llvm dev - Nov 2017 - byval vs. explicitly coded copy

Is there any disadvantage to inserting explicit llvm code in a prolog to copy a
parameter, rather than putting the 'byval' attribute on the formal or
indirect call?

I have an IR that, without some significant work, does not make enough
information
available at an indirect call site to know whether a copy is needed.  (The
original
source language knows, it just gets lost at an indirect call.)


-- 
Rodney Bates
rodney.m.bates at acm.org

On 11/3/2017 2:57 PM, Rodney M. Bates via llvm-dev
wrote:
> Is there any disadvantage to inserting explicit llvm code in a prolog 
> to copy a
> parameter, rather than putting the 'byval' attribute on the formal
or
> indirect call?
>
> I have an IR that, without some significant work, does not make enough 
> information
> available at an indirect call site to know whether a copy is needed.  
> (The original
> source language knows, it just gets lost at an indirect call.)
It's fine; actually, you'll probably get better code in many cases by 
avoiding byval.  (If you're calling code which uses the C calling 
convention, you might need to use "byval" to match the calling
convention.)

-Eli

-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux
Foundation Collaborative Project