`<bit>`: Implement `rotl()` and `rotr()` with intrinsics

[StephanTLavavej]

STL/stl/inc/bit

Lines 92 to 118 in 3c2fd04

	template <class _Ty, enable_if_t<_Is_standard_unsigned_integer<_Ty>, int> = 0>
	_NODISCARD constexpr _Ty rotl(const _Ty _Val, const int _Rotation) noexcept {
	constexpr auto _Digits = numeric_limits<_Ty>::digits;
	const auto _Remainder = _Rotation % _Digits;
	if (_Remainder > 0) {
	return static_cast<_Ty>(
	static_cast<_Ty>(_Val << _Remainder) | static_cast<_Ty>(_Val >> (_Digits - _Remainder)));
	} else if (_Remainder == 0) {
	return _Val;
	} else { // _Remainder < 0
	return _STD rotr(_Val, -_Remainder);
	}
	}
	template <class _Ty, enable_if_t<_Is_standard_unsigned_integer<_Ty>, int> _Enabled>
	_NODISCARD constexpr _Ty rotr(const _Ty _Val, const int _Rotation) noexcept {
	constexpr auto _Digits = numeric_limits<_Ty>::digits;
	const auto _Remainder = _Rotation % _Digits;
	if (_Remainder > 0) {
	return static_cast<_Ty>(
	static_cast<_Ty>(_Val >> _Remainder) | static_cast<_Ty>(_Val << (_Digits - _Remainder)));
	} else if (_Remainder == 0) {
	return _Val;
	} else { // _Remainder < 0
	return _STD rotl(_Val, -_Remainder);
	}
	}

intrin.h declares:

__MACHINE(unsigned char __cdecl _rotl8(unsigned char _Value, unsigned char _Shift))
__MACHINE(unsigned short __cdecl _rotl16(unsigned short _Value, unsigned char _Shift))
__MACHINE(unsigned int __cdecl _rotl(_In_ unsigned int _Value, _In_ int _Shift))
__MACHINE(unsigned __int64 __cdecl _rotl64(_In_ unsigned __int64 _Value, _In_ int _Shift))

__MACHINE(unsigned char __cdecl _rotr8(unsigned char _Value, unsigned char _Shift))
__MACHINE(unsigned short __cdecl _rotr16(unsigned short _Value, unsigned char _Shift))
__MACHINE(unsigned int __cdecl _rotr(_In_ unsigned int _Value, _In_ int _Shift))
__MACHINE(unsigned __int64 __cdecl _rotr64(_In_ unsigned __int64 _Value, _In_ int _Shift))

• We'll need to move these to intrin0.h (physically, intrin0.inl.h) if we want throughput.
• We'll need to make sure this works for Clang.
• <bit> is C++20, so we don't need to worry about /clr:pure (for sure), CUDA (for sure), or the Intel C++ Compiler (I think).

Originally reported as DevCom-1590917 and VSO-1440998 / AB#1440998 .

[AlexGuteniev]

We'll need to move these to intrin0.h (physically, intrin0.inl.h) if we want throughput.

_rotl is in <stdlib.h> as per documentation, I confirm it is in the SDK, can we rely on that instead of having it in <intrin0.h> ?

I suspect yes, as the same was true for _byteswap_ulong et al.

[AlexGuteniev]

Oh, actually not fully useful as there's no 8 bit and 16 bit versions

[StephanTLavavej]

Reopening as the 8-bit and 16-bit cases remain to be done. Internal MSVC-PR-368557 has moved those intrinsics to intrin0.h and will be available in VS 2022 17.2 Preview 2.

[StephanTLavavej]

The 8-bit and 16-bit cases were completed by #2611.