CSC to CSR Transformation in GPU

Specific Domains GPU
1

Is there any way to transfer A in CSC to CSR in CUDA using Adjoint?

n_rows = 1000000
n_cols = 1000000
nnz = 10000000
row_indices = rand(1:n_rows, nnz)
col_indices = rand(1:n_cols, nnz)
values = rand(nnz)

A = sparse(row_indices, col_indices, values, n_rows, n_cols)
At_cpu = sparse(adjoint(A))
A_rowptr_cpu = At_cpu.colptr
A_colval_cpu = At_cpu.rowval
A_nzval_cpu = At_cpu.nzval

2

Why not create the array in CSR to begin with by transposing the constructor(s)? Call At = sparse(col_indices, row_indices, values, n_rows, n_cols), and then you have rowptr = At.colptr, colval = At.rowval, nzval = At.nzval.

3

Because I am getting the A in CSC format, this is not the whole code.