[Debug] Assertion failure in llvm

[kiranchandramohan]

Compiling the following code with "-g" leads to assertion failure on the release_70 branch of LLVM.

module mod_vars1
  type outpatch_list
    integer, allocatable, dimension(:) :: patch_list
  end type outpatch_list
end module

The assertion failure is likely due to DIFortranSubrange taking a DIDerivedType for lowerBound and upperBound instead of a DIVariable.

!14 = !DIDerivedType(tag: DW_TAG_member, file: !3, scope: !13, name: "patch_list$sd", size: 1024, align: 64, offset: 128, baseType: !28)
!15 = !DIExpression(DW_OP_plus_uconst, 80, DW_OP_deref)
!16 = !DIExpression(DW_OP_dup, DW_OP_plus_uconst, 80, DW_OP_deref, DW_OP_swap, DW_OP_plus_uconst, 88, DW_OP_deref, DW_OP_plus, DW_OP_constu, 1, DW_OP_minus)
!17 = !DIFortranSubrange(lowerBound: !14, lowerBoundExpression: !15, upperBound: !14, upperBoundExpression: !16)

Note:

1. This issue was exposed after merging "Improve LLVM debug metadata"
   a322f86
2. It might have been introduced by "Add debug support for allocatable within modules"
   65863b1

[gklimowicz]

@jiel-nv Can you take a look at this? It may be fixed in a set of changes that are not yet in Flang.

@kiranchandramohan says this LLVM assertion failure happens with other cases as well.

[kiranchandramohan]

@gklimowicz Do you have an update on this issue? Please let me know if you need additional info.

I would like to know,

1. Can you reproduce the issue?
2. Is this issue already fixed and the fix is in your queue?
3. Is there an extension to LLVM metadata to support debug for allocatables in derived types?

[jiel-nv]

I'm not able to reproduce the reported assertion failure.

And the debug support for allocatable in derived types is working in progress.

[kiranchandramohan]

I am able to reproduce the issue with the following build script. The script builds LLVM with assertions and builds flang with FLANG_LLVM_EXTENSIONS.

Note: You will have to exclude/revert the last PR merge to avoid a build failure.
commit ca301c3
Merge: a0a67c5 a32d3ab
Author: Steve Scalpone sscalpone@nvidia.com
Date: Wed Sep 11 13:53:16 2019 -0700
Merge pull request #795 from ThePortlandGroup/nv_stage
Pull 2019-09-11T13-10 Recent NVIDIA Changes

INSTALL_DIR=$PWD
cd $INSTALL_DIR
git clone https://github.com/flang-compiler/llvm.git
cd llvm
git checkout release_80
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=ON -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR -DLLVM_CONFIG=$INSTALL_DIR/bin/llvm-config  ..
make -j24
make install

cd $INSTALL_DIR
git clone https://github.com/flang-compiler/flang-driver.git
cd flang-driver
git checkout release_80
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR -DLLVM_CONFIG=$INSTALL_DIR/bin/llvm-config  ..
make -j24
make install

cd $INSTALL_DIR
git clone https://github.com/llvm-mirror/openmp.git
cd openmp/runtime
git checkout release_80
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR -DLLVM_CONFIG=$INSTALL_DIR/bin/llvm-config -DCMAKE_CXX_COMPILER=$INSTALL_DIR/bin/clang++ -DCMAKE_C_COMPILER=$INSTALL_DIR/bin/clang  ../..
make -j24
make install

cd $INSTALL_DIR
git clone https://github.com/flang-compiler/flang.git

cd $INSTALL_DIR
cd flang/runtime/libpgmath
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR -DLLVM_CONFIG=$INSTALL_DIR/bin/llvm-config -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_Fortran_COMPILER=gfortran ..
make -j24
make install

cd $INSTALL_DIR
cd flang
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -DFLANG_LLVM_EXTENSIONS=ON -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR -DLLVM_CONFIG=$INSTALL_DIR/bin/llvm-config -DCMAKE_CXX_COMPILER=$INSTALL_DIR/bin/clang++ -DCMAKE_C_COMPILER=$INSTALL_DIR/bin/clang  -DCMAKE_Fortran_COMPILER=$INSTALL_DIR/bin/flang -DLLVM_LIT_OUTPUT_DIR=$INSTALL_DIR/llvm/build/bin ..
make -j24
make install

[jiel-nv]

Investigated, once the debug info support for allocatable in derived type gets merged in (currently in queue), this issue will be resolved.

[kiranchandramohan]

Thanks @jiel-nv.

[kiranchandramohan]

I can confirm that the assertion failure has been fixed. Thanks @gklimowicz, @jiel-nv.