Working with a global context

[digikar99]

Hi! New to both cuda and cl-cuda. I was looking for a way to avoid with-cuda; what I'm puzzled with is that the following works at the REPL, but doesn't work with C-x C-e in SLIME (at least not until it is run at least once at the REPL):

(init-cuda)
(setq *cuda-device* (get-cuda-device 0)
      *cuda-context* (create-cuda-context *cuda-device*)
      *nvcc-options*
      (if (cl-cuda.api.context::arch-exists-p *nvcc-options*)
          *nvcc-options*
          (cl-cuda.api.context::append-arch *nvcc-options* 0)))
(alloc-memory-block 'float 1)

The last statement yields the following error:

CU-MEM-ALLOC failed with driver API error No. 201.
CUDA_ERROR_INVALID_CONTEXT
   [Condition of type SIMPLE-ERROR]

Restarts:
 0: [RETRY] Retry SLIME REPL evaluation request.
 1: [*ABORT] Return to SLIME's top level.
 2: [ABORT] abort thread (#<THREAD "repl-thread" RUNNING {100F7C1DD3}>)

Backtrace:
  0: (CL-CUDA.DRIVER-API::CHECK-CUDA-ERROR CL-CUDA.DRIVER-API:CU-MEM-ALLOC 201)
  1: (ALLOC-DEVICE-MEMORY FLOAT 1)
  2: (ALLOC-MEMORY-BLOCK FLOAT 1)

Any reason why this happens?