[RFC][PATCH,reload] PR middle-end/42431, fix latent bug in gen_reload_chain_without_interm_reg_p()

[Peter Bergner]

On Thu, 2010-02-25 at 11:16 -0700, Jeff Law wrote:
> #2 really isn't that hard.  We just have a vector of locations where we 
> made a substitution, then we back them out when the routine is 
> complete.  In fact, we don't have to generate gobs of RTL for this 
> case.   I think it looks something like the attached patch.    In fact, 
> we can probably remove the copy_rtx call in this case.

Your patch bootstrapped and regtest tested on powerpc64-linux with no
regressions and I verified it fixes the bug in the PR.  Thanks!

Peter