[Bug c/19417] New: return value of some functions not properly set (in r3)

[Bug c/19417] New: return value of some functions not properly set (in r3)

[cort at fsmlabs dot com]

I have included 2 .i and .s files.  One correctly sets the return value (r3) and
one does not.  The addition of the nonleaf() call causes the return value to be
set properly.  If you search the .s files for "cort start return" "cort end
return" you can see it.

It seems that the return value isn't set properly if the function does not call
another function at the out: label.

I'm cross compiling from a 32-bit x86 to a 32-bit PowerPC (of the 6xx/74xx variety).

-- 
           Summary: return value of some functions not properly set (in r3)
           Product: gcc
           Version: 3.3.2
            Status: UNCONFIRMED
          Severity: critical
          Priority: P1
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: cort at fsmlabs dot com
                CC: gcc-bugs at gcc dot gnu dot org
  GCC host triplet: i386-linux
GCC target triplet: ppc-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug c/19417] return value of some functions not properly set (in r3)

[cort at fsmlabs dot com]

------- Additional Comments From cort at fsmlabs dot com  2005-01-12 23:48 -------
Created an attachment (id=7944)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7944&action=view)
working .i file


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug c/19417] return value of some functions not properly set (in r3)

[cort at fsmlabs dot com]

------- Additional Comments From cort at fsmlabs dot com  2005-01-12 23:48 -------
Created an attachment (id=7943)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7943&action=view)
non-working .i


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug c/19417] return value of some functions not properly set (in r3)

[cort at fsmlabs dot com]

------- Additional Comments From cort at fsmlabs dot com  2005-01-12 23:49 -------
Created an attachment (id=7946)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7946&action=view)
non-working .s file


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug c/19417] return value of some functions not properly set (in r3)

[cort at fsmlabs dot com]

------- Additional Comments From cort at fsmlabs dot com  2005-01-12 23:49 -------
Created an attachment (id=7945)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7945&action=view)
working .s file


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug c/19417] return value of some functions not properly set (in r3)

[cort at fsmlabs dot com]

------- Additional Comments From cort at fsmlabs dot com  2005-01-12 23:51 -------
More info - host system info, gcc build flags and compiler flags used to
build the source files.

Script started on Wed 12 Jan 2005 03:42:36 PM MST
cort@host115<rtlinuxpro-2.2>$ /opt/rtldk-2.2/bin/ppc-linux-gcc -v
Reading specs from /opt/rtldk-2.2/bin/../lib/gcc-lib/ppc-linux/3.3.2/specs
Configured with: ../gcc-3.3.2/configure --prefix=/opt/rtldk-2.2/
--target=ppc-linux --enable-languages=c,c++,f77 --disable-nls
--enable-threads=posix --enable-symvers=gnu --enable-__cxa_atexit
--enable-shared --enable-c99 --host=i386-linux --enable-altivec
Thread model: posix
gcc version 3.3.2
cort@host115<rtlinuxpro-2.2>$ cat /etc/issue
Fedora Core release 2 (Tettnang)
Kernel \r on an \m

cort@host115<rtlinuxpro-2.2>$ uname -a
Linux host115.fsmlabs.com 2.4.25-rtl #18 SMP Fri May 28 16:51:21 MDT 2004 i686
i686 i386 GNU/Linux
cort@host115<rtlinuxpro-2.2>$ exit

Script done on Wed 12 Jan 2005 03:42:49 PM MST

built with:

/opt/rtldk-2.2/bin/ppc-linux-gcc -D__RTCORE_KERNEL__ -nostdinc -isystem
/opt/rtldk-2.2/bin/../lib/gcc-lib/ppc-linux/3.3.2/include/
-I/sys/rtlinuxpro-2.2/include/app/rtl -I/sys/rtlinuxpro-2.2/linux/include
-I/sys/rtlinuxpro-2.2/linux/arch/ppc  -I/sys/rtlinuxpro-2.2/include -Wall
-Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common
-I/sys/linuxrtl-2.4-synergy-2.2/arch/ppc -fsigned-char -msoft-float -pipe
-ffixed-r2 -Wno-uninitialized -mmultiple -mstring -DMODULE --save-temps -O2  -o
kill.o -c kill.c
ppc-linux-gcc: warning: -pipe ignored because -save-temps specified



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug c/19417] return value of some functions not properly set (in r3)

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-13 00:02 -------
I see nothing wrong with the code:
.L139:
#APP
	/* cort start return */

#NO_APP
	mr 3,0
.L137:
--- return
-----
r0 is set before L139

r3 is set before L137:
	li 3,73
	beq- 0,.L137

So no nothing is wrong with the code as far as I can see.

Can you point out where the problem is because I cannot find it?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug middle-end/19417] return value of some functions not properly set (in r3)

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-13 00:38 -------
Oh, now I see the problem, we are using r0 twice which is just wrong.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal
          Component|c                           |middle-end
           Keywords|                            |wrong-code
           Priority|P1                          |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug middle-end/19417] return value of some functions not properly set (in r3)

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-13 00:44 -------
This works with both 3.4.4 and 4.0.0, I don't know if it is a latent bug or not.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   GCC host triplet|i386-linux                  |
 GCC target triplet|ppc-linux                   |powerpc-*-*
      Known to work|                            |3.4.4 4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug middle-end/19417] [3.3 only] return value of some functions not properly set (in r3)

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|return value of some        |[3.3 only] return value of
                   |functions not properly set  |some functions not properly
                   |(in r3)                     |set (in r3)
   Target Milestone|---                         |3.3.6


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417

[Bug middle-end/19417] [3.3 only] return value of some functions not properly set (in r3)

[dje at gcc dot gnu dot org]

------- Additional Comments From dje at gcc dot gnu dot org  2005-01-15 00:40 -------
The inlined asm rtl_a_set() does not list any outputs or clobbers, so it looks
like the compiler safely can assume that the value is unchanged.  I am not sure
why the compiler can assume that the value is zero, but the asm definitely will
confuse it about the status of the register.  Showing the register as killed in
the asm fixes the code generation.

I am going to mark this as invalid.  Please reopen if you can explain to me why
GCC should behave well with an asm statement that kills the value in a register
but only lists inputs to the data dependency and flow parts of the compiler.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19417