[LLVMdev] Value liveout (uses)

[Dan Gohman]

The pass you're referring to is in include/llvm/Analysis/LiveValues.h
and lib/Analysis/LiveValues.cpp. It computes conservative
approximations for specific liveness queries, rather than full
livein/liveout information. It's intended to be used as a heuristic.

Dan

On May 30, 2009, at 3:51 PM, Evan Cheng wrote:

> I believe Dan has added a pass to compute livein / liveout values.
>
> Evan
>
> Sent from my iPhone
>
> On May 30, 2009, at 5:03 AM, Rotem Varon <varonrotem at gmail.com> wrote:
>
>> Thank you.
>>
>> Is it possible to determine the liveout of the operands (see  
>> example bellow) ?
>>
>>         %5 = add i32 %4, %3
>>
>> For '%5': i can simply use " i->isUsedOutsideOfBlock() "
>> For '%3' and '%4' : this is the question ...
>>
>> From your answer, is it possible to determine which value is  
>> liveout ( in binary instruction)?
>>
>>
>> On Sat, May 30, 2009 at 2:57 AM, John McCall <rjmccall at apple.com>  
>> wrote:
>> On May 29, 2009, at 11:37 PM, Rotem Varon wrote:
>> > How can i know, if a value have uses outside of the current basic
>> > block (liveout), without iterating through all the basic block ?
>>
>> If the value is created within the basic block in question, and the
>> block doesn't loop to itself, then you can just iterate through the
>> uses and note if the use is an instruction in a different block:
>>
>> bool isLiveOut(Instruction* I) {
>>        BasicBlock* I_BB = I->getParent():
>>        for (Value::use_iterator ui = I->use_begin(), ue = I- 
>> >use_end(); ui !
>> = ue; ++ui)
>>                if (cast<Instruction>(ui)->getParent() != I_BB)
>>                        return true;
>>        }
>> }
>>
>> If the value is created within the block but the block loops to
>> itself, you can get away with a very slight adjustment:
>>
>>                if (cast<Instruction>(ui)->getParent() != I_BB && ! 
>> isa<PHINode>(ui))
>>
>> If the value is created in one block and you want to know if it's  
>> live
>> out of some dominated block, that's a lot more complicated (and
>> expensive).
>>
>> John.
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev