cw-b-w / fixedpointnumberlibrary

Simple header-only library for Fixed-Point Number operations

fixedpointnumberlibrary's Issues

Division is incomplete

FixedPointNumberLibrary/include/FixedPointNumber.hpp

Line 339 in 3b5a2a2

FixedPointNumber<INT_BIT_LEN, FRAC_BIT_LEN> FixedPointNumber<INT_BIT_LEN, FRAC_BIT_LEN>::operator/ (const FixedPointNumber<T1, T2> &rhs) const

is incomplete:

• Cannot do with two fixed point numbers of different scheme (bit length)
• No saturation arithmetic for overflow

No rounding when multiplying

Inside

template<int INT_BIT_LEN, int FRAC_BIT_LEN>
template<int T1, int T2>
FixedPointNumber<INT_BIT_LEN, FRAC_BIT_LEN> FixedPointNumber<INT_BIT_LEN, FRAC_BIT_LEN>::operator* (const FixedPointNumber<T1, T2> &rhs) const

The bits behind LSB are discarded, but it's better to round it to LSB.

FixedPointNumber res;
uint64_t product = lv * rv;
product >>= T2; // No rounding here

to_double(): Is using bit-operation really faster

234b58d

Originally, only one main instruction. This may be faster on modern CPU even if it's floating point number multiplication.

d /= (double)(1ULL << FRAC_BIT_LEN);

Beside, bit-operation method uses a while-loop, it may be a bigger burden than floating point number multiplication.

Need to apply saturation arithmetic

During doing arithmetic operation, if the resulting number exceeds the min/max range, it should be set to min/max.

ref: saturation arithmetic