@mattkretz Numerics can look at this first.

P2746R1 Deprecate and Replace Fenv Rounding Modes (Hans Boehm)

P2746R2 Deprecate and Replace Fenv Rounding Modes (Hans Boehm)

SG6 Varna '23

Minutes

Summary:

• Look for a different name than cr_ prefix, which C reserved for correctly rounded cmath functions.
• Consider dropping constexpr to make the feature less expensive to implement because we expect the value to be (very) low.
• Relate to from_chars for the cr_const function.
• Consider intended interaction with denormals flush-to-zero / treat-as-zero.
• Consider relation to existing numeric_limits::is_iec559 (storage layout vs. behaviour, and how do fast-math-flags tie in with this)
• We could do simd overloads for this, especially since SIMD instruction encodings may include the means to implement this efficiently. But the cost vs. utility tradeoff doesn't make us confident we want to impose the cost on the committee and implementers. We also prefer to simply ignore vector/batched operations in favor of QoI: compilers could drop unnecessary global rounding mode changes in the optimizer.

P2746R3 Deprecate and Replace Fenv Rounding Modes (Hans Boehm)

P2746R4 Deprecate and Replace Fenv Rounding Modes (Hans Boehm)

P2746R4: Deprecate and Replace Fenv Rounding Modes

2024-03-19 SG6 Minutes

Chair: Lisa Lippincott

Champion: Hans Boehm

Polls

POLL: We intend to leave floating-point exception behavior unspecified.

# of Authors: 1
Author Position: F
Outcome: Consensus

POLL: We would like to remove (at some unspecified time) fesetround and
fegetround from C++.

# of Authors: 1
Author Position: SF
Outcome: Consensus

POLL: Forward P2746R4 with the following changes to LEWG:

• Add constexpr to the constructor
• const member functions
• Add usage examples
• Make to_chars as conforming to IEC 60559 as C's printf

# of Authors: 1
Author Position: SF
Outcome: Consensus

P2746R5 Deprecate and Replace Fenv Rounding Modes (Hans Boehm)

P2746R5: Deprecate and Replace Fenv Rounding Modes

P2746R5: Deprecate and Replace Fenv Rounding Modes

2024-06-28 Library Evolution St. Louis Minutes

Champion: Hans Boehm

Chair: Ben Craig & Eddie Nolan

Minute Taker: Robert Leahy

Summary

POLL: We should promise more committee time to pursuing a library controlling floating point rounding locally, similar to P2746R5, knowing that our time is scarce and this will leave less time for other work. This poll does not encompass deprecation questions.

Attendance: 14 + 5

# of Authors: 1

Author Position: SF

Outcome: Consensus in favor

POLL: We should promise more committee time to pursuing fesetround and fegetround deprecation in the near term, knowing that our time is scarce and this will leave less time for other work.

Attendance: 14 + 5

# of Authors: 1

Author Position: SF

Outcome: Consensus in favor

Next Steps

• Get implementation experience.
• Work on wording
• Get feedback from SG22 / C liason about the deprecation of fesetround and fegetround. Provide SG22 poll results as inputs to the LEWG deprecation decision.

Related to "P3375: Reproducible floating-point results" (to be published on Sep ML).

P2746R6 Deprecate and Replace Fenv Rounding Modes (Hans Boehm)

After reviewing P3375R1, SG6 determined there's no need to revisit this paper. It's good to go for LEWG.