Null pointer in CUDA backend when using polish with no active set #463
Description
When there is no active set present in a problem and the CUDA backend tries to perform the polishing step, there is a NULL pointer error thrown by CUDA:
Filters: test_no_active_set
-----------------------------------------------------------------
OSQP v0.0.0 - Operator Splitting QP Solver
(c) Bartolomeo Stellato, Goran Banjac
University of Oxford - Stanford University 2021
-----------------------------------------------------------------
problem: variables n = 2, constraints m = 3
nnz(P) + nnz(A) = 6
settings: algebra = CUDA,
linear system solver = CUDA Preconditioned Conjugate Gradient,
eps_abs = 1.0e-05, eps_rel = 1.0e-05,
eps_prim_inf = 1.0e-04, eps_dual_inf = 1.0e-04,
rho = 1.00e-01 (adaptive),
sigma = 1.00e-06, alpha = 1.60, max_iter = 4000
check_termination: on (interval 5),
time_limit: 1.00e+10 sec,
scaling: on, scaled_termination: off
warm starting: on, polishing: off,
iter objective prim res dual res rho time
1 0.0000e+00 0.00e+00 0.00e+00 1.00e-01 8.10e-01s
5 0.0000e+00 0.00e+00 0.00e+00 1.00e-01 8.11e-01s
status: solved
number of iterations: 5
optimal objective: 0.0000
run time: 8.11e-01s
optimal rho estimate: 1.00e-06
-----------------------------------------------------------------
OSQP v0.0.0 - Operator Splitting QP Solver
(c) Bartolomeo Stellato, Goran Banjac
University of Oxford - Stanford University 2021
-----------------------------------------------------------------
problem: variables n = 2, constraints m = 3
nnz(P) + nnz(A) = 6
settings: algebra = CUDA,
linear system solver = CUDA Preconditioned Conjugate Gradient,
eps_abs = 1.0e-05, eps_rel = 1.0e-05,
eps_prim_inf = 1.0e-04, eps_dual_inf = 1.0e-04,
rho = 1.00e-01 (adaptive),
sigma = 1.00e-06, alpha = 1.60, max_iter = 4000
check_termination: on (interval 5),
time_limit: 1.00e+10 sec,
scaling: on, scaled_termination: off
warm starting: on, polishing: on,
iter objective prim res dual res rho time
1 0.0000e+00 0.00e+00 0.00e+00 1.00e-01 5.73e-02s
5 0.0000e+00 0.00e+00 0.00e+00 1.00e-01 5.80e-02s
** On entry to cusparseSpMV() parameter number 7 (vecY) had an illegal value: NULL pointer
CUDA error at /home/b63738im/dev/optimization/osqp/c_code/algebra/cuda/src/cuda_lin_alg.cu:946 code=3(CUSPARSE_STATUS_INVALID_VALUE) "cusparseSpMV( CUDA_handle->cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, A->SpMatDescr, vecx, &beta, vecy, CUDA_FLOAT, CUSPARSE_SPMV_ALG_DEFAULT, A->SpMatBuffer)"