[Bug middle-end/20371] New: Some corner cases of MS bitfields don't work

[Bug middle-end/20371] New: Some corner cases of MS bitfields don't work

[amylaar at gcc dot gnu dot org]

See comments in patch

-- 
           Summary: Some corner cases of MS bitfields don't work
           Product: gcc
           Version: 3.4.3
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P2
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: amylaar at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
OtherBugsDependingO 17652
             nThis:


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-03-07 22:00 -------
An analysis and a patch is here:
http://gcc.gnu.org/ml/gcc-patches/2004-12/msg00298.html

This patch has been used in the meantime in a 3.4 based compiler to
compile various software like Linux kernels etc.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-04-12 14:32 -------
Subject: Bug 20371

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	sh-elf-4_1-branch
Changes by:	amylaar@gcc.gnu.org	2005-04-12 14:32:06

Modified files:
	gcc            : ChangeLog stor-layout.c 

Log message:
	PR middle-end/20371:
	* stor-layout.c (place_field): Don't adjust rli->bitpos at the end
	of an MS bitfield run if we already adjusted bitpos/offset for an
	alignment as large or larger than the bitfield type size.
	Also set rli->remaining_in_alignment when we reset rli->prev_field.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=sh-elf-4_1-branch&r1=2.8142.2.4&r2=2.8142.2.5
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/stor-layout.c.diff?cvsroot=gcc&only_with_tag=sh-elf-4_1-branch&r1=1.228&r2=1.228.2.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|17652                       |
              nThis|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-04-12 18:31 -------
The patch was posted on the 3rd December 2004, well before the
25th Febuary 2005.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|                            |17652
              nThis|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-04-15 15:32 -------
Created an attachment (id=8650)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=8650&action=view)
Additional testcase

This is another testlet snipped out of gcc-dg-struct-layout-1-t026-01.
With the original patch, in struct S2503 field d overlaps field b,
and hence the store goes into the wrong place.
The idea of putting packed bit fields into rli->prev_field is bogus to
start with.  These fields are not necesasrily aligned like the underlying type,

and packed fields are not supposed to place more restrictions on the placement
of surrounding fields than ordinary non-bitfield types.  E.g. a packed
long long bitfield of size 8 is not expected to reserve more space than a
char field, even if it is followed by a non-packed bitfield of a different
type.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-04-15 22:21 -------
Created an attachment (id=8651)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=8651&action=view)
I'm currently testing this patch.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-04-29 19:58 -------
Created an attachment (id=8766)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=8766&action=view)
This patch fixes some more alignment inconsistencies.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
Attachment #8651 is|0                           |1
           obsolete|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-05-04 21:51 -------
Subject: Bug 20371

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	sh-elf-4_1-branch
Changes by:	amylaar@gcc.gnu.org	2005-05-04 21:51:35

Modified files:
	gcc            : ChangeLog tree.h stor-layout.c 

Log message:
	PR middle-end/20371:
	* tree.h (record_layout_info_s): New member prev_packed.
	* stor-layout.c (update_alignment_for_field): Fix comment about
	KNOWN_ALIGN.  For MS bitfields, if we start a new run, make sure
	we start it properly aligned.
	(place_field): At the beginning of a record, pass 0 as KNOWN_ALIGN
	to update_alignment_for_field, and recompute it afterwards using
	the alignment of the record.
	When a packed bitfield precedes an MS bitfield, don't add padding
	at the end of the packed bitfield on behalf of the base type of
	the packed bit field.
	Take possible record alignment > BIGGEST_ALIGNMENT into account
	when calculating actual_align.
	Only put packed bit fields into rli->prev_field if they end up
	suitably aligned.
	Update rli->remaining_in_alignment when we have already started a
	run of bit fields and we process a packed bit field.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=sh-elf-4_1-branch&r1=2.8142.2.8&r2=2.8142.2.9
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree.h.diff?cvsroot=gcc&only_with_tag=sh-elf-4_1-branch&r1=1.709&r2=1.709.2.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/stor-layout.c.diff?cvsroot=gcc&only_with_tag=sh-elf-4_1-branch&r1=1.228.2.1&r2=1.228.2.2



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-05-04 21:54 -------
I have found more problems with the ms bitfield code.  An
amended patch is here:
http://gcc.gnu.org/ml/gcc-patches/2005-05/msg00339.html

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-05-11 12:25 -------
Subject: Bug 20371

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	amylaar@gcc.gnu.org	2005-05-11 12:24:50

Modified files:
	gcc            : ChangeLog stor-layout.c tree.h 

Log message:
	PR middle-end/20371:
	* tree.h (record_layout_info_s): New member prev_packed.
	* stor-layout.c (update_alignment_for_field): Fix comment about
	KNOWN_ALIGN.  For MS bitfields, if we start a new run, make sure
	we start it properly aligned.
	(place_field): At the beginning of a record, pass 0 as KNOWN_ALIGN
	to update_alignment_for_field, and recompute it afterwards using
	the alignment of the record.
	When a packed bitfield precedes an MS bitfield, don't add padding
	at the end of the packed bitfield on behalf of the base type of
	the packed bit field.
	Don't adjust rli->bitpos at the end
	of an MS bitfield run if we already adjusted bitpos/offset for an
	alignment as large or larger than the bitfield type size.
	Take possible record alignment > BIGGEST_ALIGNMENT into account
	when calculating actual_align.
	Only put packed buit fields into rli->prev_field if they end up
	suitably aligned.
	Also set rli->remaining_in_alignment when we re-set rli->prev_field.
	Update rli->remaining_in_alignment when we have already started a
	run of bit fields and we process a packed bit field.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.8728&r2=2.8729
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/stor-layout.c.diff?cvsroot=gcc&r1=1.230&r2=1.231
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree.h.diff?cvsroot=gcc&r1=1.725&r2=1.726



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-05-11 13:19 -------
Fixed on mainline by patch applied today.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371

[Bug middle-end/20371] Some corner cases of MS bitfields don't work

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20371