[PATCH] D128261: [Fortran] Relax relative tolerance for FCVS tests

Mon Aug 15 04:50:23 PDT 2022

rovka added a comment.

In D128261#3718994 <https://reviews.llvm.org/D128261#3718994>, @awarzynski wrote:

> Thanks for continuing to work on this! Just to clarify:
>
> - Is this to land after https://reviews.llvm.org/D131754?

Yes.

> - Is this fix specifically for "gfortran" (i.e. required for "FM905.f" to pass when using "gfortran")?

Yes, this is for gfortran now (I haven't tested other compilers). It also helps with flang if we're passing -fdefault-real-8, although that's not a scenario I'm planning to put effort into supporting right now.

> it might be worth clarifying in the summary too.
>
> This won't be true after https://reviews.llvm.org/D131754, right? (I mean the bit in bold)

Right, sorry! I thought I had updated everything...

>> Since fpcmp treats everything as double precision, **it errors out in flang's case **because flang is printing the minimum number of decimals that would produce the same single precision value
>
> Could you replace "this" with "the output from gfortran":
>
>> Currently, the reference output contains the former, and the relative tolerance is low enough that **this** is being reported as an error.
>
> Sorry for being picky, but this is quite fiddly and easy to get wrong.

No worries, this is good feedback, I'll update!

Repository:
  rT test-suite

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D128261/new/

https://reviews.llvm.org/D128261