c++ - Arbitrary Precision Floating Point library for High Level Synthesis -

i trying make template based arbitrary precision floating point library in c++ supports variable exponents , variable mantissa can specified template arguments. have developed template based fixed point library. type of implementation want :

 template<int exponent_bits, int mantissa_bits>      struct fp_float          {            <some_data_type_to_store_exponent_and_mantissa_values>;          }; 

i unable find suitable datatype store exponents don't use more bits needed code. thought of using intn_t n = {8, 16, 32, 64} if declare fp_float<3,11> use 8 bits exponent , 16 bits mantissa.

hence makes entire library useless uses more resources ought specified precision.

i know if there other arbitrary precision data-type serves purpose.

i did come across few arbitrary precision libraries these libraries have code structure cannot synthesized hardware descriptions using high level synthesis (which reason making library).

is valid solution you? base_int_t trait type gives basic type used in subsequent bitfield definition. code below missing specializations n > 2

// gives integer type fitting in n bytes template <int n> struct base_int_t {     typedef int type; }; // specializations template <> struct base_int_t<1> {     typedef unsigned char type; };  template <> struct base_int_t<2> {     typedef unsigned short type; }; // add suitable definitions n = 3,4...8. n = 3 , 4 type unsigned int  template <int exp_bits, int mantissa_bits> struct fp_float {     // template argument number of bytes required     typedef typename base_int_t<(exp_bits + mantissa_bits + 7) / 8>::type type;     type mantissa : mantissa_bits;     type exponent : exp_bits; };  typedef fp_float<3, 11> fp_3_11_t; fp_3_11_t fp;