Looks like a straight forward mechanical change. Probably doesn't need a review.
OK. -- steve On Thu, Sep 05, 2019 at 09:55:58PM +0200, Harald Anlauf wrote: > Ping! > > On 08/28/19 21:58, Harald Anlauf wrote: > > Hi Bernhard, > > > > On 08/28/19 20:57, Bernhard Reutner-Fischer wrote: > >> I see that you copied the unfortunate error-message "commence a loop" > >> and i see that i completely forgot to adjust it as per Mike's > >> preference in > >> https://gcc.gnu.org/ml/fortran/2015-05/msg00166.html > >> > >> So can you please change your new errors (and the unroll error message > >> too) to something like the suggested > >> "directive not at the start of a loop at %C" please? > >> > >> Many thanks in advance! > >> PS: This is border obvious, i'd send the patch for review anyway, > >> maybe a native speaker can provide a better wording. > > > > see attachment and below for Changelog. Regtested OK. > > > > I have opted for the variant "directive not at the start of a loop", > > which is what Mike preferred, and what matches the current capabilities. > > > > Is that OK? > > > >> PPS: I'm still a bit unhappy about the following kludges in unroll even > >> more so since you copied the concept: > >> (1) The globals to diagnose misplaced directives are very ugly. > >> (2) putting the payload into gfc_iterator is ugly, too. I don't > >> remember offhand how ugly or intrusive it would be to provide means for > >> passing down an additional optional structure to act as sink for the > >> directive payload data. Putting those into the iterator is AFAIR not > >> all that clean. Maybe you could have a look if you can extend this area > >> to look less clumsy? "hinted_iterator" maybe, or maybe the iterator is > >> seldomly used anyway and hence we don't care? > > > > Could you please open a new PR for this? > > > > There are many things that could be improved in this context. > > There is a lot that could be learned from e.g. the Cray or NEC > > compilers, where you can explicitly control loop fusion, loop > > blocking, which are very important in (my) practice. > > > > The current implementation also does not support array notation, > > which I consider a major limitation. I'll have to learn how this > > could be done. > > > >> thanks and cheers, > >> > > > > Thanks for the feedback! > > > > Harald > > > > 2019-08-28 Harald Anlauf <anl...@gmx.de> > > > > PR fortran/91496 > > * parse.c (parse_executable): Improve error messages for > > improperly placed pragmas not preceeding a loop. > > > > 2019-08-28 Harald Anlauf <anl...@gmx.de> > > > > PR fortran/91496 > > * gfortran.dg/directive_unroll_5.f90: Adjust error message. > > > -- Steve 20170425 https://www.youtube.com/watch?v=VWUpyCsUKR4 20161221 https://www.youtube.com/watch?v=IbCHE-hONow
