P0543R3 Saturation Arithmetic

[StephanTLavavej]

WG21-P0543R3 Saturation Arithmetic
WG21-P4052R0 Renaming Saturation Arithmetic Functions

Feature-test macros:

#define __cpp_lib_saturation_arithmetic 202603L
#define __cpp_lib_freestanding_numeric  202311L

Note

The freestanding macro was added with the value 202311L by cplusplus/draft#6738 and was not increased by cplusplus/draft#8877.

Separately, it was later increased to 202502L by cplusplus/draft#7706 for #5298 WG21-P2976R1 "Freestanding Library: <algorithm>, <numeric>, <random>, And More".

Note: We are not yet accepting PRs for C++26 features.

[frederick-vs-ja]

cplusplus/draft@2b1867a indicates that there's another new freestanding feature-test macro.

#define __cpp_lib_freestanding_numeric 202311L

[StephanTLavavej]

MSVC compiler intrinsics are expected to be available by the time we implement this:

_sat_add_i8
_sat_add_i16
_sat_add_i32
_sat_add_i64
_sat_add_u8
_sat_add_u16
_sat_add_u32
_sat_add_u64
_sat_sub_i8
_sat_sub_i16
_sat_sub_i32
_sat_sub_i64
_sat_sub_u8
_sat_sub_u16
_sat_sub_u32
_sat_sub_u64

We'll need to check whether they're available on all platforms (@kbrady-msft's MSVC-PR-548218 said that codegen is initially available for x64 only), and we may need to declare them in intrin.h or intrin0.inl.h before we begin using them.

[H-G-Hristov]

I have implemented P0543R3 in libc++ a while ago. Maybe I can port it over to MS STL. What I don't understand well. If Clang builds MS STL does it need to provide the above MSVC intrinsics or will it need to use it's own __builtin_*_overflow? So the STL implementation will have a code path with the above MSVC intrinsics (which I assume are not yet available) and another for Clang? Is that so?

[StephanTLavavej]

In general we like to use the same builtins for all compilers, but we can use different builtins for Clang.