[PATCH] Fix one-sized FP arrays returning on MIPS

[Eric Botcazou]

> In other words, I think the return_in_msb hook is right to return
> true for the test case you quote.

I'm back to this and I'm not sure how to fix the ICE.  The problem is that we 
have at the tree level:

  ret (SFmode) = func() (SFmode)

and we try to expand this in expand_call to:

  ret (SFmode) = func() (DImode)

and we die when generating the move from DImode to SFmode after shifting.


The only solution I see is something along the lines of copy_blkmode_from_reg 
but this seems to be overkill for our present corner-case.  Do you have any 
suggestion as to how we could (elegantly) solve the problem?

-- 
Eric Botcazou