On Tue, Feb 25, 2020 at 09:28:59AM +0100, Tobias Burnus wrote:
> Thanks for the cleanup. Committed (with "git commit --author 'St...'" + git
> push) as Rev. r10-6836-gfe86f537f010828b7f77ff459cef138e84071660
On Wed, Feb 26, 2020 at 01:51:48PM -0500, Fritz Reese wrote:
> In fact it is a logic error that the code is not used, so the
> preferred patch would be to fix it, rather than remove it...
> That being said the solution to PR 93871
> [https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93871], which I plan to
> attend to in the coming week(s), will likely touch and possibly
> replace this code anyway.
I posted a draft patch that replaces the current
implementation with one that brings it in line with how
gfortran handles all other intrinsic subprograms (except
for a few in IEEE_ARITHMETIC). I've identified and fix
a few issues with that patch. The patch takes what you
have done and essentially in-lines the degree trig functions
during conversion by trans-intrinsic.c.
The only missing feature in my WIP is better argument
reduction or transformations as suggested by Thomas. I
can send you a diff later today if you would like to use
it as a starting point.