This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch, fortran] power.f90 FAIL in testsuite


Paul Brook wrote:
I don't think there is a "correct" place for it. We don't currently do any other transformations like this. It should probably be somehow moved into gfc_arith_power so that it works correctly with arrays.

Agreed. I wasn't certain where the best place to put this was, so I tossed out a patch for comment. I'll pursue this in gfc_arith_power.


If we do any, it makes sense to do it in a way that can easily be extended to cover similar transformations. Similar logic can be applied to x**0, x*0, x+0 and x*1. AFAICS these are optimizations rather than correctness issues.

Agreed. It simply makes sense to use arithmetic identities and axioms to produce better code.


This is an unrelated change, and should really be submitted separately.

Will do.


I don't think the gcc testsuite aims to be a complete correctness testsuite.

My concern is that a lot of people find gfortran inadequate for real world code. Yes, I know they should file bug reports; most have their own fish to fry, and gfortran is more of a "it would be nice if it worked" issue for them than it is a requirement.


Hence, I think gfortran would benefit from a more comprehensive test suite.

Having said that the gfortran testsuite is seriously lacking in some areas. We still have some things that are not tested at all.

Is it permissible to present test cases by themselves, as individual patches to the test suite, without an associated PR?


Ok, I guess. You should also update the comments in teh testcases teh reflect the new split.

Good point.


--
Scott Robert Ladd
site: http://www.coyotegulch.com
blog: http://chaoticcoyote.blogspot.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]