[cfe-commits] [PATCH] Add -fexcess-precision option.

[Chandler Carruth]

On Thu, Jun 21, 2012 at 2:34 PM, Lang Hames <lhames at gmail.com> wrote:

> Hi Hal,
>
> I object to 'never' because there may be cases in the future where we
>> can determine that extra precision cannot change the result so we can
>> use fma while maintaining strict standard compliance. 'strict' is much
>> better.
>>
>
> That's two votes for 'strict'. I'm happy with that. :)
>
> For an option whose job it is to set the default FP_CONTRACT state, I
>> think that -ffp-contract=[no|yes|default] is the best. This makes the
>> one-to-one correspondence clear to the user.
>>
>
> This flag isn't intended to set the default FP_CONTRACT behaviour, it's
> about providing FMA formation modes outside those described by FP_CONTRACT,
> in particular enabling aggressive formation of FMAs across expression
> boundaries. I had intended enable this aggressve FMA formation under the
> -fexcess-precision flag, but as Steve pointed out that flag is overly
> broad.
>
> So, proposal:
>
> New flag: -ffuse-fp-ops={fast,standard,strict}.  Default value will be
> 'standard', which means following FP_CONTRACT rules.
>
> When -ffuse-fp-ops=fast we form FMAs wherever they're profitable,
> regardless of the FP_CONTRACT state.
> When -ffuse-fp-ops=strict we form FMAs iff/when we can prove that the
> excess precision won't change the result (which to start with will mean
> never forming FMAs).
>
> -ffast-math implies -ffuse-fp-ops=fast
>

This is all fantastic.


> -fexcess-precision implies -ffuse-fp-ops=fast
>

I worry about this one, but you seem more confident, and so do others, so
carry on! =] My worries are very minor here.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20120621/41004d33/attachment.html>