Silice-float

Customizable floating point number lib for the Silice FPGA language

Using the Silice Draft branch

Features

• int to float
• float to int
• float addition
• float multiplication
• float division
• FPU module

How to use

first you need to include the float definition with :

$include('/pathToSiliceFloat/src/float_def.ice')

Then define the size of an integer :

$$int_size = 32

if you are using the FPU module make sure the int size match the size of the float you are using

Using the FPU module

To be able to use the FPU module you need to include it in your project :

$include('/pathToSiliceFloat/src/FPU_module.ice')

Then in your algorithm you need to declare it, there is multiple way to do that in Silice, Personaly I use the following one :

  uint$float_size$ in1 = uninitialized;
  uint$float_size$ in2 = uninitialized;
  uint$float_size$ out = uninitialized;
  uint3            funct = uninitialized;
  uint1            wr = uninitialized;
  uint1            busy = uninitialized;

  FPU_module FPU(
      in1   <: in1,
      in2   <: in2,
      out   :> out,
      funct <: funct,
      wr    <: wr,
      busy  :> busy,
  );
  wr ::= 0;

This way you can then call the FPU module by setting wr to 1 and setting up it's input, and you know if the computation is done when the busy value goes to 0

You can pick the function you want to using the funct value :

• 0 will add in1 and in2 with a float addition and return it's value in the out value
• 1 will substract in2 to in1 with a float substraction and return it's value in the out value
• 2 will multiply in1 by in2 with a float multiplication and return it's value in the out value
• 3 will divide in1 by in2 with a float division and return it's value in the out value
• 4 will convert the value in in1 from an int to a float and return it in the out value
• 5 will convert the value in in1 from a float to an int and return it in the out value

For example the addition can be called as following :

  in1 = 32b11001010110110001000100110101110; // a floating value
  in2 = 32b11001010100100000110000001001010; // another floating value
  wr =1;
  funct = 0;
  while(busy){}
  // now he out value should be the sum of in1 and in2

Using Custom floating point

This lib has been made to handle customisable floating The default floating point contained in the float_def.ice file is the IEEE 754 32 bits float

To change that you can create your own float definition file by copying the float_def.ice file and changing the exonant_size and the mantissa_size.

for the IEEE 754 64 bit float, the file should look like that :

$$if FLOAT_DEF == nil then
$$FLOAT_DEF = 1
//IEEE 754

$$exponant_size = 11
$$mantissa_size = 52


$$float_size = 1 + exponant_size + mantissa_size

bitfield float{
    uint1 sign,
    uint$exponant_size$ exponant,
    uint$mantissa_size$ fraction,
}

$$end

Then you can change the included float definition file by your own :

$include('pathtomyfloatdef/my_float.ice')