[Bug middle-end/11135] New: It ought to be possible to make PIC_OFFSET_TABLE_REGNUM a pseudo

[Bug middle-end/11135] New: It ought to be possible to make PIC_OFFSET_TABLE_REGNUM a pseudo

[rearnsha@gcc.gnu.org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11135

           Summary: It ought to be possible to make PIC_OFFSET_TABLE_REGNUM
                    a pseudo
           Product: gcc
           Version: unknown
            Status: NEW
          Keywords: pessimizes-code
          Severity: enhancement
          Priority: P5
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: rearnsha@gcc.gnu.org
                CC: gcc-bugs@gcc.gnu.org
OtherBugsDependingO 592
             nThis:

Currently the documentation says:

> @findex PIC_OFFSET_TABLE_REGNUM
> The register number of the register used to address a table of static
> data addresses in memory.  In some cases this register is defined by a
> processor's ``application binary interface'' (ABI)@.  When this macro
> is defined, RTL is generated for this register once, as with the stack
> pointer and frame pointer registers.  If this macro is not defined, it
> is up to the machine-dependent files to allocate such a register (if
> necessary).  Note that this register must be fixed when in use (e.g.@:
> when @code{flag_pic} is true).

This pessimizes code when the pic register is used in a leaf function, or when
there is only a single use (since the register is fixed for the entire
function).  It ought to be possible for a machine description to make the PIC
register a pseudo that is allocated as normal.  Then the register can be either
call-clobbered or call saved as appropriate and can even be spilled onto the
stack if that is preferable.

[Bug middle-end/11135] It ought to be possible to make PIC_OFFSET_TABLE_REGNUM a pseudo

[pinskia@physics.uc.edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11135



------- Additional Comments From pinskia@physics.uc.edu  2003-06-09 15:47 -------
This also effects powerpc-apple-darwin which uses a fixed PIC register of r31 which 
sometimes gets saved even though a function does not need it.

[Bug middle-end/11135] It ought to be possible to make PIC_OFFSET_TABLE_REGNUM a pseudo

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11135


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4                         |---


------- Additional Comments From pinskia at physics dot uc dot edu  2003-07-23 19:28 -------
Most likely not going to be happening for 3.4 so removing target milestone.

[Bug middle-end/11135] It ought to be possible to make PIC_OFFSET_TABLE_REGNUM a pseudo

[rearnsha at gcc dot gnu dot org]

------- Comment #3 from rearnsha at gcc dot gnu dot org  2006-01-17 20:22 -------
Subject: Bug 11135

Author: rearnsha
Date: Tue Jan 17 20:22:19 2006
New Revision: 109839

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=109839
Log:
        PR target/592
        PR middle-end/11135
        * arm.h (struct machine_function): Add pic_reg.
        * arm.c (arm_pic_register): Make unsigned.
        (arm_override_options): Only set arm_pic_register if
        TARGET_SINGLE_PIC_BASE.
        (use_return_insn): Only test for a pic register if it is fixed.
        (arm_compute_save_reg0_reg12_mask): Likewise.
        (thumb_compute_save_reg_mask): Likewise.
        (legitimate_pic_operand): Factor out some known invariants.
        (legitimize_pic_address): If we don't have a fixed pic register,
        then set up a pseudo in the function entry sequence.  Handle the
        pic base being in a pseudo.
        (arm_load_pic_register): Handle the pic register being in a pseudo.
        (arm_expand_prologue): Only set up the pic register if it is fixed.
        (thumb_expand_prologue): Likewise.
        * arm.md (pic_load_addr_based): Handle the pic base being a pseudo.
        (pic_load_addr_based_insn): Likewise.
        (builtin_setjmp_receiver): Don't restore the pic base if it isn't
        fixed.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/arm/arm.c
    trunk/gcc/config/arm/arm.h
    trunk/gcc/config/arm/arm.md


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11135

[Bug middle-end/11135] It ought to be possible to make PIC_OFFSET_TABLE_REGNUM a pseudo

[rearnsha at gcc dot gnu dot org]

------- Comment #4 from rearnsha at gcc dot gnu dot org  2006-01-17 20:32 -------
Closing as WORKSFORME.  I didn't have to change anything in the middle-end in
order to fix the ARM back-end.

Maybe the documentation should be updated to reflect this status.


-- 

rearnsha at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |WORKSFORME
   Target Milestone|---                         |4.2.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11135