[llvm-commits] [llvm] r67363 - in /llvm/trunk: include/llvm/Target/TargetOptions.h lib/Target/TargetMachine.cpp

[Mon Ping Wang]

Hi Chris,

We have some applications that wants less precise fma being generated  
but don't want unsafe math on.

   -- Mon Ping

On Mar 20, 2009, at 9:29 AM, Chris Lattner wrote:

>
> On Mar 19, 2009, at 11:04 PM, Mon Ping Wang wrote:
>
>> Hi Chris,
>>
>> This is to support FMAs that can generate less precise results
>> primarily due to how it handles rounding, e.g., some architecture  
>> will
>> truncate the result of a*b before adding it to c.
>
> Do we need a new flag just for this, or is unsafe math enough?  Is
> there an application that wants this flag but not unsafe math?
>
> -Chris
>
>>
>>
>> -- Mon Ping
>>
>>
>> On Mar 19, 2009, at 10:36 PM, Chris Lattner wrote:
>>
>>>
>>> On Mar 19, 2009, at 10:06 PM, Mon P Wang wrote:
>>>
>>>> Author: wangmp
>>>> Date: Fri Mar 20 00:06:58 2009
>>>> New Revision: 67363
>>>>
>>>> URL: http://llvm.org/viewvc/llvm-project?rev=67363&view=rev
>>>> Log:
>>>> Added option to enable generating less precise mad (multiply
>>>> addition)
>>>> for those architectures that support the instruction.
>>>
>>> Hi Mon Ping,
>>>
>>> Is this specifically to support FMA instructions that elide the
>>> rounding step, or is this for "even less precise" FMAs?
>>>
>>> -Chris
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits