[llvm-dev] Print Register Liveness Information

[Anitha B Gollamudi via llvm-dev]

On 16 February 2017 at 17:48, Quentin Colombet <qcolombet at apple.com> wrote:
>
> Good point!
> Additionally this might gets invalidated by some passes (look for MachineRegisterInfo::invalidateLiveness)

Thanks for the pointers! What is the rational behind invalidating
register liveness info? Does it consume lot of memory during compilation?

-Anitha


>
> On Feb 16, 2017, at 2:32 PM, Matthias Braun <mbraun at apple.com> wrote:
>
> Note that this only works for targets that say TargetRegisterInfo::trackLivenessAfterRegAlloc()== true.
>
> On Feb 16, 2017, at 1:21 PM, Quentin Colombet via llvm-dev <llvm-dev at lists.llvm.org> wrote:
>
> Hi Anitha,
>
> The liveness information is usually dropped after regalloc.
> Therefore the short answer is no, this information is not available before producing assembly.
>
> However, we have the live-ins and you can request the kill flag to be set. From both this information, you can recompute the liveness.
> I’d suggest you use the LivePhysReg helper class and use its result to annotate your assembly (emitComment or something along those lines).
>
> Cheers,
> -Quentin
>
> On Feb 16, 2017, at 12:46 PM, Anitha B Gollamudi via llvm-dev <llvm-dev at lists.llvm.org> wrote:
>
> Hi,
>
> Is there a way to print the register liveness information into the assembly file or as some intermediate listing?
> To be more precise, I am looking at figuring out the following information in an assembly file/listing.
>
> Say, for simplicity, our register set is only { rsp, rbp, eax }
>
>
> subq    $16, %rsp     // rsp is killed here, eax and rbp are live
> movl    $0, -4(%rbp)  // rbp is killed here, rsp and eax are live
> xorl    %eax, %eax    // eax is killed here, rsp and rbp are live
> movl    $1, %eax   // eax is killed here, rsp and rbp are live
>
> Is there a way LLVM stores this information somewhere before producing assembly?
> I haven't  dug into the source code yet, but intuitively the information must be there and hence maybe some pass dumps that info?
>
> Thanks!
>
>  Anitha
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
>



-- 
 Anitha