diff --git a/.gitignore b/.gitignore
new file mode 100644
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,17 @@
+dist/
+.hsenv/
+docs
+wiki
+TAGS
+tags
+wip
+.DS_Store
+.*.swp
+.*.swo
+*.o
+*.hi
+*~
+*#
+.cabal-sandbox/
+cabal.sandbox.config
+codex.tags
diff --git a/.travis.yml b/.travis.yml
new file mode 100644
--- /dev/null
+++ b/.travis.yml
@@ -0,0 +1,56 @@
+language: haskell
+
+env:
+  - GHCVER=7.4.2
+  - GHCVER=7.6.3
+  - GHCVER=7.8.3
+  - GHCVER=head
+
+matrix:
+  allow_failures:
+    - env: GHCVER=head
+
+before_install:
+  # If $GHCVER is the one travis has, don't bother reinstalling it.
+  # We can also have faster builds by installing some libraries with
+  # `apt`. If it isn't, install the GHC we want from hvr's PPA along
+  # with cabal-1.18.
+  - |
+    if [ $GHCVER = `ghc --numeric-version` ]; then
+      # Try installing some of the build-deps with apt-get for speed.
+      travis/cabal-apt-install --enable-tests $MODE
+      export CABAL=cabal
+    else
+      # Install the GHC we want from hvr's PPA
+      travis_retry sudo add-apt-repository -y ppa:hvr/ghc
+      travis_retry sudo apt-get update
+      travis_retry sudo apt-get install cabal-install-1.18 ghc-$GHCVER happy
+      export CABAL=cabal-1.18
+      export PATH=/opt/ghc/$GHCVER/bin:$PATH
+    fi
+  # Uncomment whenever hackage is down.
+  # - mkdir -p ~/.cabal && cp travis/config ~/.cabal/config && $CABAL update
+  - $CABAL update
+
+  # Update happy when building with GHC head
+  - |
+    if [ $GHCVER = "head" ] || [ $GHCVER = "7.8.3" ]; then
+      $CABAL install happy alex
+      export PATH=$HOME/.cabal/bin:$PATH
+    fi
+
+install:
+  - $CABAL install --dependencies-only --enable-tests
+  - $CABAL configure --enable-tests $MODE
+
+script:
+  - $CABAL build
+  - $CABAL test --show-details=always
+
+notifications:
+  irc:
+    channels:
+      - "irc.freenode.org#haskell-lens"
+    skip_join: true
+    template:
+      - "\x0313gl\x03/\x0306%{branch}\x03 \x0314%{commit}\x03 %{build_url} %{message}"
diff --git a/CHANGELOG.markdown b/CHANGELOG.markdown
new file mode 100644
--- /dev/null
+++ b/CHANGELOG.markdown
@@ -0,0 +1,4 @@
+0.1
+---
+* Initial release
+
diff --git a/LICENSE b/LICENSE
new file mode 100644
--- /dev/null
+++ b/LICENSE
@@ -0,0 +1,26 @@
+Copyright 2014 Edward Kmett
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions
+are met:
+
+1. Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
+IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGE.
diff --git a/README.markdown b/README.markdown
new file mode 100644
--- /dev/null
+++ b/README.markdown
@@ -0,0 +1,17 @@
+half
+====
+
+[![Build Status](https://secure.travis-ci.org/ekmett/half.png?branch=master)](http://travis-ci.org/ekmett/half)
+
+This package supplies half-precision floating point values w/ 1 bit of sign, 5 bits of exponent, 11 bits of mantissa trailing a leading 1 bit with proper underflow.
+
+These arise commonly in GPU applications.
+
+Contact Information
+-------------------
+
+Contributions and bug reports are welcome!
+
+Please feel free to contact me through github or on the #haskell IRC channel on irc.freenode.net.
