[Bug target/39386] New: different computation results for O1 and O0 executables

[Bug target/39386] New: different computation results for O1 and O0 executables

[jxyang at cs dot utah dot edu]

This is seen on the version of avr-gcc 4.3.3 that gets built by the script that
comes with FemtoOS 0.88.

The program performs simple arithmatic and logical operations on a global
variable, and store the result in register R30:R31.

If the program is compiled with "avr-gcc -mmcu=atmega128 -O0", the final result
is 00. On the other hand, if the program is compiled with "avr-gcc
-mmcu=atmega128 -O1", the final result is 0xFF. Compiling at O2 gives me 0xFF
too.

Obviously, avr-gcc compiled the program wrong at one of the these optimization
levels.

This bug is observed in avr studio 4.15.


-- 
           Summary: different computation results for O1 and O0 executables
           Product: gcc
           Version: 4.3.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jxyang at cs dot utah dot edu
 GCC build triplet: --target=avr --with-gnu-ld --with-gnu-as --enable-
                    languages=c,c+
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: avr


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] different computation results for O1 and O0 executables

[jxyang at cs dot utah dot edu]

------- Comment #1 from jxyang at cs dot utah dot edu  2009-03-05 21:22 -------
Created an attachment (id=17400)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17400&action=view)
mis-calculated program


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[eric dot weddington at atmel dot com]

------- Comment #2 from eric dot weddington at atmel dot com  2009-03-18 17:47 -------
Confirmed with gcc 4.3.2.


-- 

eric dot weddington at atmel dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
      Known to fail|                            |4.3.3 4.3.2
   Last reconfirmed|0000-00-00 00:00:00         |2009-03-18 17:47:31
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[eric dot weddington at atmel dot com]

------- Comment #3 from eric dot weddington at atmel dot com  2009-03-21 16:52 -------
Please try using gcc 4.4 (HEAD). Anatoly Sokolov (AVR port maintainer) has
indicated to me that he does not see this bug when using HEAD/4.4.


-- 

eric dot weddington at atmel dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[eric dot weddington at atmel dot com]

------- Comment #4 from eric dot weddington at atmel dot com  2009-04-07 15:18 -------
*** Bug 39635 has been marked as a duplicate of this bug. ***


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[eric dot weddington at atmel dot com]

------- Comment #5 from eric dot weddington at atmel dot com  2009-04-07 15:23 -------
Adding additional version numbers in the known-to-fail field from comment in
duplicate bug #39635. Still waiting to hear if bug is present in 4.4.


-- 

eric dot weddington at atmel dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |wvangulik at xs4all dot nl
   GCC host triplet|i686-pc-linux-gnu           |
           Keywords|                            |wrong-code
      Known to fail|4.3.3 4.3.2                 |4.3.3 4.3.2 4.1.2 4.2.2
                   |                            |4.3.0 4.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[anitha dot boyapati at atmel dot com]

------- Comment #6 from anitha dot boyapati at atmel dot com  2009-08-10 13:27 -------
Confirmed with gcc 4.4.0. Using switch -O0 with avr-gcc 4.4.0 (-mmcu=atmega128)
gave the result 0 while -O1 gave 0xFF


-- 

anitha dot boyapati at atmel dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |anitha dot boyapati at atmel
                   |                            |dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[eric dot weddington at atmel dot com]

-- 

eric dot weddington at atmel dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
   Last reconfirmed|2009-03-18 17:47:31         |2009-08-10 13:40:54
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[anitha dot boyapati at atmel dot com]

------- Comment #7 from anitha dot boyapati at atmel dot com  2010-08-23 13:08 -------
Created an attachment (id=21547)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21547&action=view)
Reduced testcase

Reduced testcase to reproduce the bug.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[anitha dot boyapati at atmel dot com]

------- Comment #8 from anitha dot boyapati at atmel dot com  2010-08-23 13:25 -------
When -O2 is enabled result is 0xFF (255) which is incorrect. The code generated
for the following shift operation:

