[Bug target/49699] New: Aligned load on unaligned address

[Bug target/49699] New: Aligned load on unaligned address

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49699

           Summary: Aligned load on unaligned address
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com
                CC: ubizjak@gmail.com


[hjl@gnu-6 tmp]$ cat a.c
#include <emmintrin.h>

struct foo
{
  __m128 x;
} __attribute__ ((aligned(8)));

extern struct foo x;

__m128
foo ()
{
 return x.x;
}
[hjl@gnu-6 tmp]$ gcc -msse2 -O a.c -S    
[hjl@gnu-6 tmp]$ cat a.s
    .file    "a.c"
    .text
    .globl    foo
    .type    foo, @function
foo:
.LFB516:
    .cfi_startproc
    movaps    x(%rip), %xmm0
    ret
    .cfi_endproc
.LFE516:
    .size    foo, .-foo
    .ident    "GCC: (GNU) 4.6.0 20110530 (Red Hat 4.6.0-9)"
    .section    .note.GNU-stack,"",@progbits
[hjl@gnu-6 tmp]$

[Bug middle-end/49699] Aligned load on unaligned address

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49699

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |middle-end

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2011-07-11 03:13:16 UTC ---
It looks like a middle-end bug.

[Bug middle-end/49699] Aligned load on unaligned address

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49699

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.07.11 09:24:48
     Ever Confirmed|0                           |1

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-11 09:24:48 UTC ---
Confirmed.

The issue is that alignment info in FIELD_DECLs is "broken".  I think that
in this case get_inner_reference simply chooses to tell the caller that
V4SF mode is to be used for the access and that's all it gets.

Very old issue.  The aligned attribute cannot be reliably used to decrease
alignment.

[Bug middle-end/49699] Aligned load on unaligned address

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49699

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-07-11 09:47:27 UTC ---
I think you'd need to make it packed as well, but then it depends on gcc
version which alignment the aggregate actually has.

[Bug middle-end/49699] Aligned load on unaligned address

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49699

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> 2011-07-12 14:05:08 UTC ---
(In reply to comment #2)
> 
> Very old issue.  The aligned attribute cannot be reliably used to decrease
> alignment.

Can we issue a warning for things like this?

[Bug middle-end/49699] Aligned load on unaligned address

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49699

--- Comment #5 from H.J. Lu <hjl.tools at gmail dot com> 2011-07-12 14:06:08 UTC ---
(In reply to comment #3)
> I think you'd need to make it packed as well, but then it depends on gcc
> version which alignment the aggregate actually has.

This issue came from

http://sourceware.org/bugzilla/show_bug.cgi?id=12872

[Bug middle-end/49699] Aligned load on unaligned address

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=49699

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2011-07-11 09:24:48         |2021-9-1
           Keywords|                            |wrong-code

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
What is interesting is clang gets it wrong too.