+
+-Edward Kmett
diff --git a/Setup.hs b/Setup.hs
new file mode 100644
--- /dev/null
+++ b/Setup.hs
@@ -0,0 +1,2 @@
+import Distribution.Simple
+main = defaultMain
diff --git a/cbits/half.c b/cbits/half.c
new file mode 100644
--- /dev/null
+++ b/cbits/half.c
@@ -0,0 +1,62 @@
+unsigned short hs_floatToHalf (float f) {
+  union { float d; unsigned int i; } u = { f };
+  int s =  (u.i >> 16) & 0x8000;
+  int e = ((u.i >> 23) & 0xff) - 112;
+  int m =          u.i & 0x7fffff;
+  if (e <= 0) {
+    if (e < -10) return s; /* underflowed */
+    /* force leading 1 and round */
+    m |= 0x800000;
+    int t = 14 - e;
+    int a = (1 << (t - 1)) - 1;
+    int b = (m >> t) & 1;
+    return s | ((m + a + b) >> t);
+  } 
+  if (e == 143) {
+    if (m == 0) return s | 0x7c00; /* +/- infinity */
+
+    /* NaN, m == 0 forces us to set at least one bit and not become an infinity */
+    m >>= 13;
+    return s | 0x7c00 | m | (m == 0); 
+  }
+  
+  /* round the normalized float */
+  m = m + 0xfff + ((m >> 13) & 1);
+
+  /* significand overflow */
+  if (m & 0x800000) {
+     m =  0;
+     e += 1;
+  }
+
+  /* exponent overflow */
+  if (e > 30) return s | 0x7c00;
+
+  return s | (e << 10) | (m >> 13);
+}
+
+int hs_halfToFloatRep (unsigned short c) {
+  int s = (c >> 15) & 0x001;
+  int e = (c >> 10) & 0x01f;
+  int m =         c & 0x3ff;
+  if (e == 0) {
+    if (m == 0) /* +/- 0 */ return s << 31;
+    /* denormalized, renormalize it */
+    while (!(m & 0x400)) {
+      m <<= 1;
+      e -=  1;
+    }
+    e += 1;
+    m &= ~0x400;
+  } else if (e == 31) return (s << 31) | 0x7f800000 | (m << 13); /* NaN or +/- infinity */
+  e += 112;
+  m <<= 13;
+  return (s << 31) | (e << 23) | m;
+}
+
+float hs_halfToFloat (unsigned short c) {
+  union { float d; unsigned int i; } u;
+  u.i = hs_halfToFloatRep(c);
+  return u.d;
+}
+
diff --git a/half.cabal b/half.cabal
new file mode 100644
--- /dev/null
+++ b/half.cabal
@@ -0,0 +1,33 @@
+name:          half
+category:      Numeric
+version:       0.1
+license:       BSD3
+cabal-version: >= 1.8
+license-file:  LICENSE
+author:        Edward A. Kmett
+maintainer:    Edward A. Kmett <ekmett@gmail.com>
+stability:     provisional
+homepage:      http://github.com/ekmett/half
+bug-reports:   http://github.com/ekmett/half/issues
+copyright:     Copyright (C) 2014 Edward A. Kmett
+build-type:    Simple
+tested-with:   GHC == 7.8.3
+synopsis:      Half-precision floating-point
+description:   Half-precision floating-point
+
+extra-source-files:
+  .travis.yml
+  .gitignore
+  README.markdown
+  CHANGELOG.markdown
+
+source-repository head
+  type: git
+  location: git://github.com/analytics/half.git
+
+library
+  hs-source-dirs: src
+  c-sources: cbits/half.c
+  build-depends: base >= 4.3 && < 5
+  ghc-options: -Wall -fwarn-tabs -O2
+  exposed-modules: Numeric.Half
diff --git a/src/Numeric/Half.hs b/src/Numeric/Half.hs
new file mode 100644
--- /dev/null
+++ b/src/Numeric/Half.hs
@@ -0,0 +1,158 @@
+{-# LANGUAGE ForeignFunctionInterface #-}
+{-# LANGUAGE PatternSynonyms #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+{-# LANGUAGE GeneralizedNewtypeDeriving #-}
+{-# LANGUAGE DeriveDataTypeable #-}
+-----------------------------------------------------------------------------
+-- |
+-- Copyright   :  (C) 2014 Edward Kmett
+-- License     :  BSD-style (see the file LICENSE)
+-- Maintainer  :  Edward Kmett <ekmett@gmail.com>
+-- Stability   :  experimental
+-- Portability :  PatternSynonyms
+--
+-- Half-precision floating-point values. These arise commonly in GPU work
+-- and it is useful to be able to compute them and compute with them on the
+-- CPU as well.
+----------------------------------------------------------------------------
+
+module Numeric.Half 
+  ( Half(..)
