Fix device_functions macro

[KSepetanc]

I made if-else to capture new temp variable from definition in Julia 1.13.
If I did not error, this should fix the macro.

After this, swap of device_functions for device_function should be reverted.

[eschnett]

Can you add to your PR some code that uses the macro? This will show that the macro works. For example, you could undo the change away from @device_functions.

[KSepetanc]

No errors on my end. Tried it with undone change of @device_functions (deved local branch that matches this PR).
Works both on 1.12 and 1.13. Try it yourself before merge.

To test this and to avoid unrelated issue, I got system CUDA updated to 13.1.

   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.13.0-beta2 (2026-02-04)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org release
|__/                   |

(@v1.13) pkg> st
Status `C:\Users\karlo\.julia\environments\v1.13\Project.toml`
  [052768ef] CUDA v5.9.6 `E:\CUDA.jl`

julia> using CUDA

julia>

   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.12.5 (2026-02-09)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org release
|__/                   |

(@v1.12) pkg> st
Status `C:\Users\karlo\.julia\environments\v1.12\Project.toml`
  [052768ef] CUDA v5.9.6 `E:\CUDA.jl\`

julia> using CUDA

julia>

[eschnett]

I am currently (and have been for two weeks) fighting with the CUDA.jl CI infrastructure to make sure that the package passes with my changes. This is turning out to be quite a bit more difficult and frustrating than I expected. (One reason is that CUDA.jl currently only tests Julia 1.11 and CUDA 12.9, and trying anything more modern uncovers problems that are unrelated to supporting Julia 1.13.)

Unfortunately, the CUDA.jl CI infrastructure does not run on your PR because buildkite isn't set up for my clone of the CUDA.jl repo, so I would be flying blind when merging your changes.

In any way, I will hold off merging your changes until my current PR passes the CUDA.jl CI pipeline to avoid working on too many fronts simultaneously.

[KSepetanc]

CUDA.jl CI is passing now on main PR. Could you try this fix? The macro is needed so other PR can be merged (for AOT compilation).

[eschnett]

Can we merge the two PRs one after the other?

Alternative, you could open a PR against the main CUDA.jl repository which includes my changes.

[KSepetanc]

Check it out JuliaGPU#3031.