[llvm-commits] CVS: llvm/lib/CodeGen/LiveIntervalAnalysis.cpp RegAllocLinearScan.cpp VirtRegMap.cpp VirtRegMap.h

[Evan Cheng]

On Mar 30, 2007, at 2:44 PM, Chris Lattner wrote:

>> First cut trivial re-materialization support.
>
> Woot!
>
>>              unsigned reg = rep(mop.getReg());
>>              mii->getOperand(i).setReg(reg);
>>
>> +            // If the definition instruction is re- 
>> materializable, its spill
>> +            // weight is zero.
>>              LiveInterval &RegInt = getInterval(reg);
>> -            RegInt.weight +=
>> -              (mop.isUse() + mop.isDef()) * pow(10.0F, (int) 
>> loopDepth);
>> +            if (!RegInt.remat) {
>> +              RegInt.weight +=
>> +                (mop.isUse() + mop.isDef()) * pow(10.0F, (int) 
>> loopDepth);
>> +            }
>
> Should it really be zero?  It should definitely be lower that a  
> normal value, but zero seems low.  The loop depth should still be  
> taken into consideration.

Funny you should mention this. I am re-evaluating this right now. My  
thinking is loads (load from constantpool or stack slots) would have  
the normal weight. The rest will have half the weight.

Evan

>
>> @@ -497,6 +515,9 @@
>>      }
>>
>>    } else {
>> +    // Can't safely assume definition is rematierializable anymore.
>
> typo in comment.
>
>> @@ -53,14 +60,20 @@
>>      /// read/written by this instruction.
>>      MI2VirtMapTy MI2VirtMap;
>>
>> +    /// ReMatMap - This is irtual register to re-materialized  
>> instruction
> typo
>
>> +    /// mapping. Each virtual register whose definition is going  
>> to be
>> +    /// re-materialized has an entry in it.
>> +    std::map<unsigned, const MachineInstr*> ReMatMap;
>
>> +    /// ReMatId - Instead of assigning a stack slot to a to be  
>> rematerialized
>> +    /// virtaul register, an unique id is being assinged. This  
>> keeps track of
> 2 x typo
>
>> +    /// the highest id used so far. Note, this starts at (1<<18)  
>> to avoid
>> +    /// conflicts with stack slot numbers.
>> +    int ReMatId;
>
> -Chris
>
>