ExecutionSpaceRulesΒΆ

CUDA execution space rules

Nested Rules

CUDA-3.1

Use functions only in the execution spaces they target

CUDA-3.2

Use function pointers only on the host or device where their address was taken

CUDA-3.3

Only use objects in the execution space they are associated with

CUDA-3.4

Only use one __shared__ array-of-unknown-bound declaration in a kernel function

CUDA-3.5

Do not create aliases between restricted pointer arguments

Options