On Mon, Dec 19, 2016 at 6:58 PM, James Greenhalgh
<james.greenha...@arm.com> wrote:
>
>> On Thu, Dec 8, 2016 at 10:44 PM, Uros Bizjak <ubiz...@gmail.com> wrote:
>> > Hello!
>> >
>> > Attached patch fixes fall-out from excess-precision improvements
>> > patch. As shown in the PR, the code throughout the compiler assumes
>> > FLAG_PRECISION_FAST when flag_unsafe_math_optimizations flag is in
>> > effect. The patch puts back two lines, removed by excess-precision
>> > improvements patch.
>> >
>> > 2016-12-08 Uros Bizjak <ubiz...@gmail.com>
>> >
>> > PR middle-end/78738
>> > * toplev.c (init_excess_precision): Initialize flag_excess_precision
>> > to EXCESS_PRECISION_FAST for flag_unsafe_math_optimizations.
>> >
>> > testsuite/ChangeLog:
>> >
>> > 2016-12-08 Uros Bizjak <ubiz...@gmail.com>
>> >
>> > PR middle-end/78738
>> > * gcc.target/i386/pr78738.c: New test.
>> >
>> > Patch was bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.
>> >
>> > OK for mainline?
>>
>> Hmm, I think it belongs to set_unsafe_math_optimization_flags instead
>> (and be consistent if -fexcess-precision was manually specified).
>
> I think it would be better if this were implied by -ffast-math/-Ofast
> than by -funsafe-math-optimizations . That's what I've implemented here,
> and tagged the option as SetByCombined to allow us to honour what the
> user requests.
>
> This should give us the behaviour you were looking for Uros.
>
> I've bootstrapped and tested the behaviour on x86_64, and I've hacked up
> the AArch64 backend to validate that we're setting the flag in the right
> circumstances (but that meant changing the AArch64 behaviour, so isn't
> something we'd want on trunk, and therefore I can't write a testcase for
> this patch).
>
> OK?
Looks good to me, but please also adjust invoke.texi to list -fexcess-precision
as affected by -ffast-math.
Ok with that change.
Richard.
> Thanks,
> James
>
> ---
> 2016-12-19 James Greenhalgh <james.greenhal...@arm.com>
>
> * common.opt (excess_precision): Tag as SetByCombined.
> * opts.c (set_fast_math_flags): Also set
> flag_excess_precision_cmdline.
> (fast_math_flags_set_p): Also check flag_excess_precision_cmdline.
>
