Hi,
FYI, people interested in this thread might be interested in
pgsql-bugs #18610. There are two related issues here:
1. Some people want to use LSE on modern ARM servers so they want to
use -moutline-atomics, which IIUC adds auto-detection logic with
fallback code so it can still run on the first generation of aarch64
ARMv8 (without .1) hardware. That was being discussed as a feature
proposal for master. (People could already access that if they
compile from source by using -march=something_modern, but the big
distributions are in charge of what they target and AFAIK mostly still
choose ARMv8, so this outline atomics idea is a nice workaround to
make everyone happy, I haven't studied exactly how it works.)
2. Clang has started assuming -moutline-atomics in some version, so
it's already compiling .bc files that way, so it breaks if our JIT
system decides to inline SQL-callable functions, so we'll need to
decide what to do about that and back-patch something. Conservative
choice would be to stop it from doing that with -mno-outline-atomics,
until this thread makes progress, but perhaps people closer to the
subject have another idea...
[1] https://www.postgresql.org/message-id/flat/18610-37bf303f904fede3%40postgresql.org
