[Bug c/123482] in -frounding-math, GCC moves floating-point operation across fesetround call

[pascal_cuoq at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123482

--- Comment #4 from Pascal Cuoq <pascal_cuoq at hotmail dot com> ---
@Vincent Lefèvre 
Since the rounding mode is set by calling a function, technically, the same
rounding mode can only be in effect everywhere in a program that doesn't call
fesetround or only calls it with FE_TONEAREST.

I understand -frounding-math is partially implemented, but the description of
how it is implemented is not useful if it amounts to “GCC will put ‘false’ in
the hypotheses of the optimizer if the program calls fesetround with an
argument other than FE_TONEAREST”.