[PATCH v5 06/12] configure: Only use -fno-fp-int-builtin-inexact if compiler supports it

[Alexander Monakov]

On Thu, 20 Nov 2025, Sam James wrote:

> Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:
> 
> > Checked on x86_64-linux-gnu.
> 
> Reviewed-by: Sam James <sam@gentoo.org>

I wonder if this only covers the (mechanical) change done in the patch, or is
really meant as an acknowledgement that omitting -fno-fp-int-builtin-inexact
is the proper solution here.

The proposed commit message does not seem to explain it.

I guess a part of why it works is because a previous (already pushed) patch
enabled -ftrapping-math for Clang, making its floating-point codegen very
conservative. But that cannot be the complete explanation.

If the idea is that if there are targets where LLVM emits instructions with
undesired effect on floating-point exception flags, the tests are expected
to catch it, might be nice to say so in the commit message (along with the
list of targets tested so far).

Alexander