diff --git a/Data/ByteString/FastBuilder/Internal.hs b/Data/ByteString/FastBuilder/Internal.hs
--- a/Data/ByteString/FastBuilder/Internal.hs
+++ b/Data/ByteString/FastBuilder/Internal.hs
@@ -463,7 +463,8 @@
 
 -- | Turn a value of type @a@ into a 'Builder', using the given 'PI.BoundedPrim'.
 primBounded :: PI.BoundedPrim a -> a -> Builder
-primBounded prim !x = mappend (ensureBytes $ PI.sizeBound prim) $ mkBuilder $ do
+primBounded prim x = rebuild $ seq x $ mkBuilder $ do
+  useBuilder $ ensureBytes $ PI.sizeBound prim
   cur <- getCur
   cur' <- io $ PI.runB prim x cur
   setCur cur'
diff --git a/benchmarks/map.hs b/benchmarks/map.hs
new file mode 100644
--- /dev/null
+++ b/benchmarks/map.hs
@@ -0,0 +1,55 @@
+{-# LANGUAGE DeriveFunctor #-}
+import Control.Concurrent
+import Criterion.Main
+import qualified Data.IntMap as IM
+import Data.Monoid
+import qualified System.IO as IO
+
+import qualified Data.ByteString.Builder as Bstr
+import qualified Data.ByteString.FastBuilder as Fast
+
+main :: IO ()
+main = runInUnboundThread $ do
+  h <- IO.openFile "/dev/null" IO.WriteMode
+  let
+    size sz m = bgroup sz
+      [ bench "lazy/fast" $ nf (Fast.toLazyByteString . fastMap) m
+      , bench "lazy/bstr" $ nf (Bstr.toLazyByteString . bstrMap) m
+      ]
+  map10 `seq` map100 `seq` map1000 `seq` map10000 `seq` defaultMain
+    [ size "10" map10
+    , size "100" map100
+    , size "1000" map1000
+    , size "10000" map10000
+    ]
+
+type Map = IM.IntMap Int
+
+fastMap :: Map -> Fast.Builder
+fastMap = Fast.rebuild . IM.foldMapWithKey f
+  where
+    f key val =
+      Fast.int32LE (fromIntegral key) <> Fast.int32LE (fromIntegral val)
+
+bstrMap :: Map -> Bstr.Builder
+bstrMap = IM.foldMapWithKey f
+  where
+    f key val =
+      Bstr.int32LE (fromIntegral key) <> Bstr.int32LE (fromIntegral val)
+
+map10 :: Map
+map10 = makeItems 10
+
+map100 :: Map
+map100 = makeItems 100
+
+map1000 :: Map
+map1000 = makeItems 1000
+
+map10000 :: Map
+map10000 = makeItems 10000
+
+makeItems :: Int -> Map
+makeItems n = IM.fromList $ do
+  i <- [0..n-1]
+  return (i * 3, i)
diff --git a/fast-builder.cabal b/fast-builder.cabal
--- a/fast-builder.cabal
+++ b/fast-builder.cabal
@@ -1,8 +1,10 @@
 name:                fast-builder
-version:             0.0.0.0
+version:             0.0.0.1
 synopsis:            Fast ByteString Builder
 description:         An efficient implementation of ByteString builder. It should be faster than
                      the standard implementation in most cases.
+                     .
+                     In many benchmarks, the performance improvement is 2x-10x.
 homepage:            http://github.com/takano-akio/fast-builder
 license:             PublicDomain
 license-file:        LICENSE
@@ -20,7 +22,7 @@
   other-modules:       Data.ByteString.FastBuilder.Internal.Prim
 
   -- other-extensions:    
-  build-depends:       base ==4.8.0.0, bytestring >= 0.10.6.0, ghc-prim
+  build-depends:       base >= 4.8 && < 4.9, bytestring >= 0.10.6.0, ghc-prim
   -- hs-source-dirs:      
   default-language:    Haskell2010
   ghc-options:         -Wall -g
@@ -39,6 +41,13 @@
   main-is:             vector.hs
   hs-source-dirs:      benchmarks
   build-depends:       base, fast-builder, criterion, bytestring, vector, deepseq
+  ghc-options:         -Wall -threaded -g -rtsopts -eventlog
+
+benchmark map
+  type:                exitcode-stdio-1.0
+  main-is:             map.hs
+  hs-source-dirs:      benchmarks
+  build-depends:       base, fast-builder, criterion, bytestring, containers, deepseq
   ghc-options:         -Wall -threaded -g -rtsopts -eventlog
 
 test-suite prop