On 11/07/2011 01:44 AM, Richard Guenther wrote: > It just catched my eye... moving it to expand_call_stmt would be nice > indeed, but I was suggesting to add that note where we produce the > CALL rtx, not sure if that's reasonably straight-forward (I suppose there > was a reason to go with the hack above ...).
Because targets emit all sorts of stuff in gen_call that isn't a CALL_INSN. We have to search anyway. And the guts of expand_call are complex enough already; no need to make it worse. r~
