[Bug rtl-optimization/54524] New: Spurious add on sum of bitshifts

[Bug rtl-optimization/54524] New: Spurious add on sum of bitshifts

[jan.smets@alcatel-lucent.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

             Bug #: 54524
           Summary: Spurious add on sum of bitshifts
    Classification: Unclassified
           Product: gcc
           Version: 4.6.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: jan.smets@alcatel-lucent.com


There is an spurious addiu    a2,a2,1   which causes the result to be
incorrect.


tUint32 is unsigned long long


MIPS O32
GCC 4.6.3

 -nostdinc -funsigned-char  -fno-builtin -march=mips2 -G 0 -mno-gpopt
-msoft-float -fno-merge-constants -pipe -Wall -O2 -fno-optimize-sibling-calls
-fno-partial-inlining -fno-reorder-blocks -fno-strict-aliasing 


PUBLIC void function(
    tUint32                  index160,
    tUint8 *                 data)
{
    tUint8        temp_data[160/8];
    tUint32       scrambled_address;

    *(tUint64*)(temp_data+8)  = *(tUint64*)(data+8);

    scrambled_address = index160;

    printf("%16llX.\n",
           0x4000000000400000ull +
           (((tUint64)*(tUint16*)(temp_data+12)) << 24) +
           ((tUint64)(scrambled_address & 0xFF) << 56));
}


0x364c70c  27bdffd0    addiu    sp,sp,-48
0x364c710  afbf002c    sw       ra,44(sp)
0x364c714  8ca8000c    lw       t0,12(a1)   t0 = pp qq rr ss
0x364c718  3c030040    lui      v1,0x40     v1 = 00 40 00 00
0x364c71c  3c024000    lui      v0,0x4000   v0 = 40 00 00 00
0x364c720  00085402    srl      t2,t0,16    t2 = 00 00 pp qq
0x364c724  000a4e00    sll      t1,t2,24    t1 = qq 00 00 00
0x364c728  01233821    addu     a3,t1,v1    a3 = qq 40 00 00
0x364c72c  000a5202    srl      t2,t2,8     t2 = 00 00 00 pp
0x364c730  00e9482b    sltu     t1,a3,t1    t1 = carry?
0x364c734  01421021    addu     v0,t2,v0    v0 = 40 00 00 pp
0x364c738  00042600    sll      a0,a0,24    a0 = index160 << 24
0x364c73c  8ca30008    lw       v1,8(a1)
0x364c740  01221021    addu     v0,t1,v0    v0 += carry
0x364c744  00443021    addu     a2,v0,a0    a2 = v0 + index160 << 24
0x364c748  3c04066e    lui      a0,0x66e    \_ String address
0x364c74c  24842084    addiu    a0,a0,8324  /
0x364c750  24c60001    addiu    a2,a2,1     - Spurious add
0x364c754  afa30018    sw       v1,24(sp)
0x364c758  0cb402e1    jal      printf      - Takes a0 = string. a1 = low word,
a2 = high word
0x364c75c  afa8001c    sw       t0,28(sp)
0x364c760  8fbf002c    lw       ra,44(sp)
0x364c764  8fe00000    lw       zero,0(ra)   (internal extension)
0x364c768  03e00008    jr       ra
0x364c76c  27bd0030    addiu    sp,sp,48

[Bug rtl-optimization/54524] Spurious add on sum of bitshifts

[jan.smets@alcatel-lucent.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

--- Comment #1 from Jan Smets <jan.smets@alcatel-lucent.com> 2012-09-08 13:18:48 UTC ---
Created attachment 28152
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28152
testcase

The result is correct with -fno-forward-propagate

tUint64 is unsigned long long I meant. 


