Curious about rounding

[JPHutchins]

Often, my use case may be to avoid soft floats on hardware that has no FPU. See this section of code:

libfixmath/libfixmath/fix16.h

Lines 72 to 89 in d308e46

	static inline fix16_t fix16_from_float(float a)
	{
	float temp = a * fix16_one;
	#ifndef FIXMATH_NO_ROUNDING
	temp += (temp >= 0) ? 0.5f : -0.5f;
	#endif
	return (fix16_t)temp;
	}
	static inline fix16_t fix16_from_dbl(double a)
	{
	double temp = a * fix16_one;
	/* F16() and F16C() are both rounding allways, so this should as well */
	//#ifndef FIXMATH_NO_ROUNDING
	temp += (double)((temp >= 0) ? 0.5f : -0.5f);
	//#endif
	return (fix16_t)temp;
	}

First, I think it's problematic that "from double" is disregarding the rounding compile def while "from float" respects it.

Second, I'm curious about the statement:

temp += (temp >= 0) ? 0.5f : -0.5f;

In my use case, won't this bring in the soft fadd that I don't want? Then again, I suppose that the user is just doing it wrong if they don't want any floats yet the end up using fix16_from_float at runtime!

[PetteriAimonen]

I agree that the rounding difference is weird and should be fixed. Not sure which way makes more sense though.

To avoid any float use, I think you'd use F16() or F16C() macro. Though probably in most cases compiler will optimize away the inline functions.

[JPHutchins]

Totally agreed re: F16 and F16C.