[m-rev.] for review: optimization_options.m

Peter Wang novalazy at gmail.com
Tue Sep 29 14:35:16 AEST 2020


On Mon, 28 Sep 2020 18:51:31 +1000 "Zoltan Somogyi" <zoltan.somogyi at runbox.com> wrote:
> 
> 2020-09-28 18:42 GMT+10:00 "Zoltan Somogyi" <zoltan.somogyi at runbox.com>:
> > 
> > 2020-09-28 17:06 GMT+10:00 "Zoltan Somogyi" <zoltan.somogyi at runbox.com>:
> >> A related issue is: if an optimization level setting causes an integer option
> >> to be set to e.g. 3, then the current code handles a latter setting of that option
> >> to a higher value just fine (using int.max), but ignores any later setting of that
> >> option to a lower value.
> >> 
> >> I will fix this by applying the max *only* to option settings implicit in -O<n>
> >> arguments, not when a specific option is set explicitly. Again, in a separate
> >> diff for further review.
> > 
> > Here is that diff, for post-commit review by Peter.
> 
> That would work better with actual attachments :-(
> 
> Zoltan.

> Make --optx=N work after -O... sets it to >N.
> 
> tools/make_optimization_options_middle:
> tools/make_optimization_options_end:
>     Differentiate between an option being set explicitly and an option
>     being set implicitly by -O<N>. In the explicit case, always set integer
>     values exactly as they are, which allows such options to both increase
>     and decrease their values. In the implicit case, set the maximum of
>     the old and new values, which allows -O<N> to increase their values
>     but not to decrease them.
> 
> compiler/optimization_options.m:
>     Update the auto-generated code.

That's fine.

Peter


More information about the reviews mailing list