An even more reduced testcase is attached. Compile it with a MIPS cross
compiler:

./gcc  -nostdinc -funsigned-char  -fno-builtin -G 0 -mno-gpopt -msoft-float
-fno-merge-constants -pipe -Wall -O2 -fno-optimize-sibling-calls
-fno-partial-inlining -fno-reorder-blocks -fno-strict-aliasing  -c /tmp/test.c
-o test -mabi=32

[Bug rtl-optimization/54524] Spurious add on sum of bitshifts (forward-propagate issue)

[jan.smets@alcatel-lucent.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

Jan Smets <jan.smets@alcatel-lucent.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |major

[Bug rtl-optimization/54524] Spurious add on sum of bitshifts (forward-propagate issue)

[jan.smets@alcatel-lucent.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

Jan Smets <jan.smets@alcatel-lucent.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|mips                        |mipsisa64-octeon-elf
      Known to work|                            |4.5.0, 4.5.4
      Known to fail|                            |4.6.0, 4.6.3, 4.7.1, 4.8.0

--- Comment #2 from Jan Smets <jan.smets@alcatel-lucent.com> 2012-09-18 12:41:43 UTC ---
GCC 4.5.4 => CORRECT
====================

   0:   27bdffd0        addiu   sp,sp,-48
   4:   97a5001c        lhu     a1,28(sp)
   8:   000427c0        sll     a0,a0,0x1f
   c:   00053600        sll     a2,a1,0x18
  10:   3c030040        lui     v1,0x40
  14:   00052a02        srl     a1,a1,0x8
  18:   00a42821        addu    a1,a1,a0
  1c:   00c33821        addu    a3,a2,v1
  20:   3c024000        lui     v0,0x4000
  24:   00e6302b        sltu    a2,a3,a2
  28:   00a21021        addu    v0,a1,v0
  2c:   3c040000        lui     a0,0x0
                        2c: R_MIPS_HI16 .rodata
  30:   24840000        addiu   a0,a0,0
                        30: R_MIPS_LO16 .rodata
  34:   afbf002c        sw      ra,44(sp)
  38:   0c000000        jal     0 <function>
                        38: R_MIPS_26   printf
  3c:   00c23021        addu    a2,a2,v0         <= !
  40:   8fbf002c        lw      ra,44(sp)
  44:   03e00008        jr      ra
  48:   27bd0030        addiu   sp,sp,48



GCC 4.6.3
=========
   0:   27bdffd0        addiu   sp,sp,-48
   4:   97a5001c        lhu     a1,28(sp)
   8:   3c030040        lui     v1,0x40
   c:   3c024000        lui     v0,0x4000
  10:   00054600        sll     t0,a1,0x18
  14:   01033821        addu    a3,t0,v1
  18:   00052a02        srl     a1,a1,0x8
  1c:   00e8402b        sltu    t0,a3,t0
  20:   00a21021        addu    v0,a1,v0
  24:   000427c0        sll     a0,a0,0x1f
  28:   01023021        addu    a2,t0,v0
  2c:   00c43021        addu    a2,a2,a0
  30:   3c040000        lui     a0,0x0
                        30: R_MIPS_HI16 .rodata
  34:   24840000        addiu   a0,a0,0
                        34: R_MIPS_LO16 .rodata
  38:   afbf002c        sw      ra,44(sp)
  3c:   0c000000        jal     0 <function>
                        3c: R_MIPS_26   printf
  40:   24c60001        addiu   a2,a2,1       <= !
  44:   8fbf002c        lw      ra,44(sp)
  48:   03e00008        jr      ra
  4c:   27bd0030        addiu   sp,sp,48


GCC 4.8.0
=========
   0:   27bdffd0        addiu   sp,sp,-48
   4:   97a6001c        lhu     a2,28(sp)
   8:   3c030040        lui     v1,0x40
   c:   00062e00        sll     a1,a2,0x18
  10:   00a33821        addu    a3,a1,v1
  14:   00063202        srl     a2,a2,0x8
  18:   3c024000        lui     v0,0x4000
  1c:   00e5282b        sltu    a1,a3,a1
  20:   00c21021        addu    v0,a2,v0
  24:   000427c0        sll     a0,a0,0x1f
  28:   00a21021        addu    v0,a1,v0
  2c:   00443021        addu    a2,v0,a0
  30:   3c040000        lui     a0,0x0
                        30: R_MIPS_HI16 .rodata
  34:   24840000        addiu   a0,a0,0
                        34: R_MIPS_LO16 .rodata
  38:   afbf002c        sw      ra,44(sp)
  3c:   0c000000        jal     0 <function>
                        3c: R_MIPS_26   printf
  40:   24c60001        addiu   a2,a2,1        <= !
  44:   8fbf002c        lw      ra,44(sp)
  48:   03e00008        jr      ra
  4c:   27bd0030        addiu   sp,sp,48

