README.md

   1 # sfpy
   2 softfloat and softposit in Python
   3   * support for softfloat float16, float32, and float64
   4   * support for softposit posit8, quire8, posit16, and quire16
   5
   6 ## Demo
   7 ```
   8 >>> import sfpy
   9 >>> sfpy.Float32(1.3) + sfpy.Float32(1.4) # <-- construct from doubles
  10 Float32(2.6999998092651367)
  11 >>> sfpy.Float32(3) # <-- construct from raw bits
  12 Float32(4.203895392974451e-45)
  13 >>> sfpy.Float32(3).bits
  14 3
  15 >>> x = sfpy.Float16(0)
  16 >>> x
  17 Float16(0.0)
  18 >>> x += sfpy.Float16(10.0) # <-- in-place operators have better performance
  19 >>> x
  20 Float16(10.0)
  21 >>> sfpy.Posit16(1.3) + sfpy.Posit16(1.4) # <-- posits work the same way as floats
  22 Posit16(2.7001953125)
  23 >>> q = sfpy.Quire16(0) # <-- quire is also supported
  24 >>> q
  25 Quire16(0.0)
  26 >>> q.iqam(sfpy.Posit16(3), sfpy.Posit16(5))
  27 >>> q
  28 Quire16(3.725290298461914e-09)
  29 >>> q.iqam(sfpy.Posit16(3.0), sfpy.Posit16(5.0))
  30 >>> q
  31 Quire16(15.0)
  32 >>> q.bits
  33 1080863910568919232
  34 >>> bin(q.bits)
  35 '0b111100000000000000000000000000000000000000000000000011000000'
  36 ```
  37
  38 ## Building (on Linux)
  39 The Cython module can be built in place in the usual way:
  40
  41 `python setup.py build_ext --inplace`
  42
  43 This requires the submodules to be checked out, and the static libraries `SoftPosit/build/Linux-x86_64-GCC/softposit.a` and `berkeley-softfloat-3/build/Linux-x86_64-GCC/softfloat.a` to be built. Note that in order for Cython to be able to build the shared objects for the module, the static libraries must be compiled with -fPIC, which currently requires modifying the appropriate Makefiles manually. SoftPosit can be build with -fPIC using its python2 and python3 targets.
  44
  45 The package can also be installed to a local Python distribution with pip, i.e. `pip install .` from the top level of the repository using the appropriate pip. This requires that Cython be installed.