[Feature Request] exact floating-point mantissa overflow round to byte

[jopadan]

#include <float.h>

        inline BX_CONSTEXPR_FUNC uint8_t ubround(float _a)
        {
                _a += (1u << (FLT_MANT_DIG - 1)) + (1u << (FLT_MANT_DIG - 2));
                return *(uint8_t*)&_a;
        }

        inline BX_CONSTEXPR_FUNC int8_t bround(float _a)
        {
                _a += (1u << (FLT_MANT_DIG - 1)) + (1u << (FLT_MANT_DIG - 2));
                return *(int8_t*)&_a;
        }

Causes constexpr warning for reinterpret cast.
Works for HALF_MANT_DIG = 11 C23 _Float16 types too but uses software emulation if no hardware support is present

[bkaradzic]

Where are you using this?

[jopadan]

VGA6 to RGB8 level up see bround