[Bug c/43700] New: global register variables defect

[Bug c/43700] New: global register variables defect

[steffenschmidt dot ext at siemens dot com]

We're making use of global register variables, which work fine in GCC version
4.3.2 and 4.3.3. GCC 4.4.1 and 4.4.3 however produce slightly different code,
which seems to be incorrect.

A brief example is:

//---------------test_register_call.c-------------------
register unsigned long long *g_reg_counter asm("$20");

void register_add (void)
{
  *g_reg_counter++;
}
//---------------test_register_call.c-------------------

Compiled with: 
mips-elf-gcc.exe -save-temps test_register_call.c
This produces the following assembler file using GCC 4.3.2

//---------------test_register_call_4.3.2.s-------------
        .file   1 "test_register_call.c"
        .section .mdebug.abi32
        .previous
        .gnu_attribute 4, 1
        .text
        .align  2
        .globl  register_add
        .set    nomips16
        .ent    register_add
register_add:
        .frame  $fp,8,$31               # vars= 0, regs= 1/0, args= 0, gp= 0
        .mask   0x40000000,-4
        .fmask  0x00000000,0
        .set    noreorder
        .set    nomacro

        addiu   $sp,$sp,-8
        sw      $fp,4($sp)
        move    $fp,$sp
        move    $2,$20
        addiu   $2,$2,8
        move    $20,$2
        move    $sp,$fp
        lw      $fp,4($sp)
        addiu   $sp,$sp,8
        j       $31
        nop

        .set    macro
        .set    reorder
        .end    register_add
        .size   register_add, .-register_add
        .ident  "GCC: 4.3.2"
//---------------test_register_call_4.3.2.s-------------

GCC 4.4.1 produces different assembly code (lines are marked):

//---------------test_register_call_4.4.1.s-------------
        .file   1 "test_register_call.c"
        .section .mdebug.abi32
        .previous
        .gnu_attribute 4, 1
        .text
        .align  2
        .globl  register_add
        .set    nomips16
        .ent    register_add
        .type   register_add, @function
register_add:
        .frame  $fp,8,$31               # vars= 0, regs= 2/0, args= 0, gp= 0
        .mask   0x40100000,-4
        .fmask  0x00000000,0
        .set    noreorder
        .set    nomacro

        addiu   $sp,$sp,-8
        sw      $fp,4($sp)
        sw      $20,0($sp)              # <== $20 is stored to stack
        move    $fp,$sp
        move    $2,$20
        addiu   $2,$2,8
        move    $20,$2
        move    $sp,$fp
        lw      $fp,4($sp)
        lw      $20,0($sp)              # <== $20 is restored from stack
        addiu   $sp,$sp,8
        j       $31
        nop

        .set    macro
        .set    reorder
        .end    register_add
        .size   register_add, .-register_add
        .ident  "GCC: 4.4.1"
//---------------test_register_call_4.4.1.s-------------

As can be seen in the example, MIPS register $20 is the global variable
register. The variable is modified in the function and the register should
therefore not be saved and restored. GCC 4.3.2 seems to be correct here. GCC
4.4.1, however, stores and restores the register $20 content.


-- 
           Summary: global register variables defect
           Product: gcc
           Version: 4.4.3
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: steffenschmidt dot ext at siemens dot com
 GCC build triplet: i686-mingw32
  GCC host triplet: i686-mingw32
GCC target triplet: mips-elf


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] global register variables defect

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2010-04-09 09:50 -------
Must be a target issue, can't reproduce this on x86_64 with

register unsigned long *g_reg_counter asm("%r10");
void register_add (void)
{
  *g_reg_counter++;
}


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c                           |target


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] global register variables defect

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2010-04-09 18:12 -------
Well $20 is a callee save register.

The back-end emits a save for some reason ...


