<random>: generate_canonical() could avoid calling log2() at runtime

[StephanTLavavej]

Gautham Beeraka reported a performance issue in generate_canonical(), where it calls log2():

STL/stl/inc/random

Lines 248 to 259 in 4c862ee

	template <class _Real, size_t _Bits, class _Gen>
	_NODISCARD _Real generate_canonical(_Gen& _Gx) { // build a floating-point value from random sequence
	_RNG_REQUIRE_REALTYPE(generate_canonical, _Real);
	const size_t _Digits = static_cast<size_t>(numeric_limits<_Real>::digits);
	const size_t _Minbits = _Digits < _Bits ? _Digits : _Bits;
	const _Real _Gxmin = static_cast<_Real>((_Gx.min)());
	const _Real _Gxmax = static_cast<_Real>((_Gx.max)());
	const _Real _Rx = (_Gxmax - _Gxmin) + _Real{1};
	const int _Ceil = static_cast<int>(_STD ceil(static_cast<_Real>(_Minbits) / _STD log2(_Rx)));

For engines like mt19937 and mt19937_64, this is a really expensive way of computing 32.0 and 64.0.

Solving this in a completely general manner might be somewhat difficult, but perhaps we could simply special-case the Standard engines that are known to have a power-of-2 range.

[pjessesco]

Hi, if I understand this issue correctly, is the suggested solution comparing _Gen with standard engines and specifing the denominator manually?

For example,

if(_Gen == mt19937) 
    denominator = 32
else if(_Gen == other pre-defined engines)
    denominator = something
....
else
    denominator = _STD log2(_Rx)