public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug middle-end/65315] New: incorrect alignment of local variable with aligned attribute
@ 2015-03-04 18:38 sje at gcc dot gnu.org
  2015-03-04 18:58 ` [Bug middle-end/65315] " sje at gcc dot gnu.org
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: sje at gcc dot gnu.org @ 2015-03-04 18:38 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65315

            Bug ID: 65315
           Summary: incorrect alignment of local variable with aligned
                    attribute
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: sje at gcc dot gnu.org

Created attachment 34953
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34953&action=edit
Test case

When GCC has multiple local variables with different aligned attributes on
them, it creates an aligned block of space aligned to the least aligned
variable instead of the most aligned variable.

In the attached program, when run on MIPS, GCC is creating a 32 byte aligned
block of memory for foo1, foo3, and foo4 and a 128 byte aligned memory block
for foo2.  foo3 and foo4 should have 128 byte aligned memory.

I thought the problem was in stack_var_cmp where it checks the alignment of
two variables and returns '(int) largeb - (int) largea' to sort the variables
based on their alignment but when I swapped the two arguments I got an ICE
in expand_stack_vars [gcc_assert (large_base != NULL);].

It also tried changing:

large_align = stack_vars[stack_vars_sorted[0]].alignb * BITS_PER_UNIT;

to

large_align = stack_vars[stack_vars_sorted[n-1]].alignb * BITS_PER_UNIT;

but that caused the same ICE.

It may be that we need a loop to check the alignment of each variable.


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [Bug middle-end/65315] incorrect alignment of local variable with aligned attribute
  2015-03-04 18:38 [Bug middle-end/65315] New: incorrect alignment of local variable with aligned attribute sje at gcc dot gnu.org
@ 2015-03-04 18:58 ` sje at gcc dot gnu.org
  2015-03-05 16:34 ` sje at gcc dot gnu.org
  2015-10-21 23:43 ` sje at gcc dot gnu.org
  2 siblings, 0 replies; 4+ messages in thread
From: sje at gcc dot gnu.org @ 2015-03-04 18:58 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65315

--- Comment #1 from Steve Ellcey <sje at gcc dot gnu.org> ---
It might be easier to see this bug if you apply this patch:

diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c
index 7dfe1f6..7beb00e 100644
--- a/gcc/cfgexpand.c
+++ b/gcc/cfgexpand.c
@@ -973,6 +973,8 @@ expand_stack_vars (bool (*pred) (size_t), struct sta
ck_vars_data *data)
       i = stack_vars_sorted[si];
       alignb = stack_vars[i].alignb;

+      gcc_assert (alignb*BITS_PER_UNIT <= large_align);
+
       /* Stop when we get to the first decl with "small" alignment.  */
       if (alignb * BITS_PER_UNIT <= MAX_SUPPORTED_STACK_ALIGNMENT)
         break;


This way you get an ICE instead of just incorrectly aligned vectors.


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [Bug middle-end/65315] incorrect alignment of local variable with aligned attribute
  2015-03-04 18:38 [Bug middle-end/65315] New: incorrect alignment of local variable with aligned attribute sje at gcc dot gnu.org
  2015-03-04 18:58 ` [Bug middle-end/65315] " sje at gcc dot gnu.org
@ 2015-03-05 16:34 ` sje at gcc dot gnu.org
  2015-10-21 23:43 ` sje at gcc dot gnu.org
  2 siblings, 0 replies; 4+ messages in thread
From: sje at gcc dot gnu.org @ 2015-03-05 16:34 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65315

--- Comment #3 from Steve Ellcey <sje at gcc dot gnu.org> ---
Author: sje
Date: Thu Mar  5 16:34:03 2015
New Revision: 221219

URL: https://gcc.gnu.org/viewcvs?rev=221219&root=gcc&view=rev
Log:
2015-03-05  Steve Ellcey  <sellcey@imgtec.com>

    PR middle-end/65315
    * cfgexpand.c (expand_stack_vars): Update large_align to maximum
    needed alignment.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/cfgexpand.c


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [Bug middle-end/65315] incorrect alignment of local variable with aligned attribute
  2015-03-04 18:38 [Bug middle-end/65315] New: incorrect alignment of local variable with aligned attribute sje at gcc dot gnu.org
  2015-03-04 18:58 ` [Bug middle-end/65315] " sje at gcc dot gnu.org
  2015-03-05 16:34 ` sje at gcc dot gnu.org
@ 2015-10-21 23:43 ` sje at gcc dot gnu.org
  2 siblings, 0 replies; 4+ messages in thread
From: sje at gcc dot gnu.org @ 2015-10-21 23:43 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65315

Steve Ellcey <sje at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |FIXED
   Target Milestone|---                         |5.0

--- Comment #4 from Steve Ellcey <sje at gcc dot gnu.org> ---
A patch for this was checked in before 5.0 branched.


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2015-10-21 23:43 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-03-04 18:38 [Bug middle-end/65315] New: incorrect alignment of local variable with aligned attribute sje at gcc dot gnu.org
2015-03-04 18:58 ` [Bug middle-end/65315] " sje at gcc dot gnu.org
2015-03-05 16:34 ` sje at gcc dot gnu.org
2015-10-21 23:43 ` sje at gcc dot gnu.org

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).