[llvm] r224583 - use -0.0 when creating an fneg instruction

[Sanjay Patel]

Hi David,

Yes - that's the intent of this follow-on patch:
http://reviews.llvm.org/D6731

On Fri, Dec 19, 2014 at 10:05 AM, David Majnemer <david.majnemer at gmail.com>
wrote:
>
> This improves things when forming fneg but what if we see fsub 0.00, %V in
> the IR? Shouldn't we add a canonicalization for that?
>
> On Fri, Dec 19, 2014 at 11:44 AM, Sanjay Patel <spatel at rotateright.com>
> wrote:
>>
>> Author: spatel
>> Date: Fri Dec 19 10:44:08 2014
>> New Revision: 224583
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=224583&view=rev
>> Log:
>> use -0.0 when creating an fneg instruction
>>
>> Backends recognize (-0.0 - X) as the canonical form for fneg
>> and produce better code. Eg, ppc64 with 0.0:
>>
>>    lis r2, ha16(LCPI0_0)
>>    lfs f0, lo16(LCPI0_0)(r2)
>>    fsubs f1, f0, f1
>>    blr
>>
>> vs. -0.0:
>>
>>    fneg f1, f1
>>    blr
>>
>> Differential Revision: http://reviews.llvm.org/D6723
>>
>>
>> Modified:
>>     llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp
>>     llvm/trunk/test/Transforms/InstCombine/fast-math.ll
>>
>> Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp?rev=224583&r1=224582&r2=224583&view=diff
>>
>> ==============================================================================
>> --- llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp (original)
>> +++ llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp Fri Dec
>> 19 10:44:08 2014
>> @@ -759,7 +759,7 @@ Value *FAddCombine::createFSub(Value *Op
>>  }
>>
>>  Value *FAddCombine::createFNeg(Value *V) {
>> -  Value *Zero = cast<Value>(ConstantFP::get(V->getType(), 0.0));
>> +  Value *Zero =
>> cast<Value>(ConstantFP::getZeroValueForNegation(V->getType()));
>>    Value *NewV = createFSub(Zero, V);
>>    if (Instruction *I = dyn_cast<Instruction>(NewV))
>>      createInstPostProc(I, true); // fneg's don't receive instruction
>> numbers.
>>
>> Modified: llvm/trunk/test/Transforms/InstCombine/fast-math.ll
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/fast-math.ll?rev=224583&r1=224582&r2=224583&view=diff
>>
>> ==============================================================================
>> --- llvm/trunk/test/Transforms/InstCombine/fast-math.ll (original)
>> +++ llvm/trunk/test/Transforms/InstCombine/fast-math.ll Fri Dec 19
>> 10:44:08 2014
>> @@ -93,7 +93,7 @@ define float @fold9(float %f1, float %f2
>>    ret float %t3
>>
>>  ; CHECK-LABEL: @fold9(
>> -; CHECK: fsub fast float 0.000000e+00, %f2
>> +; CHECK: fsub fast float -0.000000e+00, %f2
>>  }
>>
>>  ; Let C3 = C1 + C2. (f1 + C1) + (f2 + C2) => (f1 + f2) + C3 instead of
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141219/8909b1bb/attachment.html>