[llvm-commits] [PATCH] Add support for lowering FNEG with AVX

[Chad Rosier]

On Dec 14, 2011, at 4:53 PM, Bruno Cardoso Lopes wrote:

> On Wed, Dec 14, 2011 at 10:39 PM, Jakob Stoklund Olesen <stoklund at 2pi.dk> wrote:
>> 
>> On Dec 14, 2011, at 4:37 PM, Chad Rosier wrote:
>> 
>>> 
>>> On Dec 14, 2011, at 4:27 PM, Bruno Cardoso Lopes wrote:
>>>> 
>>>> +    if (NumBits == 256) {
>>>> +      CV.push_back(C);
>>>> +      CV.push_back(C);
>>>> +    }
>>>>  } else {
>>>>    Constant *C = ConstantFP::get(*Context, APFloat(APInt(32, 1U << 31)));
>>>>    CV.push_back(C);
>>>>    CV.push_back(C);
>>>>    CV.push_back(C);
>>>>    CV.push_back(C);
>>>> +    if (NumBits == 256) {
>>>> +      CV.push_back(C);
>>>> +      CV.push_back(C);
>>>> +      CV.push_back(C);
>>>> +      CV.push_back(C);
>>>> +    }
>>>> 
>>>> If you iterate over the number of elements to push_back the constants,
>>>> you can make it more general and don't need to get the size here.
>>> 
>>> Sure, easy fix.
>> 
>> Note that SmallVector has these methods:
>> 
>>  void append(size_type NumInputs, const T &Elt);
>>  void assign(unsigned NumElts, const T &Elt);
> 
> Better yet! :)
> 

Committed revision 146625.

Thanks, Bruno and Jakob.

 Chad

> -- 
> Bruno Cardoso Lopes
> http://www.brunocardoso.cc