[llvm-dev] Issues with new Attributor (replaceAllUses fails with type mismatch)

[Johannes Doerfert via llvm-dev]

Hi Suresh,


thanks for reporting this!

I thought I fixed this with 8d94d3c3b44c3a27a69b153cef9be4b8e481150e.

Did you run before or after that commit?


Cheers,

Johannes


On 5/15/20 7:17 AM, Mani, Suresh via llvm-dev wrote:
> [AMD Public Use]
>
> Hi ,
>
> Please ignore the earlier header of Internal and Official use only.
>
> Thanks
> M Suresh
>
> From: llvm-dev <llvm-dev-bounces at lists.llvm.org> On Behalf Of Mani, Suresh via llvm-dev
> Sent: Friday, May 15, 2020 4:37 PM
> To: llvm-dev at lists.llvm.org
> Subject: [llvm-dev] Issues with new Attributor (replaceAllUses fails with type mismatch)
>
> [CAUTION: External Email]
>
> [AMD Official Use Only - Internal Distribution Only]
>
> Hi ,
>
> There seems to be some issue with attributor, exactly in  File Attributor.cpp,  Function Attributor::rewriteFunctionSignatures(),  Line No: 1600 (approrimate).
>
> The llvm source code at above address is as follows:
>
>      // Eliminate the instructions *after* we visited all of them.
>      for (auto &CallSitePair : CallSitePairs) {
>        CallBase &OldCB = *CallSitePair.first;
>        CallBase &NewCB = *CallSitePair.second;
>        ModifiedFns.insert(OldCB.getFunction());
>        CGUpdater.replaceCallSite(OldCB, NewCB);
>        OldCB.replaceAllUsesWith(&NewCB);   <---------------------------------------- Failing location
>        OldCB.eraseFromParent();
>      }
>
> Failure symptom:
> Assertion failed: New->getType() == getType() && "replaceAllUses of value with new value of different type!"
>
> Snaps for IR from failing llvm IR and debug info while debugging is as follows:
>
> Sample situation1:
> Original code in IR --          %133 = call i32 (i32, ...) bitcast (void (i32)* @set_yydebug to i32 (i32, ...)*)(i32 1) #21
> Attribute info ----                attributes #21 = { nounwind }
>
> Debug info:
> OldCB--------------------  %133 = call i32 (i32, ...) bitcast (void (i32)* @0 to i32 (i32, ...)*)(i32 1) #27
> NewCB-------------------- call void @set_yydebug() #27
>
>
> OldCB.replaceAllUsesWith(&NewCB);   fails because of a type mismatch b/w OldCB (i32) and NewCB (void)  .
> ------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
>
> Sample situation 2:
>
> Original code in IR --     %63 = tail call %struct.cv* bitcast (%struct.cv.231* (i8, i64, i32*)* @Perl_find_runcv_where to %struct.cv* (i8, i64, i32*)*)(i8 zeroext 1, i64 %62, i32* null) #14
> Attribute info ----   attributes #14 = { nounwind }
>
> Debug info:
> OldCB --- %63 = tail call %struct.cv.231* @Perl_find_runcv_where(i8 zeroext 1, i64 %62) #14
> NewCB ---%64 = tail call %struct.cv* bitcast (%struct.cv.231* (i8, i64, i32*)* @15 to %struct.cv* (i8, i64, i32*)*)(i8 zeroext 1, i64 %62, i32* noalias align 536870912 null) #14
>
> And hence OldCB.replaceAllUsesWith(&NewCB);   fails
> ------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> The complete IR file is very huge, and had no luck in getting a reduced test case.
>
> Thanks
> M Suresh
>
>
>
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200515/0d6575e3/attachment.html>