_decode;
 
 use deflate::symbol::{self, HuffmanCodec};
 use lz77;
 use non_blocking::transaction::TransactionalBitReader;
-use util;
-
 /// DEFLATE decoder which supports non-blocking I/O.
 #[derive(Debug)]
 pub struct Decoder<R> {
@@ -210,19 +209,7 @@ impl BlockDecoder {
                             distance
                         ));
                     }
-                    let old_len = self.buffer.len();
-                    self.buffer.reserve(length as usize);
-                    unsafe {
-                        self.buffer.set_len(old_len + length as usize);
-                        let start = old_len - distance as usize;
-                        let ptr = self.buffer.as_mut_ptr();
-                        util::ptr_copy(
-                            ptr.add(start),
-                            ptr.add(old_len),
-                            length as usize,
-                            length > distance,
-                        );
-                    }
+                    rle_decode(&mut self.buffer, usize::from(distance), usize::from(length));
                 }
                 symbol::Symbol::EndOfBlock => {
                     self.eob = true;
diff --git a/src/util.rs b/src/util.rs
index 682bfa7..82f58d1 100644
--- a/src/util.rs
+++ b/src/util.rs
@@ -1,17 +1,5 @@
 #[cfg(test)]
 use std::io::{self, Read};
-use std::ptr;
-
-#[inline]
-pub unsafe fn ptr_copy(src: *const u8, dst: *mut u8, count: usize, is_overlapping: bool) {
-    if !is_overlapping {
-        ptr::copy_nonoverlapping(src, dst, count);
-    } else {
-        for i in 0..count {
-            ptr::copy_nonoverlapping(src.add(i), dst.add(i), 1);
-        }
-    }
-}
 
 #[cfg(test)]
 pub struct WouldBlockReader<R> {