Update reqd_sub_group_size argument name.

[bader]

dim name is tipically used to denote the dimension (see reqd_work_group_size attribute description). In this case we set the size characterisitic, rather than the dimension.

[Pennycook]

I think it was originally written as "dim" to match reqd_work_group_size.

Should we change to reqd_work_group_size(size0, size1, size2) as well? Note that what the attribute really sets in that case is the size of the work-group in each dimension, so I think the same logic could apply.

[bader]

I think it was originally written as "dim" to match reqd_work_group_size.

Should we change to reqd_work_group_size(size0, size1, size2) as well? Note that what the attribute really sets in that case is the size of the work-group in each dimension, so I think the same logic could apply.

According to my understanding, the reqd_sub_group_size doesn't specify the size in any particular dimension.
E.g. sub-group size 8 can be implemented as group of work-items with dimension sizes (1, 1, 8) or (1, 2, 4) or (2, 2, 2).
Therefore, using dim for reqd_sub_group_size doesn't make sense.

I don't have a strong opinion about reqd_work_group_size, but reqd_work_group_size(size0, size1, size2) looks clearer to me than reqd_work_group_size(dim0, dim1, dim2).

[gmlueck]

I agree that it makes sense to change reqd_sub_group_size because the parameter is not related to the dimensions of the invocation range.

I could go either way with the naming of reqd_work_group_size. The parameters here describe the size of the work-group in each dimension, so the words "size" and "dim" both make sense.

[gmlueck]

This needs one more reviewer. Can someone volunteer?

[keryell]

Yes, that looks more uniform. Thanks.