[LLVMdev] live information

Jakob Stoklund Olesen stoklund at 2pi.dk
Mon Jul 25 09:39:53 PDT 2011

On Jul 25, 2011, at 7:18 AM, Jonas Paulsson wrote:

> I'm working on live intervals and I'm curious about how LLVM typically would represent live info. Take the following case (after SimpleRegisterCoalescer):
> BB#3: derived from LLVM BB %bb.nph
>     Live Ins: %a0_h
>     Predecessors according to CFG: BB#2
>     %reg16500<def> = COPY %reg16499
>     Successors according to CFG: BB#4
> BB#4: derived from LLVM BB %for.body21
>     Live Ins: %a0_h
>     Predecessors according to CFG: BB#3 BB#4
>     %reg16500<def> = add_twoAddr %reg16500, 1
>     cmp_branch <BB#4>, %reg16500, %a0_h
>     Successors according to CFG: BB#4 BB#5
> BB#4 is a loop.
> %reg16500 is first defined by a COPY, then updated at each iteration and used at the end as well of BB#4. 
> Live ins(BB#4) should include %reg16500 (Simple register coalescer did not care for this)
> LiveRanges should be from the def in BB#3 to end of BB#3, from start of BB#4 to the add MI, from the add MI to the compare&branch MI.
> Am I correct?

The live-in lists on basic blocks are only used for physical registers. The are filled in after register allocation is complete by RegAllocBase::addMBBLiveIns.

> I am trying to recompute live-info after the coalescer, so I also wonder if any or both of the add and cmp_branch MI's should have kill-flags for %reg16500?

Kill flags are very difficult to keep updated when coalescing and allocating registers, so they are simply cleared by the coalescer and reinserted afterwards by LiveIntervalAnalysis::addKillFlags.

Thinking about it, it may make sense to also use kill flags for physical registers only.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20110725/3e8071f4/attachment.html>

More information about the llvm-dev mailing list