[Bug target/60337] New: s390: Function arguments are not aligned

[Bug target/60337] New: s390: Function arguments are not aligned

[vogt at linux dot vnet.ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60337

            Bug ID: 60337
           Summary: s390: Function arguments are not aligned
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vogt at linux dot vnet.ibm.com

On s390, if a type with __attribute__((aligned(...))) is passed as a function
parameter, the parameters are not aligned properly.  Also, a parameter that is
passed in a register and then copied to the stack is not aligned.  Output of
the attached test program on s390x:

$ ./foo      
0x3ffff8f0000      <--- Address on main's stack is aligned
0x3ffff8e2980 5    <--- foo's parameter is not aligned
0x3ffff8e2980 5    <--- bar's parameter is not aligned
0x3ffff8e2980 123  <--- baz's parameter is not aligned

[Bug target/60337] s390: Function arguments are not aligned

[vogt at linux dot vnet.ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60337

--- Comment #1 from Dominik Vogt <vogt at linux dot vnet.ibm.com> ---
Created attachment 32209
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32209&action=edit
test program

[Bug target/60337] s390: Function arguments are not aligned

[vogt at linux dot vnet.ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60337

--- Comment #2 from Dominik Vogt <vogt at linux dot vnet.ibm.com> ---
(libffi does not honour the alignment either.)

[Bug target/60337] s390: Function arguments are not aligned

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60337

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ABI, diagnostic
             Target|                            |s390x-*-*

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
It's the same on x86 I suppose - the aligned() attribute would have to change
the ABI for non-register passing as aligning the argument stack cannot fulfil
alignment requirements for more than one argument.

IMHO this should be simply diagnosed.

I don't see in your testcase where "Also, a parameter that is passed in a
register and then copied to the stack is not aligned." happens.

[Bug target/60337] s390: Function arguments are not aligned

[vogt at linux dot vnet.ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60337

--- Comment #5 from Dominik Vogt <vogt at linux dot vnet.ibm.com> ---
With -O2 and -O3 all three cases generate the proper alignment
With -O1 only foo() has the proper alignment
With -O0 none of the cases has the proper alignment

==> Code compiled with -O0 and -O3 accidentally uses a different ABI.

Power shows the same erratic behaviour, on amd64 alignment works with all
-O{0,1,2,3}.