[Bug c/22362] New: Register allocation problem in combination with -funit-at-a-time and global register variables

[Bug c/22362] New: Register allocation problem in combination with -funit-at-a-time and global register variables

[peter at hawkins dot emu dot id dot au]

Hi...

I'm using debian's unstable gcc 4.0:
$ gcc -v
Using built-in specs.
Target: i486-linux-gnu
Configured with: ../src/configure -v
--enable-languages=c,c++,java,f95,objc,ada,treelang --prefix=/usr
--enable-shared --with-system-zlib --libexecdir=/usr/lib --enable-nls
--without-included-gettext --enable-threads=posix --program-suffix=-4.0
--enable-__cxa_atexit --enable-libstdcxx-allocator=mt --enable-clocale=gnu
--enable-libstdcxx-debug --enable-java-gc=boehm --enable-java-awt=gtk
--with-java-home=/usr/lib/jvm/java-1.4.2-gcj-4.0-1.4.2.0/jre --enable-mpfr
--disable-werror --enable-checking=release i486-linux-gnu
Thread model: posix
gcc version 4.0.1 20050701 (prerelease) (Debian 4.0.0-12)

I'm trying to compile the following code (which is a reduced test case generated
from mercury_deep_copy.c in the Mercury compiler version ROTD 20050707 from
http://www.cs.mu.oz.au/mercury/):

register unsigned int reg0 __asm__("esi");
register unsigned int reg1 __asm__("edi");
register unsigned int reg2 __asm__("ebx");

static unsigned int
MR_deep_copy_typeclass_info(unsigned int *typeclass_info,
    void *lower_limit, void *upper_limit)
{
    int i;

    for (i = 5; i > 0; i--)
    {
        typeclass_info[i] = (unsigned int) MR_deep_copy_typeclass_info(
            (unsigned int *)typeclass_info[i], lower_limit, lower_limit);
    }
    return 0;
}

unsigned int func(void)
{
    return MR_deep_copy_typeclass_info(0, 0, 0);
}

using this command line:
$ gcc -c test3.c -O -funit-at-a-time

which produces this:
test3.c: In function 'MR_deep_copy_typeclass_info':
test3.c:17: error: unable to find a register to spill in class 'GENERAL_REGS'
test3.c:17: error: this is the insn:
(insn 27 26 28 1 (set (reg:SI 0 ax)
        (mem:SI (plus:SI (reg:SI 70)
                (const_int 4 [0x4])) [0 S4 A32])) 41 {*movsi_1}
(insn_list:REG_DEP_TRUE 22 (nil))
    (nil))
test3.c:17: confused by earlier errors, bailing out

The same code compiles with -O1 and no -funit-at-a-time. It also compiles with
gcc-3.4, so this is a gcc 4 regression. I suspect this means -funit-at-a-time
behaves badly under high register pressure.

=)
Peter

-- 
           Summary: Register allocation problem in combination with -funit-
                    at-a-time and global register variables
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: peter at hawkins dot emu dot id dot au
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22362

[Bug target/22362] [3.4/4.0/4.1 Regression] static function calls and global register variables

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-07-08 14:37 -------
What is happening is that MR_deep_copy_typeclass_info is not being inlined but instead GCC changes 
the how the arguments are passed, via registers.  This is a target bug.
The bug is in:
      /* Use register calling convention for local functions when possible.  */
      if (!TARGET_64BIT && !user_convention && decl
          && flag_unit_at_a_time && !profile_flag)
        {

This is a latent bug in 3.4.0 also.  In a way this could be reproduced in 3.4.0.  This is a 3.4.0 and 4.0.0 
and 4.1.0 regression (even though it is harder to reproduce in 3.4.0 and 4.1.0).

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
          Component|c                           |target
     Ever Confirmed|                            |1
           Keywords|                            |ice-on-valid-code
   Last reconfirmed|0000-00-00 00:00:00         |2005-07-08 14:37:52
               date|                            |
            Summary|Register allocation problem |[3.4/4.0/4.1 Regression]
                   |in combination with -funit- |static function calls and
                   |at-a-time and global        |global register variables
                   |register variables          |
   Target Milestone|---                         |3.4.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22362

[Bug target/22362] [3.4/4.0/4.1 Regression] static function calls and global register variables

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-07-22 21:13 -------
Moving to 4.0.2 pre Mark.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4.5                       |4.0.2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22362

[Bug target/22362] [3.4/4.0/4.1 Regression] static function calls and global register variables

[bonzini at gcc dot gnu dot org]

------- Additional Comments From bonzini at gcc dot gnu dot org  2005-08-17 12:36 -------
A solution could be to count the number of global register vars, and decrease
the number of regparm'd parameters accordingly.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22362

[Bug target/22362] [3.4/4.0/4.1 Regression] static function calls and global register variables

[jakub at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |jakub at gcc dot gnu dot org
                   |dot org                     |
                URL|                            |http://gcc.gnu.org/ml/gcc-
                   |                            |patches/2005-
                   |                            |09/msg00279.html
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2005-07-08 14:37:52         |2005-09-05 13:22:14
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22362

[Bug target/22362] [3.4/4.0/4.1 Regression] static function calls and global register variables

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-09-06 19:58 -------
Subject: Bug 22362

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	jakub@gcc.gnu.org	2005-09-06 19:57:57

Modified files:
	gcc            : ChangeLog 
	gcc/config/i386: i386.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.target/i386: pr22362.c 

Log message:
	PR target/22362
	* config/i386/i386.c (ix86_function_regparm): Make sure automatic regparm
	for internal functions doesn't use registers used by global registers
	variables.  Use fewer register parameters if there are global register
	variables.
	
	* gcc.target/i386/pr22362.c: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.9905&r2=2.9906
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/i386/i386.c.diff?cvsroot=gcc&r1=1.857&r2=1.858
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.6019&r2=1.6020
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.target/i386/pr22362.c.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22362

[Bug target/22362] [3.4/4.0/4.1 Regression] static function calls and global register variables

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-09-06 20:03 -------
Subject: Bug 22362

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-4_0-branch
Changes by:	jakub@gcc.gnu.org	2005-09-06 20:03:28

Modified files:
	gcc            : ChangeLog 
	gcc/config/i386: i386.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.target/i386: pr22362.c 

Log message:
	PR target/22362
	* config/i386/i386.c (ix86_function_regparm): Make sure automatic regparm
	for internal functions doesn't use registers used by global registers
	variables.  Use fewer register parameters if there are global register
	variables.
	
	* gcc.target/i386/pr22362.c: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=2.7592.2.414&r2=2.7592.2.415
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/i386/i386.c.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=1.795.6.10&r2=1.795.6.11
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=1.5084.2.385&r2=1.5084.2.386
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.target/i386/pr22362.c.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=NONE&r2=1.1.2.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22362

[Bug target/22362] [3.4 Regression] static function calls and global register variables

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-06 20:13 -------
Still a latent bug/regression on the 3.4 branch.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |4.0.0 4.0.1
      Known to work|                            |3.3.3 4.0.2 4.1.0
            Summary|[3.4/4.0/4.1 Regression]    |[3.4 Regression] static
                   |static function calls and   |function calls and global
                   |global register variables   |register variables
   Target Milestone|4.0.2                       |3.4.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22362