On 9/6/21 14:16, Richard Biener wrote:
> On Mon, Sep 6, 2021 at 1:46 PM Jakub Jelinek <jakub@redhat.com> wrote:
>>
>> On Mon, Sep 06, 2021 at 01:37:46PM +0200, Martin Liška wrote:
>>> --- a/gcc/opts.c
>>> +++ b/gcc/opts.c
>>> @@ -1323,6 +1323,14 @@ finish_options (struct gcc_options *opts, struct gcc_options *opts_set,
>>>         = (opts->x_flag_unroll_loops
>>>            || opts->x_flag_peel_loops
>>>            || opts->x_optimize >= 3);
>>> +
>>> +  /* With -fcx-limited-range, we do cheap and quick complex arithmetic.  */
>>> +  if (opts->x_flag_cx_limited_range)
>>> +    flag_complex_method = 0;
>>> +
>>> +  /* With -fcx-fortran-rules, we do something in-between cheap and C99.  */
>>> +  if (opts->x_flag_cx_fortran_rules)
>>> +    flag_complex_method = 1;
>>
>> That should then be opts->x_flag_complex_method instead of flag_complex_method.
>>
>> Ok with that change.
> 
> But the C/C++ langhooks also set flag_complex_method so I fail to see how
> this helps?  As said I was referring to -fcx-limited-range on the command-line
> and -fno-cx-limited-range in the optimize node to undo this which should
> get you the langhook setting of flag_complex_method = 2.

You are right, it's even more complicated as -fno-cx-limited-range is target specific.
Option handling has been introducing surprises every time ...

The following tested patch should handle it.

Ready to be installed?
Thanks,
Martin

> 
>> Note, I think we want to do much more in finish_options and less in
>> process_options, anything that is about Optimization options rather than
>> just the global ones.  Though one needs to be careful with the cases where
>> the code diagnoses something.
>>
>>          Jakub
>>