[Bug c/20102] New: Incorrect floating point code generated when assigning to "packed" structure

[Bug c/20102] New: Incorrect floating point code generated when assigning to "packed" structure

[oakad at yahoo dot com]

gcc generates incorrect code when assigning to stack residing struct 
bearing "packed" attribute. Correct code is generated when assigning to global 
struct of the same type.
Example:
//-----
typedef struct
{
  char x1;
  char x2;
  char x3;
  float v1;
  float v2;
  float v3;
  char x4;
  char x5;
} __attribute__((packed,aligned(1))) ptype1;

ptype1 g;

void foo()
{
//....
 ptype1 p;
//....
 p.v1=g.v1;
 p.v2=p.v3;
//....
}
//-----
When compiled with:
powerpc-eabi-gcc -std=gnu99 -mcpu=505 -fsingle-precision-constant -mmultiple -
mstring -fno-common -ffast-math -O3 -g3 -memb -msdata=eabi -G 26624
the following incorrect code is generated:
//---
lfs   f3, -31984(r13)  //g.v1
lfs   f2, 395(r1)      //p.v3 - incorrect misaligned load (throws exception)
stfs  f3, 387(r1)      //p.v1 - incorrect misaligned store (throws exception)
stfs  f2, 391(r1)      //p.v2 - incorrect misaligned store (throws exception)
//---
Desired behavior: at least error message; at most - consistent, byte by byte 
load/store.

-- 
           Summary: Incorrect floating point code generated when assigning
                    to "packed" structure
           Product: gcc
           Version: 3.4.2
            Status: UNCONFIRMED
          Severity: critical
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: oakad at yahoo dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-cygwin
  GCC host triplet: i686-pc-cygwin
GCC target triplet: powerpc-eabi-*


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-20 13:34 -------
lfs   f2, 395(r1)      //p.v3 - incorrect misaligned load (throws exception)
Yes but the OS/kernel is supposed to handle the exception as any other OS on PPC would.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c                           |target


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[oakad at yahoo dot com]

------- Additional Comments From oakad at yahoo dot com  2005-02-20 21:16 -------
Three points: 
1. Handling of this case shall be consistent. When dealing with global 
variables compiler emits byte by byte copy code. With automatic variables, 
compiler emits single lfs/stfs. 
2. I believe that PPC architecture explicitly forbids unaligned floating loads 
and stores. 
3. The whole point in my code sniplet is to accelerate some communication 
protocol. Exception handler invocation is obviously out of question. 
 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-20 21:24 -------
(In reply to comment #2)
> 2. I believe that PPC architecture explicitly forbids unaligned floating loads 
> and stores. 

Wrong.  The PPC architecutures says that it will most likely to produce an unalignment exception but 
IIRC all unalignment exceptions have to be handled.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[oakad at yahoo dot com]

------- Additional Comments From oakad at yahoo dot com  2005-02-21 10:39 -------
I would like to notice that the issue only arises with packed structs, which 
are mostly found in places where exception handling is undesirable. Non-packed 
structs are always aligned anyway. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[amodra at bigpond dot net dot au]

------- Additional Comments From amodra at bigpond dot net dot au  2005-08-23 01:20 -------
Using -mstrict-align will avoid the unaligned accesses,  I haven't looked into
why there is a difference between the global and auto var..

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[oakad at yahoo dot com]

------- Additional Comments From oakad at yahoo dot com  2005-08-23 10:27 -------
I was dealing with packed structs in this case. Are you sure that 
-mstrict-align will not break packing? And I'm talking of both inter- and 
intra-struct packing, as in packed array of packed structs. 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[amodra at bigpond dot net dot au]

------- Additional Comments From amodra at bigpond dot net dot au  2005-08-24 00:44 -------
-mstrict-align should not affect aggregate layout.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[amodra at bigpond dot net dot au]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amodra at bigpond dot net
                   |                            |dot au


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-08-25 03:26 -------
-mstrict-align means don't produce loads which have unaligned loads.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

[Bug target/20102] Incorrect floating point code generated when assigning to "packed" structure

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20102

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

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

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-12-22 00:45:57 UTC ---
AS mentioned, GCC for powerpc defaults to non-strict alignment which means you
need to supply -mstrict-align to get the strict alignment requirements.