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c                           |target
           Keywords|                            |wrong-code


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
  GCC build triplet|i686-mingw32                |
   GCC host triplet|i686-mingw32                |
 GCC target triplet|mips-elf                    |mips*-*
            Summary|global register variables   |[4.4/4.5/4.6 Regression]
                   |defect                      |global register variables
                   |                            |defect
   Target Milestone|---                         |4.4.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2010-04-09 21:52 -------
Confirmed.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2010-04-09 21:52:01
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[jakub at gcc dot gnu dot org]

------- Comment #4 from jakub at gcc dot gnu dot org  2010-04-12 08:41 -------
Isn't this just a user error then?  You should have used -ffixed-20 if you use
a call saved register as global register IMHO.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[mikpe at it dot uu dot se]

------- Comment #5 from mikpe at it dot uu dot se  2010-04-12 09:02 -------
(In reply to comment #4)
> Isn't this just a user error then?  You should have used -ffixed-20 if you use
> a call saved register as global register IMHO.

gcc's documentation (I'm looking at the global register variables node in 4.5's
extend.texi) states that passing -ffixed-reg is only needed if the translation
unit doesn't contain a declaration of that global register variable.

In this PR's test case the declaration is present, so -ffixed-20 should not be
needed.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[pinskia at gcc dot gnu dot org]

------- Comment #6 from pinskia at gcc dot gnu dot org  2010-04-12 18:18 -------
*** Bug 43732 has been marked as a duplicate of this bug. ***


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |lance604 at gmail dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[pinskia at gcc dot gnu dot org]

------- Comment #7 from pinskia at gcc dot gnu dot org  2010-04-12 18:20 -------
Well -ffixed-reg-r20 is also broken the same way :) See the duplicated bug
which has a patch which I have not looked into to see if it is the correct fix
yet or not.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[mikpe at it dot uu dot se]

------- Comment #8 from mikpe at it dot uu dot se  2010-04-15 19:55 -------
4.3 is also broken.  Using the simpler test case in PR43732 and a collection of
cross-compilers to mips-unknown-linux I see:

gcc-4.1.2: -O2 only: ok; -O2 -ffixed-23: ok
gcc-4.2.4: -O2 only: ok; -O2 -ffixed-23: ok
gcc-4.3.x: -O2 only: ok; -O2 -ffixed-23: bug (for x in 0..4)
gcc-4.4.x: -O2 only: bug; -O2 -ffixed-23: bug (x=0, x=3)
gcc-4.5.0: -O2 only: bug; -O2 -ffixed-23: bug

So -ffixed- broke between 4.2 and 4.3, and global register variables broke
between 4.3 and 4.4.


-- 

mikpe at it dot uu dot se changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mikpe at it dot uu dot se


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[mikpe at it dot uu dot se]

------- Comment #9 from mikpe at it dot uu dot se  2010-04-15 22:30 -------
MIPS global register variables worked up to r144963 but broke in r144964:

URL: http://gcc.gnu.org/viewcvs?root=3Dgcc&view=3Drev&rev=3D144964
Log:
2009-03-19  Alexandre Oliva  <aoliva@redhat.com>

        * reginfo.c (globalize_reg): Recompute derived reg sets.

The corresponding gcc-patches thread started here:
http://gcc.gnu.org/ml/gcc-patches/2009-03/msg00891.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[mikpe at it dot uu dot se]

------- Comment #10 from mikpe at it dot uu dot se  2010-04-16 13:04 -------
The -ffixed-$reg option worked for MIPS up to r128346 but broke in r128347:

Author: rsandifo
Date: Mon Sep 10 15:21:18 2007
New Revision: 128347

URL: http://gcc.gnu.org/viewcvs?root=3Dgcc&view=3Drev&rev=3D128347
Log:
gcc/
        * config/mips/mips.c (mips_global_pointer): Check
        call_really_used_regs instead of call_used_regs.
        (mips_save_reg_p): Likewise.  Save all call-saved registers
        if current_function_saves_all_registers.  Fix indentation.
        No longer treat $18 as a special case.
        (compute_frame_size): Guard FPR loop with TARGET_HARD_FLOAT.

See also the thread around:
http://gcc.gnu.org/ml/gcc-patches/2007-09/msg00886.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700

[Bug target/43700] [4.4/4.5/4.6 Regression] global register variables defect

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.4.4                       |4.4.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43700