https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124819
--- Comment #9 from Christopher Albert <albert at tugraz dot at> --- (In reply to Jerry DeLisle from comment #8) > (In reply to Christopher Albert from comment #7) > > Created attachment 64177 [details] > > testsuite-only patch: relax bessel_6 tolerance near x=475.78 > > > > The testcase point at x = 475.78 is worth keeping, but the final tolerance > > entry is too tight for the generic non-FMA path. > > > > This testsuite-only patch changes the last factor from 15 to 32 and keeps > > the same evaluation point. > > > > Tested with the existing build via direct runtest: > > - `cd /tmp/gcc-bessel-bisect-build/gcc/testsuite/gfortran && runtest --tool > > gfortran dg.exp=bessel_6.f90` -> 12 expected passes > > - `cd /tmp/gcc-bessel-bisect-build/gcc/testsuite/gfortran && runtest --tool > > gfortran dg.exp=bessel_7.f90` -> 12 expected passes > > If there is no failure in the testsuite now, there is no point to changing > the testcase. The problem is, I still frequently hit this problem when compiling and testing locally on my machine - no idea why this configuration is so sensitive to it. Should I just apply FMA flags then per default?