[Bug rtl-optimization/54524] [4.6/4.7/4.8 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.6.4
            Summary|Spurious add on sum of      |[4.6/4.7/4.8 Regression]
                   |bitshifts                   |Spurious add on sum of
                   |(forward-propagate issue)   |bitshifts
                   |                            |(forward-propagate issue)

[Bug rtl-optimization/54524] [4.6/4.7/4.8 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[jan.smets@alcatel-lucent.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

--- Comment #3 from Jan Smets <jan.smets@alcatel-lucent.com> 2012-09-18 21:49:21 UTC ---
This commit introduced the bug:

http://gcc.gnu.org/viewcvs?view=revision&revision=161655
(or 182cf5a9a415f31df0f9a10e46faed1221484a35 in git)

[Bug rtl-optimization/54524] [4.6/4.7/4.8 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-09-19
     Ever Confirmed|0                           |1
           Severity|major                       |normal

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-09-19 01:39:38 UTC ---
As far as I can tell, Richard's patch just exposed a latent bug as I was
looking into the dumps.



In insn 28, replacing
 (ltu:SI (reg:SI 231 [+4 ])
        (reg:SI 229 [+4 ]))
 with (const_int 1 [0x1])
Changed insn 28
deferring rescan insn with uid = 28.

after cse:
(insn 27 43 28 2 (set (reg:SI 231 [+4 ])
        (plus:SI (reg:SI 229 [+4 ])
            (const_int 0 [0]))) t.c:24 10 {*addsi3}
     (nil))

(insn 28 27 29 2 (set (reg:SI 212)
        (ltu:SI (reg:SI 231 [+4 ])
            (reg:SI 229 [+4 ]))) t.c:24 521 {*sltu_sisi}
     (nil))

So this is a bug in CSE which is not simplifying that addsi3 into just a move
for some reason.

[Bug rtl-optimization/54524] [4.6/4.7/4.8 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-09-19 08:46:37 UTC ---
Still non-canonical RTL should not result in wrong-code.

[Bug rtl-optimization/54524] [4.6/4.7/4.8 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-09-22 22:26:37 UTC ---
I think this code:
  /* (LTU/GEU (PLUS a C) C), where C is constant, can be simplified to
     (GEU/LTU a -C).  Likewise for (LTU/GEU (PLUS a C) a).  */

Is what is causing the issue as that is only true if C is non-zero.

[Bug rtl-optimization/54524] [4.6/4.7/4.8 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-11-02 23:32:38 UTC ---
Author: pinskia
Date: Fri Nov  2 23:32:32 2012
New Revision: 193111

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=193111
Log:
2012-11-02  Andrew Pinski  <apinski@cavium.com>

        PR rtl-opt/54524
        * simplify-rtx.c (simplify_relational_operation_1): Don't simplify
        (LTU/GEU (PLUS a 0) 0) into (GEU/LTU a 0) since they are not
equivalent.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/simplify-rtx.c

[Bug rtl-optimization/54524] [4.6/4.7 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.8.0
            Summary|[4.6/4.7/4.8 Regression]    |[4.6/4.7 Regression]
                   |Spurious add on sum of      |Spurious add on sum of
                   |bitshifts                   |bitshifts
                   |(forward-propagate issue)   |(forward-propagate issue)
      Known to fail|4.8.0                       |

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-11-02 23:33:46 UTC ---
Fixed on the trunk.

[Bug rtl-optimization/54524] [4.6/4.7 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[jan.smets@alcatel-lucent.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

--- Comment #9 from Jan Smets <jan.smets@alcatel-lucent.com> 2012-11-03 18:25:32 UTC ---
Verified. Thanks Andrew.
(Should I mark it as Resolved?)

[Bug rtl-optimization/54524] [4.6/4.7 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug rtl-optimization/54524] [4.6/4.7 Regression] Spurious add on sum of bitshifts (forward-propagate issue)

[jan.smets@alcatel-lucent.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54524

Jan Smets <jan.smets@alcatel-lucent.com> changed:

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

--- Comment #10 from Jan Smets <jan.smets@alcatel-lucent.com> 2013-01-07 13:30:40 UTC ---
Verified