[PATCH] Fix missing definition for Linker alignment limit

[Richard Guenther]

On Mon, May 12, 2008 at 2:32 AM, Andy H <hutchinsonandy@aim.com> wrote:
> This is fix for AVR target which does not have explicit definition for
> MAX_OFILE_ALIGNMENT.
>
>  So it defaults to  value of BIGGEST_ALIGNMENT = 8 bits (1 byte).
>
>  So any alignment attempt  will produce warning :
>
>    debugstuff.c:4: warning: alignment of 'Foo' is greater than maximum
> object file alignment.  Using 1
>
>  Note alignment is forced to 1 (byte)
>
>  Simple testcase:
>
>    typedef int aligned_int __attribute__ ((aligned (2)));
>    struct  { char c;aligned_int a,b;} xa = {1,2};
>
>  The attached patch corrects this by defining a more appropriate  linker
> alignment limit.
>
>  Tested  on full testsuite. There were 32 less failures and no regressions
>
>  # of expected passes        43485
>  # of unexpected failures    564
>  # of unexpected successes    3
>  # of expected failures        92
>  # of unresolved testcases    172
>  # of untested testcases        67
>  # of unsupported tests        1466
>
>  xgcc  version 4.4.0 20080511
>
>
> -------------------------------------------------------------------------------------
>
>  2008-05-11  Andy Hutchinson <hutchinsonandy@aim.com>
>
>    * config/avr/avr.h (MAX_OFILE_ALIGNMENT_: Define.
>
>
>  Can you approve for gcc head.

I would consider this also appropriate for the 4.3 branch.

Thanks,
Richard.

>
>
>
>
>
>
>
> Index: gcc/config/avr/avr.h
>  ===================================================================
>  --- gcc/config/avr/avr.h    (revision 135174)
>  +++ gcc/config/avr/avr.h    (working copy)
>  @@ -147,6 +147,9 @@
>   /* No data type wants to be aligned rounder than this.  */
>   #define BIGGEST_ALIGNMENT 8
>
>  +/* Maximum alignment permitted by linker */
>  +#define MAX_OFILE_ALIGNMENT (32768 * 8)
>  +
>   #define TARGET_VTABLE_ENTRY_ALIGN 8
>
>   #define STRICT_ALIGNMENT 0
>
>