Adding defines as _additional_ preprocessor definitions

[loostrum]

I have a kernel where I tune a bunch of parameters, but need to insert some static values as well (think problem size etc.) as preprocessor definitions. I thought I'd be able to use the defines option for this, but using that means the tunable parameters are no longer inserted. It would be great if there'd be a way to insert both the tunable parameters and some more defines.

Currently I'm using single-value tunable parameters as a workaround, but it clutters the tuning output a bit.

[stijnh]

You should be able to use:

defines = dict(
block_size_x=lambda p: p["block_size_x"],
use_smem=lambda p: p["use_smem"],
...
)

If I remember correctly, this should also work:

defines = dict(
block_size_x="block_size_x",
use_smem="use_smem",
...
)

In which case, you can generate the dict automatically:

defines = {key: key for key in tune_params}
defines["another_variable"] = 1
defines["more_variables"] = 2

[loostrum]

Thanks for the tip, that should solve my issue!

[stijnh]

Great! This should be documented better since currently its a bit hidden away in the API documentation between all the other keyword argument descriptions.

[benvanwerkhoven]

At least the API documentation is visible again! haha