+  , isZero
+  , toFloat
+  , toHalf
+  , pattern POS_INF
+  , pattern NEG_INF
+  , pattern QNaN
+  , pattern SNaN
+  , pattern HALF_MIN
+  , pattern HALF_NRM_MIN
+  , pattern HALF_MAX
+  , pattern HALF_EPSILON
+  , pattern HALF_DIG
+  , pattern HALF_MIN_10_EXP
+  , pattern HALF_MAX_10_EXP
+  ) where
+
+import Data.Bits
+import Data.Function (on)
+import Data.Typeable
+import Foreign.C.Types
+import Foreign.Storable
+
+-- | Convert a 'Float' to a 'Half' with proper rounding, while preserving NaN and dealing appropriately with infinity
+foreign import ccall unsafe "hs_floatToHalf" toHalf :: Float -> Half
+{-# RULES "toHalf"  realToFrac = toHalf #-}
+
+-- | Convert a 'Half' to a 'Float' while preserving NaN
+foreign import ccall unsafe "hs_halfToFloat" toFloat :: Half -> Float
+{-# RULES "toFloat" realToFrac = toFloat #-}
+
+newtype Half = Half { getHalf :: CUShort } deriving (Storable, Typeable)
+
+instance Show Half where
+  showsPrec d h = showsPrec d (toFloat h)
+
+instance Eq Half where
+  (==) = (==) `on` toFloat
+
+instance Ord Half where
+  compare = compare `on` toFloat
+
+instance Real Half where
+  toRational = toRational . toFloat
+
+instance Fractional Half where
+  fromRational = toHalf . fromRational
+  recip = toHalf . recip . toFloat
+  a / b = toHalf $ toFloat a / toFloat b
+
+instance RealFrac Half where
+  properFraction a = case properFraction (toFloat a) of
+    (b, c) -> (b, toHalf c)
+  truncate = truncate . toFloat
+  round = round . toFloat
+  ceiling = ceiling . toFloat
+  floor = floor . toFloat
+
+instance Floating Half where
+  pi = toHalf pi
+  exp = toHalf . exp . toFloat
+  sqrt = toHalf . sqrt . toFloat
+  log = toHalf . log . toFloat
+  a ** b = toHalf $ toFloat a ** toFloat b
+  logBase a b = toHalf $ logBase (toFloat a) (toFloat b)
+  sin = toHalf . sin . toFloat
+  tan = toHalf . tan . toFloat
+  cos = toHalf . cos . toFloat
+  asin = toHalf . asin . toFloat
+  atan = toHalf . atan . toFloat
+  acos = toHalf . acos . toFloat
+  sinh = toHalf . sinh . toFloat
+  tanh = toHalf . tanh . toFloat
+  cosh = toHalf . cosh . toFloat
+  asinh = toHalf . asinh . toFloat
+  atanh = toHalf . atanh . toFloat
+  acosh = toHalf . acosh . toFloat
+
+instance RealFloat Half where
+  floatRadix  _ = 2
+  floatDigits _ = 11
+  decodeFloat = decodeFloat . toFloat
+  isInfinite (Half h) = unsafeShiftR h 10 .&. 0x1f >= 32
+  isIEEE _ = isIEEE (undefined :: Float)
+  atan2 a b = toHalf $ atan2 (toFloat a) (toFloat b)
+  isDenormalized (Half h) = unsafeShiftR h 10 .&. 0x1f == 0 && h .&. 0x3ff /= 0
+  isNaN (Half h) = unsafeShiftR h 10 .&. 0x1f == 0x1f && h .&. 0x3ff /= 0
+  isNegativeZero (Half h) = h == 0x8000
+  floatRange _ = (16,-13)
+  encodeFloat i j = toHalf $ encodeFloat i j
+  exponent = exponent . toFloat
+  significand = toHalf . significand . toFloat
+  scaleFloat n = toHalf . scaleFloat n . toFloat
+
+-- | Is this 'Half' equal to 0?
+isZero :: Half -> Bool
+isZero (Half h) = h .&. 0x7fff == 0
+
+-- | Positive infinity
+pattern POS_INF = Half 0x7c00
+
+-- | Negative infinity
+pattern NEG_INF = Half 0xfc00
+
+-- | Quiet NaN
+pattern QNaN    = Half 0x7fff
+
+-- | Signalling NaN
+pattern SNaN    = Half 0x7dff
+
+-- | Smallest positive half
+pattern HALF_MIN = 5.96046448e-08 :: Half 
+
+-- | Smallest positive normalized half
+pattern HALF_NRM_MIN = 6.10351562e-05 :: Half
+
+-- | Largest positive half
+pattern HALF_MAX = 65504.0 :: Half
+
+-- | Smallest positive e for which half (1.0 + e) != half (1.0)
+pattern HALF_EPSILON = 0.00097656 :: Half
+
+-- | Number of base 10 digits that can be represented without change
+pattern HALF_DIG = 2 
+
+-- Minimum positive integer such that 10 raised to that power is a normalized half
+pattern HALF_MIN_10_EXP = -4
+
+-- Maximum positive integer such that 10 raised to that power is a normalized half
+pattern HALF_MAX_10_EXP = 4
+
+instance Num Half where
+  a * b = toHalf (toFloat a * toFloat b)
+  a - b = toHalf (toFloat a - toFloat b)
+  a + b = toHalf (toFloat a + toFloat b)
+  negate (Half a) = Half (xor 0x8000 a)
+  abs = toHalf . abs . toFloat
+  signum = toHalf . signum . toFloat
+  fromInteger a = toHalf (fromInteger a)