https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124819

--- Comment #13 from Christopher Albert <albert at tugraz dot at> ---
(In reply to anlauf from comment #12)
> bessel_6.f90 has already received multiple updates.
> 
> Let's make a choice: increase the tolerance as done previously (multiple
> times),
> or choose a different value to test (-> comment#11).

a) What speaks for increasing tolerance and keeping edge case: Subtle behaviour
changes from gcc code will be picked up with high sensitivity. 

b) What speaks for choosing a different value to test: Subtle behavior changes
from hardware floating point arithmetics will not break tests anymore.

So what's realistically more probable? From the history of this problem I tend
to believe that when we have a test that is affected by FMA vs non-FMA, subtle
hardware behavior is more risky than breaking exactly this one test-case but
nothing else by code changes in gcc.

So I tend to choose a different value to test after giving it some thought.

Reply via email to