This is related to the changes to the groupstaticsize intrinsic in
D61494 which would otherwise make the related tests in these files
fail or much less useful.
Note that for some reason, SOPK generation is less effective in the
amdhsa OS, which is why I chose PAL. I haven't investigated this