llvm dev - Jan 2014 - [LLVMdev] byval attr for base+range parameters

Hello guys,

For single scalar pointer parameter, we have 'byval' to specify the
pointee is not changed. But for base+range parameters like:
declare void @_gfortran_transfer_integer_write(%struct.__st_parameter_dt*,
i8*, i32)
(This is the fortran runtime api to 'printf' one integer) The 2nd &
3rd args are used to specify the memory of that integer. I didn't find
a comfortable way to specify that it's 'byval'.  I need this to make
argpromotion work. Right now I just create an alloca, load value into
it, and specify this alloca(base+range) as parameter.   Is it a
missing feature or there is a solution already?


thanks,
yuanfang

That is not what byval does.  byval implicitly copies the pointee into the
argument slots used for the call or registers on some architectures.


On Thu, Jan 23, 2014 at 7:42 PM, Yuanfang Chen <cyfmxc at gmail.com>
wrote:
> Hello guys,
>
> For single scalar pointer parameter, we have 'byval' to specify the
> pointee is not changed. But for base+range parameters like:
> declare void @_gfortran_transfer_integer_write(%struct.__st_parameter_dt*,
> i8*, i32)
> (This is the fortran runtime api to 'printf' one integer) The 2nd
&
> 3rd args are used to specify the memory of that integer. I didn't find
> a comfortable way to specify that it's 'byval'.  I need this to
make
> argpromotion work. Right now I just create an alloca, load value into
> it, and specify this alloca(base+range) as parameter.   Is it a
> missing feature or there is a solution already?
>
>
> thanks,
> yuanfang
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140123/86fc5081/attachment.html>

On Thu, Jan 23, 2014 at 10:57 PM, Reid Kleckner <rnk at google.com>
wrote:
> That is not what byval does.  byval implicitly copies the pointee into the
> argument slots used for the call or registers on some architectures.
Thanks Reid.  I looked into the argpromotion code and  you are right.
'byval' does not work for this case. I think I'll stick to my
workaround for now. But I don't  quite understand why argpromotion
does not check for readonly use other than (GEPorNonGEP) LoadInst.
For case like

define void @checksum_(i32* noalias %i) {
  . . .  load %i
  call void @_gfortran_transfer_integer_write(%xx, %i, 4)
}

If type of
@_gfortran_transfer_integer_write(%struct.__st_parameter_dt*, i8*,
i32)  is changed to
@_gfortran_transfer_integer_write(%struct.__st_parameter_dt*, i8* byval, i32)

It makes sense to argpro %i of checksum_ , right?