reinterpret-cast

Memory reinterpretation casts for Float/Double and Word32/Word64 in Haskell.

Takes the bit representation of a number and uses it for a different numeric type. This is reinterpret_cast from C++ and float f = 1.23; int i = * (int *) &f; from C.

We offer you fast casts for:

floatToWord  :: Float  -> Word32
wordToFloat  :: Word32 -> Float
doubleToWord :: Double -> Word64
wordToDouble :: Word64 -> Double

If you need something like Int32 or similar, fromIntegral will do the job within the integral types (so to/from Word* are the only conversions needed).

On Hackage: cabal install reinterpret-cast

Fast casting

The way implemented in this package is the fastest possible way known as of now. In particular, it is faster than what data-binary-ieee754 does at the moment.

For a discussion, see this StackOverflow question.

Benchmark results for bench/Bench.hs are available here.

In theory, these should all be no-ops, but you must not use unsafeCoerce or even unsafeCoerce# for this.

The real solution will be this GHC feature request - please support it.

TODO

• Compare to the performance of an unsafe pure FFI call that does float f = 1.23; int i = * (int *) &f; inside.

Contributions welcome!