llvm dev - Jan 2019 - Problem using BlockFrequencyInfo's getBlockProfileCount

Hey,

I am trying to use the BlockFrequencyInfoWrapperPass to obtain hotness 
information in my LLVM pass. Attached is a minimal example of code which 
creates a SIGSEGV. The pass calls 
AU.addRequired<BlockFrequencyInfoWrapperPass>(); in 
getAnalysisUsage(..). The problem exists with changed and unchanged IR.

The binary is instrumented like this: clang input.bc -fprofile-generate 
-o output

The binary runs with selected inputs and the profiles are merged: 
llvm-profdata merge input.profraw -output=output.prof

Then opt runs with arguments: opt input.bc -o output.bc -load mypass.so 
-block-freq -pgo-instr-use -pgo-test-profile-file=output.prof 
-profile-sample-accurate -mypass

Is this a bug or am can someone provide an example on how to use 
BlockFrequencyInfo correctly?


Example code:


for (auto& F : M) {
     if (F.isDeclaration()) {
         continue;
     }
     auto& bfiPass = getAnalysis<BlockFrequencyInfoWrapperPass>(F);
     llvm::BlockFrequencyInfo* BFI = &bfiPass.getBFI();

     //Works - shows all info I want
     BFI->print(llvm::dbgs());

     for (const llvm::BasicBlock& B : F) {
         //Fails with SIGSEGV -> *getFunction() returns 0xa
         dbgs() << BFI->getBlockProfileCount(&B).getValueOr(0)
<< "\n";
     }
}


lib/Analysis/BlockFrequencyInfo.cpp#L211

return BFI->getBlockProfileCount(*getFunction(), BB);

On Sun, Jan 13, 2019 at 9:24 AM Dennis Fischer via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Hey,
>
> I am trying to use the BlockFrequencyInfoWrapperPass to obtain hotness
> information in my LLVM pass. Attached is a minimal example of code which
> creates a SIGSEGV. The pass calls
> AU.addRequired<BlockFrequencyInfoWrapperPass>(); in
> getAnalysisUsage(..). The problem exists with changed and unchanged IR.
>
> The binary is instrumented like this: clang input.bc -fprofile-generate
> -o output
>
> The binary runs with selected inputs and the profiles are merged:
> llvm-profdata merge input.profraw -output=output.prof
>
> Then opt runs with arguments: opt input.bc -o output.bc -load mypass.so
> -block-freq -pgo-instr-use -pgo-test-profile-file=output.prof
> -profile-sample-accurate -mypass
>
> Is this a bug or am can someone provide an example on how to use
> BlockFrequencyInfo correctly?
>
>
> Example code:
>
>
> for (auto& F : M) {
>      if (F.isDeclaration()) {
>          continue;
>      }
>      auto& bfiPass =
getAnalysis<BlockFrequencyInfoWrapperPass>(F);
>      llvm::BlockFrequencyInfo* BFI = &bfiPass.getBFI();
>
>      //Works - shows all info I want
>      BFI->print(llvm::dbgs());
>
I don't know offhand why your code below is segfaulting, but note that
BFI->print() will dereference the F pointer just as *getFunction below
will. I would look in the debugger to see what the value of F is during
print() since it seems to be ok there, and then see if you can figure out
where it is changing if it is returning a bad value below.
Teresa

>      for (const llvm::BasicBlock& B : F) {
>          //Fails with SIGSEGV -> *getFunction() returns 0xa
>          dbgs() << BFI->getBlockProfileCount(&B).getValueOr(0)
<< "\n";
>      }
> }
>
>
> lib/Analysis/BlockFrequencyInfo.cpp#L211
>
> return BFI->getBlockProfileCount(*getFunction(), BB);
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>

-- 
Teresa Johnson |  Software Engineer |  tejohnson at google.com |
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190114/711530e0/attachment.html>

Hey,

I am trying to use the BlockFrequencyInfoWrapperPass to obtain hotness
information in my LLVM pass. Attached is a minimal example of code which
creates a SIGSEGV. The pass calls
AU.addRequired<BlockFrequencyInfoWrapperPass>(); in
getAnalysisUsage(..). The problem exists with changed and unchanged IR.

The binary is instrumented like this: clang input.bc -fprofile-generate
-o output

The binary runs with selected inputs and the profiles are merged:
llvm-profdata merge input.profraw -output=output.prof

Then opt runs with arguments: opt input.bc -o output.bc -load mypass.so
-block-freq -pgo-instr-use -pgo-test-profile-file=output.prof
-profile-sample-accurate -mypass

Is this a bug or am can someone provide an example on how to use
BlockFrequencyInfo correctly?


Example code:


for (auto& F : M) {
     if (F.isDeclaration()) {
         continue;
     }
     auto& bfiPass = getAnalysis<BlockFrequencyInfoWrapperPass>(F);
     llvm::BlockFrequencyInfo* BFI = &bfiPass.getBFI();

     //Works - shows all info I want
     BFI->print(llvm::dbgs());

I don't know offhand why your code below is segfaulting, but note that
BFI->print() will dereference the F pointer just as *getFunction below will.
I would look in the debugger to see what the value of F is during print() since
it seems to be ok there, and then see if you can figure out where it is changing
if it is returning a bad value below.
Teresa

Thanks for the feedback! I've tried debugging but this didn't get me
far. I found a different solution. First, you're right that my code is
correct. The problem existed elsewhere.
I've upgraded to the latest LLVM 8 commit and - after recompiling my
projects - the above code executed without segfaults. Going back to my LLVM 7.01
source build, and again I could no longer observe any segfaults. I know that one
of my projects was compiled with clang 7, while LLVM build from source was done
with the system's default gcc/g++ 8. Not sure if combining differently
compiled libraries is considered UB in C++, but after using the same compiler
for both parts errors stopped occurring.


     for (const llvm::BasicBlock& B : F) {
         //Fails with SIGSEGV -> *getFunction() returns 0xa
         dbgs() << BFI->getBlockProfileCount(&B).getValueOr(0)
<< "\n";
     }
}


lib/Analysis/BlockFrequencyInfo.cpp#L211

return BFI->getBlockProfileCount(*getFunction(), BB);

_______________________________________________
LLVM Developers mailing list
llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev


--
Teresa Johnson |         Software Engineer |     tejohnson at
google.com<mailto:tejohnson at google.com> |

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190117/5f1b2a88/attachment-0001.html>