Warning with newer GCC versions in matrix_free/fe_evaluation.h

[bangerth]

[226/939] Building CXX object examples/CMakeFiles/step-76.release.dir/step-76/step-76.cc.o
In file included from /home/bangerth/p/deal.II/1/dealii/examples/step-76/step-76.cc:46:
In member function ‘void dealii::FEEvaluationBase<dim, n_components_, Number, <anonymous>, <template-parameter-1-5> >::read_write_operation(const VectorOperation&, const std::array<VectorType*, n_components_>&, const std::array<const std::vector<dealii::ArrayView<const typename VectorType::value_type> >*, n_components_>&, const std::bitset<VectorizedArrayType::size()>&, bool) const [with VectorType = dealii::LinearAlgebra::distributed::Vector<double>; VectorOperation = dealii::internal::VectorDistributorLocalToGlobal<double, dealii::VectorizedArray<double, 2> >; int dim = 2; int n_components_ = 4; Number = double; bool is_face = false; VectorizedArrayType = dealii::VectorizedArray<double, 2>]’,
    inlined from ‘void dealii::FEEvaluationBase<dim, n_components_, Number, <anonymous>, <template-parameter-1-5> >::distribute_local_to_global(VectorType&, unsigned int, const std::bitset<VectorizedArrayType::size()>&) const [with VectorType = dealii::LinearAlgebra::distributed::Vector<double>; int dim = 2; int n_components_ = 4; Number = double; bool is_face = false; VectorizedArrayType = dealii::VectorizedArray<double, 2>]’ at /home/bangerth/p/deal.II/1/dealii/include/deal.II/matrix_free/fe_evaluation.h:4598:23,
    inlined from ‘Euler_DG::EulerOperator<2, 5, 7>::perform_stage(unsigned int, Euler_DG::Number, Euler_DG::Number, Euler_DG::Number, const dealii::LinearAlgebra::distributed::Vector<double>&, dealii::LinearAlgebra::distributed::Vector<double>&, dealii::LinearAlgebra::distributed::Vector<double>&) const::<lambda(const auto:56&, auto:57&, const auto:58&, auto:59)> [with auto:56 = dealii::MatrixFree<2, double, dealii::VectorizedArray<double, 2> >; auto:57 = dealii::LinearAlgebra::distributed::Vector<double>; auto:58 = dealii::LinearAlgebra::distributed::Vector<double>; auto:59 = std::pair<unsigned int, unsigned int>]’ at /home/bangerth/p/deal.II/1/dealii/examples/step-76/step-76.cc:939:47:
/home/bangerth/p/deal.II/1/dealii/include/deal.II/matrix_free/fe_evaluation.h:3744:24: warning: ‘void* __builtin_memset(void*, int, long unsigned int)’ offset [24, 2040] is out of the bounds [0, 16] of object ‘dof_indices’ with type ‘const unsigned int* [2]’ [-Warray-bounds]
 3744 |         dof_indices[v] = nullptr;
      |         ~~~~~~~~~~~~~~~^~~~~~~~~
/home/bangerth/p/deal.II/1/dealii/include/deal.II/matrix_free/fe_evaluation.h: In lambda function:
/home/bangerth/p/deal.II/1/dealii/include/deal.II/matrix_free/fe_evaluation.h:3669:23: note: ‘dof_indices’ declared here
 3669 |   const unsigned int *dof_indices[n_lanes];
      |                       ^~~~~~~~~~~

[bangerth]

Also:

/home/bangerth/p/deal.II/1/dealii/include/deal.II/matrix_free/fe_evaluation.h:3710:24: warning: ‘void* __builtin_memset(void*, int, long unsigned int)’ offset [24, 2040] is out of the bounds [0, 16] of object ‘dof_indices’ with type ‘const unsigned int* [2]’ [-Warray-bounds]
 3710 |         dof_indices[v] = nullptr;
      |         ~~~~~~~~~~~~~~~^~~~~~~~~

[kronbichler]

This compiler seems to be pedantic, doesn't it? It tries to call memset with bounds 24, 2040, but the array length is only of 2 size_t (n_lanes=2) and the index range in the line above is also within n_vectorization_actual and n_lanes. The only way this would happen is that the compiler believes the integer could need to wrap around max_unsigned_int. It might help if we fix this

dealii/include/deal.II/matrix_free/fe_evaluation.h

Lines 3417 to 3419 in 090def8

	unsigned int n_vectorization_actual =
	this->dof_info
	->n_vectorization_lanes_filled[this->dof_access_index][this->cell];

to std::min(... what we do now ..., n_lanes);?

[kronbichler]

Alternatively, we can declare both variables as std::size_t, then the compiler should again know that we cannot wrap around the max index as this is a memory index.