[vta] fix PCH testsuite regression with DW_AT_MIPS_linkage_name

[Geoffrey Keating]

Alexandre Oliva <aoliva@redhat.com> writes:

> Turns out compilations with PCH get DECL_ASSEMBLER_NAME set earlier
> than without it, so the code I added to defer the computation of D_A_N
> in dwarf2out behaved differently, adding the DW_AT_M_l_n attribute in
> a different order.  Not just that: it also attached the attribute to
> the wrong die.  Oops.
> 
> Fixed as follows.  This is kind of a kludge; I can't convince myself
> it makes sense to investigate why PCH makes it different.

It's intentional, a patch Mark contributed to improve performance; it
takes quite a lot of time to compute the DECL_ASSEMBLER_NAME so you
want to do it at PCH generation time rather than when using the PCH if
possible.

> 	* dwarf2out.c (move_linkage_attr): New.
> 	(dwarf2out_finish): Add MIPS_linkage_name attribute to the correct
> 	die, at the right place.