P3111R0: Atomic Reduction Operations

2024-06-24 SG6 Minutes

Chair: Matthias Kretz

Champion: Gonzalo Brito Gadeschi

Minute Taker: Frank Birbacher

Summary

SG6 has no concerns about the ability of the new atomic add function to have
"fast-math-add" behavior. The main concern, of how this can be worded, should
be clarified with CWG and LWG experts.
It came up that it would be very simple to word if we had a
std::"fast_math_add"(floating-point, floating-point) function...

SG6 gave sufficient feedback and does not need to see another revision of this paper.

SG1 in St Louis

Reduction operations are not a step, but
infinite loops around reductions operations
are not UB (practically implementations can
assume reductions are finite)
SF F N A SA
3 9 1 0 0

The single floating point reduction operations
should allow "fast math" (allow tree reductions),
you can get precise results using fetch_add/sub
SF F N A SA
5 5 1 1 0

Any objection to approve the design of P3111R0
No objection

P3111R1 implemented the changes requested SG1, and we forward to LEWG for C++26
SF F N A SA
3 5 0 0 0
Unanimous consent

SG1 forgot to send to EWG, author/chair emailed to notify.