llvm dev - May 2014 - [LLVMdev] It is possible to somehow turn off coercion of struct parameters into ints?

In particular, I would for example like to prevent that two fields of
type i32 are packed into an i64 parameter. And so on...

Thanks!
-- Zvonimir

There isn't a flag to do this, but depending on what you need, you can hack
clang to get it.

Someone else wanted Clang to pass structs directly as LLVM aggregates, but
that would definitely be an ABI break:
http://lists.cs.uiuc.edu/pipermail/cfe-dev/2014-April/036615.html

This might be a reasonable lowering, if it works for you:

struct A { int a; int b; };
void foo(A x) { }

%struct.A = type { i32, i32 }
define void @foo(%struct.A* byval %x) {
  ret void
}

Something like this would be easy to implement and might even be ABI
compatible with the integral coercion lowering, but I make no guarantees.
 You'd change TargetInfo.cpp to return ABIArgInfo::getIndirect(0,
/*ByVal=*/true) for every record that is currently expanded or passed
directly.


On Fri, May 16, 2014 at 8:55 AM, Zvonimir Rakamaric <zvonimir at
cs.utah.edu>wrote:
> In particular, I would for example like to prevent that two fields of
> type i32 are packed into an i64 parameter. And so on...
>
> Thanks!
> -- Zvonimir
> _______________________________________________
> 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/20140516/2716364b/attachment.html>

Thanks for your help Reid.

After poking around a little (a lot actually) I figured out that
selecting DefaultTargetCodeGenInfo would work for us. And that could
be accomplished by selecting UnknownArch as the target architecture.

Now, I figured our a one line hack that accomplishes that in clang.
What I could not figure out was how to set UnknownArch from the
command line using --target or something like that.

Invoking clang like this:
clang --target=x86_64-unknown-linux-gnu simple.c
is fine.

But for this:
clang --target=unknown-unknown-linux-gnu simple.c

I get an error:
error: unknown target triple 'unknown-unknown-linux-gnu', please use
-triple or -arch

It would really be awesome if UnknownArch could somehow be selected
from the command line. Is that possible? Am I missing something?

Thanks!
-- Zvonimir


--
http://zvonimir.info
http://soarlab.org/


On Fri, May 16, 2014 at 11:22 AM, Reid Kleckner <rnk at google.com>
wrote:
> There isn't a flag to do this, but depending on what you need, you can
hack
> clang to get it.
>
> Someone else wanted Clang to pass structs directly as LLVM aggregates, but
> that would definitely be an ABI break:
> http://lists.cs.uiuc.edu/pipermail/cfe-dev/2014-April/036615.html
>
> This might be a reasonable lowering, if it works for you:
>
> struct A { int a; int b; };
> void foo(A x) { }
>
> %struct.A = type { i32, i32 }
> define void @foo(%struct.A* byval %x) {
>   ret void
> }
>
> Something like this would be easy to implement and might even be ABI
> compatible with the integral coercion lowering, but I make no guarantees.
> You'd change TargetInfo.cpp to return ABIArgInfo::getIndirect(0,
> /*ByVal=*/true) for every record that is currently expanded or passed
> directly.
>
>
> On Fri, May 16, 2014 at 8:55 AM, Zvonimir Rakamaric <zvonimir at
cs.utah.edu>
> wrote:
>>
>> In particular, I would for example like to prevent that two fields of
>> type i32 are packed into an i64 parameter. And so on...
>>
>> Thanks!
>> -- Zvonimir
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>