Bug: `ks_solver=cusolver` not working properly reporting CUDA error

[AsTonyshment]

Describe the bug

When I set ks_solver=cusolver, ABACUS quits with CUDA error (for example, a simple SCF task). After my testing, this seems to occur after PR #5225. Maybe @Cstandardlib can have a look?

 START CHARGE      : atomic
 DONE(1.21413    SEC) : INIT SCF
 * * * * * *
 << Start SCF iteration.
CUDA error at /home/itztony/Documents/Research/Coding/abacus-develop/source/module_hsolver/kernels/cuda/diag_cusolver.cu:132 code=1(cudaErrorInvalidValue) "cudaMemcpy(d_A2, A, sizeof(cuDoubleComplex) * lda * m, cudaMemcpyHostToDevice)" 
--------------------------------------------------------------------------
Primary job  terminated normally, but 1 process returned
a non-zero exit code. Per user-direction, the job has been aborted.
--------------------------------------------------------------------------
--------------------------------------------------------------------------
mpirun detected that one or more processes exited with non-zero status, thus causing
the job to be terminated. The first process to do so was:

  Process name: [[48991,1],0]
  Exit code:    1
--------------------------------------------------------------------------

Expected behavior

No response

To Reproduce

No response

Environment

No response

Additional Context

No response

Task list for Issue attackers (only for developers)

• Verify the issue is not a duplicate.
• Describe the bug.
• Steps to reproduce.
• Expected behavior.
• Error message.
• Environment details.
• Additional context.
• Assign a priority level (low, medium, high, urgent).
• Assign the issue to a team member.
• Label the issue with relevant tags.
• Identify possible related issues.
• Create a unit test or automated test to reproduce the bug (if applicable).
• Fix the bug.
• Test the fix.
• Update documentation (if necessary).
• Close the issue and inform the reporter (if applicable).

[Cstandardlib]

I will reproduce it locally.

[Cstandardlib]

When ks_solver is set to cusolver,
The error arises from:
line 154 of hsolver_lcao.cpp where an object of DiagoCusolver is created and diag method is called

DiagoCusolver<T> cs(this->ParaV);
cs.diag(hm, psi, eigenvalue);

line 160 to 168 of diago_cusolver.cpp is excuted

if (num_procs > 1)
    {
        MPI_Barrier(MPI_COMM_WORLD);
        // global psi for distribute
        int psi_len = myid == root_proc ? h_mat_g.row * h_mat_g.col : 1;
        std::vector<T> psi_g(psi_len);
        if (myid == root_proc)
        {
            this->dc.Dngvd(h_mat_g.col, h_mat_g.row, h_mat_g.p.get(), s_mat_g.p.get(), eigen.data(), psi_g.data());
        }
...

Then dc.Dngvd in diago_cusolver.cpp is invoked

// diago_cusolver.cpp
this->dc.Dngvd(h_mat.row, h_mat.col, h_mat.p, s_mat.p, eigen.data(), eigenvectors.data());

// diago_cusolver.cuh
void Dngvd(int N, int M, std::complex<double> *A, std::complex<double> *B, double *W, std::complex<double> *V)
    {
        return Dngvd_complex(N, M, A, B, W, V);
    };

// diago_cusolver.cu
void Diag_Cusolver_gvd::Dngvd_complex(int N, int M, std::complex<double> *A, std::complex<double> *B, double *W, std::complex<double> *V){
    
    // copy A, B to the GPU
        assert(N == M);
        if (M != m) {
            this->finalize();
            this->init_complex(M);
        }
        checkCudaErrors( cudaMemcpy(d_A2, A, sizeof(cuDoubleComplex) * lda * m, cudaMemcpyHostToDevice) );
        checkCudaErrors( cudaMemcpy(d_B2, B, sizeof(cuDoubleComplex) * lda * m, cudaMemcpyHostToDevice) );
    ...
}

line 132 of diago_cusolver.cu is where the error is thrown.
checkCudaErrors( cudaMemcpy(d_A2, A, sizeof(cuDoubleComplex) * lda * m, cudaMemcpyHostToDevice) );

[Cstandardlib]

When I removed the 'DiagH' base class that 'DiagoCusolver' inherits from, the scf example using cusolver on the GPU works correctly. How peculiar!