Undefined behavior warnings

[rhornung67]

The GNU compiler emits a warning 'iteration 3 invokes undefined behavior' at the following line when compiling RAJA tests:

https://github.com/LLNL/RAJA/blob/develop/include/RAJA/pattern/tensor/internal/MatrixRegisterImpl.hpp#L764

Is this fixable?

[rchen20]

Is this fixable?

I'm not sure, but I'm looking in to this. Another data point with gcc/10.2.1 on TOSS3:

In file included from /usr/WS1/chen59/allraja/rajavecarg/raja_git_vecarg/include/RAJA/pattern/tensor.hpp:31,
                 from /usr/WS1/chen59/allraja/rajavecarg/raja_git_vecarg/include/RAJA/RAJA.hpp:52,
                 from /usr/WS1/chen59/allraja/rajavecarg/raja_git_vecarg/test/include/RAJA_test-base.hpp:15,
                 from /usr/WS1/chen59/allraja/rajavecarg/raja_git_vecarg/build_lc_toss3-gcc-10.2.1/test/functional/tensor/matrix/test-tensor-matrix-double-ColMajor-ET_MatrixVector.cpp:11:
/usr/WS1/chen59/allraja/rajavecarg/raja_git_vecarg/include/RAJA/pattern/tensor/internal/MatrixRegisterImpl.hpp: In static member function 'static RAJA::internal::expt::ET::MultiplyOperator<LEFT_OPERAND_TYPE, RIGHT_OPERAND_TYPE, typename std::enable_if<((LEFT_OPERAND_TYPE::s_num_dims == 2) && (RIGHT_OPERAND_TYPE::s_num_dims == 1))>::type>::result_type RAJA::internal::expt::ET::MultiplyOperator<LEFT_OPERAND_TYPE, RIGHT_OPERAND_TYPE, typename std::enable_if<((LEFT_OPERAND_TYPE::s_num_dims == 2) && (RIGHT_OPERAND_TYPE::s_num_dims == 1))>::type>::multiply(const TILE_TYPE&, const LEFT_OPERAND_TYPE&, const RIGHT_OPERAND_TYPE&) [with TILE_TYPE = RAJA::internal::expt::TensorTile<int, RAJA::internal::expt::TENSOR_FULL, 1>; LEFT_OPERAND_TYPE = RAJA::internal::expt::ET::TensorLoadStore<RAJA::expt::TensorRegister<RAJA::expt::avx2_register, double, RAJA::expt::TensorLayout<1, 0>, camp::int_seq<long int, 2, 4> >, RAJA::internal::expt::StaticTensorRef<double*, long int, RAJA::internal::expt::TENSOR_MULTIPLE, camp::int_seq<long int, 8, 1>, camp::int_seq<long int, 0, 0>, camp::int_seq<long int, 8, 8>, 1> >; RIGHT_OPERAND_TYPE = RAJA::internal::expt::ET::TensorLoadStore<RAJA::expt::TensorRegister<RAJA::expt::avx2_register, double, RAJA::expt::TensorLayout<0>, camp::int_seq<long int, 4> >, RAJA::internal::expt::StaticTensorRef<double*, long int, RAJA::internal::expt::TENSOR_MULTIPLE, camp::int_seq<long int, 1>, camp::int_seq<long int, 0>, camp::int_seq<long int, 8>, 0> >]':
/usr/WS1/chen59/allraja/rajavecarg/raja_git_vecarg/include/RAJA/pattern/tensor/internal/MatrixRegisterImpl.hpp:764:51: warning: iteration 1 invokes undefined behavior [-Waggressive-loop-optimizations]
  764 |               camp::idx_t reg_num_cols = num_cols - i*s_major_dim_per_register;
      |                                          ~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~
/usr/WS1/chen59/allraja/rajavecarg/raja_git_vecarg/include/RAJA/pattern/tensor/internal/MatrixRegisterImpl.hpp:762:37: note: within this loop 
  762 |             for(camp::idx_t i = 0;i < s_num_registers;++ i){
      |                                   ~~^~~~~~~~~~~~~~~~~