13:              return  (left << right);
+000000AB:   C004        RJMP      PC+0x0005      Relative jump
+000000AC:   0F22        LSL       R18            Logical Shift Left
+000000AD:   1F33        ROL       R19            Rotate Left Through Carry
+000000AE:   1F44        ROL       R20            Rotate Left Through Carry
+000000AF:   1F55        ROL       R21            Rotate Left Through Carry
+000000B0:   952A        DEC       R18            Decrement
+000000B1:   F7D2        BRPL      PC-0x05        Branch if plus
19:          g_6= func_1 (g_6, g_6);
+000000B2:   93200100    STS       0x0100,R18     Store direct to data space

'left' is stored in registers R21:R20:R19:R18 as it is typecasted to uint32_t
type while 'right' is stored in R18. Because of shift logic, R18 becomes 0xFF
in first iteration which is then stored.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[wvangulik at xs4all dot nl]

------- Comment #9 from wvangulik at xs4all dot nl  2010-08-23 13:36 -------
I already generated a simple testcase (although different)in
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39635. The comments from Andy in
the linked report might be worth reproducing here


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

Georg-Johann Lay <gjl at gcc dot gnu.org> changed:

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

--- Comment #14 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2011-07-25 16:08:04 UTC ---
Closed as FIXED.

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

--- Comment #13 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2011-07-25 15:45:51 UTC ---
Author: gjl
Date: Mon Jul 25 15:45:47 2011
New Revision: 176757

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=176757
Log:

    PR target/39386
    Backport from mainline r176756
    2011-07-25  Georg-Johann Lay
    * config/avr/avr.c (out_shift_with_cnt): Use tmp_reg as
    shift counter for x << x and x >> x shifts.


Modified:
    branches/gcc-4_6-branch/gcc/ChangeLog
    branches/gcc-4_6-branch/gcc/config/avr/avr.c

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

--- Comment #12 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2011-07-25 15:42:00 UTC ---
Author: gjl
Date: Mon Jul 25 15:41:55 2011
New Revision: 176756

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=176756
Log:

    PR target/39386
    * config/avr/avr.c (out_shift_with_cnt): Use tmp_reg as
    shift counter for x << x and x >> x shifts.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/avr/avr.c

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

--- Comment #11 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2011-07-23 13:48:14 UTC ---
Created attachment 24814
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24814
Minimal testcase

Minimal testcase that also miscompiles for newer versions of the compiler like
4.5.2 or 4.6.1.

Compiles with -Os to

shift:
    rjmp 2f
1:    lsl r24
    rol r25
2:    dec r24
    brpl 1b
    ret

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

Georg-Johann Lay <gjl at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.6.2

[Bug target/39386] [avr] different computation results for O1 and O0 executables

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39386

Georg-Johann Lay <gjl at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gjl at gcc dot gnu.org
      Known to fail|                            |

--- Comment #10 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2011-07-23 10:08:33 UTC ---
The problem arises from the following insns:

*ashlqi3
ashlhi3
ashlsi3

ashrqi3
ashrhi3
ashrsi3

*lshrqi3
lshrhi3
lshrsi3

For variable shifts the first constraint alternative is "=r,0,r"

avr.c:out_shift_with_cnt() reads 

  else if (register_operand (operands[2], QImode))
    {
      if (reg_unused_after (insn, operands[2]))
    op[3] = op[2];
      else
    {
      op[3] = tmp_reg_rtx;
      if (!len)
        strcat (str, (AS2 (mov,%3,%2) CR_TAB));
    }
    }

so that in the very rare, pathological case of op0 = op1 = op2 and op2 is
unused after the insn, wrong code gets generated. Fix is:

      if (reg_unused_after (insn, operands[2])
          && !reg_overlap_mentioned_p (operands[0], operands[2]))

or something like that.