[Bug rtl-optimization/47763] New: Useless initialization of register

[Bug rtl-optimization/47763] New: Useless initialization of register

[jiez at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

           Summary: Useless initialization of register
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: jiez@gcc.gnu.org


For arm-eabi target and options "-O2 -funroll-loops", gcc compiles the
following empty function

foo()
{
}

to

foo:
        mov     r0, #0
        bx      lr

Apparently the first instruction is useless.

[Bug rtl-optimization/47763] Useless initialization of register

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-02-16 07:35:09 UTC ---
The function is not empty, it is returning an uninitialized value as int.  If
you want an empty function, use void foo () {}

[Bug rtl-optimization/47763] Useless initialization of register

[jiez at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #2 from Jie Zhang <jiez at gcc dot gnu.org> 2011-02-16 07:42:45 UTC ---
OK. From this point, it's not empty. But if it returns an uninitialized value,
why bother initialize r0 to 0.

Btw, the patch in reviewing:

http://gcc.gnu.org/ml/gcc-patches/2011-02/msg00787.html

I just realized that we need a new PR number for the test case. So I reported
this bug.

[Bug rtl-optimization/47763] Useless initialization of register

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-02-16 07:48:24 UTC ---
Why should we care about code quality of function with undefined behavior?
Various optimizations add zero initialization or assume zero initialization of
undefined variables.

[Bug rtl-optimization/47763] Useless initialization of register

[jiez at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #4 from Jie Zhang <jiez at gcc dot gnu.org> 2011-02-16 07:59:19 UTC ---
Yeah, normally we don't care about such cases. But this one comes from
dhrystone. If it can be fixed cleanly, why not do it?

[Bug rtl-optimization/47763] Useless initialization of register

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-02-16 08:08:33 UTC ---
(In reply to comment #4)
> Yeah, normally we don't care about such cases. But this one comes from
> dhrystone. If it can be fixed cleanly, why not do it?

Then the whole dhrystone benchmark is invalid and should not be trusted.  I
think people should report the bug to the benchmark maker and then declare all
previous results of the benchmark as invalid.  In fact any benchmark which
depends on undefined code should not be trusted.

[Bug rtl-optimization/47763] Useless initialization of register

[jiez at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #6 from Jie Zhang <jiez at gcc dot gnu.org> 2011-02-16 08:32:14 UTC ---
I think we all know that dhrystone is a bad benchmark. But some users don't.
Which is more difficult: educating all users about that vs apply a simple patch
in GCC to remove the useless initialization? 4.4 does has such initialization.
But 4.5 has. Fixing this is an improvement and does not hurt. From another
point, leaving it unfixed does not help GCC.

[Bug rtl-optimization/47763] Useless initialization of register

[jiez at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #7 from Jie Zhang <jiez at gcc dot gnu.org> 2011-02-16 08:49:46 UTC ---
Sorry, in my last comment, I meant to say "4.4 does NOT have such
initialization".

[Bug rtl-optimization/47763] Useless initialization of register

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.02.16 11:04:15
     Ever Confirmed|0                           |1

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-02-16 11:04:15 UTC ---
The difference is that with -funroll-loops the init_regs pass isn't presented
with

(insn 8 7 6 2 (clobber (reg:SI 59 [ <retval> ])) t.c:1 -1
     (nil))

and thus inserts a zero-initialization instruction.  I wonder whether
it could instead insert a clobber.

[Bug rtl-optimization/47763] Useless initialization of register

[jiez at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #9 from Jie Zhang <jiez at gcc dot gnu.org> 2011-02-16 11:29:50 UTC ---
The clobber is optimized away in 172r.cprop3 because the register renaming in
171r.web breaks the def-use relationship between the clobber and the use in the
following instruction when -funroll-loops is used. My patch prevents the
renaming in such case.

[Bug rtl-optimization/47763] Useless initialization of register

[schwab@linux-m68k.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #10 from Andreas Schwab <schwab@linux-m68k.org> 2011-02-16 18:26:31 UTC ---
There is no undefined behaviour if all calls to foo ignore the returned value.

[Bug rtl-optimization/47763] Useless initialization of register

[jiez at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

--- Comment #11 from Jie Zhang <jiez at gcc dot gnu.org> 2011-02-23 00:25:38 UTC ---
Author: jiez
Date: Wed Feb 23 00:25:34 2011
New Revision: 170422

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=170422
Log:
    PR rtl-optimization/47763
    * web.c (web_main): Ignore naked clobber when replacing register.

    testsuite/
    PR rtl-optimization/47763
    * gcc.dg/pr47763.c: New test.

Added:
    trunk/gcc/testsuite/gcc.dg/pr47763.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/web.c

[Bug rtl-optimization/47763] Useless initialization of register

[jiez at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47763

Jie Zhang <jiez at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.6.0

--- Comment #12 from Jie Zhang <jiez at gcc dot gnu.org> 2011-02-23 00:27:15 UTC ---
Fixed.