[llvm-commits] [PATCH] To enable struct byval for ARM

[Tim Northover]

Hi Manman,

> This patch is to enable struct byval for ARM.
> It inserts a pseudo op for each struct byval argument and the pseudo ops will be expanded to a loop or unrolled scalar copies.

Is there any particular reason you're not using SelectionDAG's
getMemcpy function? It looks like EmitTargetCodeForMemCpy is a little
naive on ARM at the moment (it seems to generate no loops), but that
seems like a more natural place to put the bulk of your code when it
really is the best way to expand the copy.

I may be missing something fundamental, of course.

Tim.