[Bug target/14580] New: [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[Bug target/14580] New: [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

This is a recent regression and it causes libjava not to build.
Here is a simple testcase:

#include <sys/types.h>
#include <sys/mman.h>
template <typename T_implPtr, typename T_implLen>
static inline int
munmap_adaptor(int (*munmap)(T_implPtr caddr, T_implLen sizet),
                 void* caddr, size_t sizet)
{
  return munmap ((T_implPtr) caddr, (T_implLen) sizet);
}

void *implPtr;
size_t implLen;

void
unmapImpl ()
{
  munmap_adaptor(munmap, implPtr, implLen);
}
int main()
{
  unmapImpl();
}

-- 
           Summary: [3.5 Regression]  symbol emitted instead of a
                    non_lazy_symbol
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Keywords: wrong-code, build
          Severity: critical
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org
                CC: andreast at gcc dot gnu dot org,austern at apple dot
                    com,gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-15 14:41 -------
I almost think it is caused by:
2004-03-12  Matt Austern  <austern@apple.com>
        
        * target.h (struct gcc_target): New target hook, unwind_label.
        * target-def.h (TARGET_ASM_EMIT_UNWIND_LABEL): New hook.
        * output.h (default_emit_unwind_label): New function.
        * default.h (TARGET_EXPLICIT_INSTANTIATIONS_ONE_ONLY): New macro.
        (TARGET_USES_WEAK_UNWIND_INFO): New target macro.
        (TARGET_SUPPORTS_HIDDEN): New target macro.
        * dwarf2out.c (struct dw_fde_struct): Add field for function decl
        that corresponds to this FDE.
        (FRAME_BEGIN_LABEL): Allow target to override default label.
        (output_call_frame_info): If FDEs are linknonce, then use extra
        indirection for FDE encoding, output a label for each FDE, and
        output an empty label for each function without an FDE.
        (dwarf2out_begin_prologue): Set up decl field when creating an FDE.
        * varasm.c (globalize_decl): Call ASM_MAKE_LABEL_LINKONCE for
        decls with DECL_ONE_ONLY set, if that macro is defined.
        (make_decl_one_only): Don't use DECL_COMMON if we're compiling
        for a SUPPORTS_ONE_ONLY target.
        * config/darwin-protos.h (darwin_unique_section): Declare.
        (darwin_asm_named_section): Likewise.
        (darwin_section_type_flags): Likewise.
        (darwin_non_lazy_pcrel): Likewise.
        (darwin_emit_unwind_label): Likewise.
        * config/darwin.c (machopic_finish): Get rid of tweak that
        eliminate stubs for symbols that are defined.
        (darwin_encode_section_info): Don't treat weak functions as defined.
        (darwin_make_decl_one_only): Define.
        (darwin_asm_named_section): Likewise.
        (darwin_section_type_flags): Likewise.
        (darwin_unique_section): Likewise.
        (darwin_emit_unwind_label): Likewise.
        (darwin_non_lazy_pcrel): Likewise.
        (darwin_asm_output_dwarf_delta): Difference between two labels is
        local only if both labels are local.
        * config/darwin.h (MAKE_DECL_ONE_ONLY): Define.
        (ASM_MAKE_LABEL_LINKONCE): Likewise.
        (TARGET_SUPPORTS_HIDDEN): Likewise.
        (TARGET_USES_WEAK_UNWIND_INFO): Likewise.
        (TARGET_EXPLICIT_INSTANTIATIONS_ONE_ONLY): Likewise.
        (FRAME_BEGIN_LABEL): Likewise.
        (ASM_DECLARE_OBJECT_NAME): Make references to weak symbols indirect.
        (ASM_DECLARE_FUNCTION_NAME): Likewise.
        (darwin_eh_frame_section): Give __eh_frame section the coalesced flag.
        (TARGET_ASM_UNIQUE_SECTION): Define.
        (EH_FRAME_SECTION_NAME): Define.
        (EH_FRAME_SECTION_ATTR): Likewise.
        (ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX): Likewise.
        (TARGET_ASM_NAMED_SECTION): Likewise.
        (TARGET_SECTION_TYPE_FLAGS): Likewise.
        * doc/tm.texi: Document TARGET_USES_WEAK_UNWIND_INFO,
        TARGET_EXPLICIT_INSTANTIATIONS_ONE_ONLY, 
TARGET_SUPPORTS_HIDDEN,
        TARGET_ASM_EMIT_UNWIND_LABEL.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-15 14:48 -------
Also it worked in 20040307 but does not today.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |3.5.0
      Known to work|                            |3.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-15 15:25 -------
Actually it looks more like as .combine is the stage where it is messed up:
2004-03-12  Roger Sayle  <roger@eyesopen.com>

        * combine.c (unmentioned_reg_p): New function to check whether an
        expression is a "specialization" of another, i.e. that there are
        no registers or memory references mentioned in the first that don't
        appear in the second.
        (unmentioned_reg_p_1): New helper subroutine of unmentioned_reg_p.
        (combine_instructions): Also try combining instructions using the
        REG_EQUAL note from a preceding log-linked instruction.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |roger at eyesopen dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-15 16:09 -------
Before .combine:
(call_insn 33 32 41 0 (parallel [
            (set (reg:SI 3 r3)
                (call (mem:SI (reg/v/u/f:SI 118 [ munmap ]) [0 S4 A8])
                    (const_int 32 [0x20])))
            (use (const_int 0 [0x0]))
            (clobber (scratch:SI))
        ]) 388 {*call_value_indirect_nonlocal_sysv} (insn_list 16 (insn_list 31 (insn_list 32 
(nil))))
    (expr_list:REG_DEAD (reg:SI 4 r4 [ sizet ])
        (expr_list:REG_DEAD (reg/v/u/f:SI 118 [ munmap ])
            (expr_list:REG_UNUSED (scratch:SI)
                (expr_list:REG_UNUSED (reg:SI 3 r3)
                    (nil)))))
    (expr_list (use (reg:SI 4 r4 [ sizet ]))
        (expr_list (use (reg:SI 3 r3 [ caddr ]))
            (nil))))
After .combine:
(call_insn 33 32 41 0 (parallel [
            (set (reg:SI 3 r3)
                (call (mem:SI (symbol_ref:SI ("!t__munmap") [flags 0x41] <function_decl 
0x6c5600 munmap>) [0 S4 A8])
                    (const_int 32 [0x20])))
            (use (const_int 0 [0x0]))
            (clobber (scratch:SI))
        ]) 389 {*call_value_nonlocal_sysv} (insn_list 32 (insn_list 31 (nil)))
    (expr_list:REG_UNUSED (reg:SI 3 r3)
        (expr_list:REG_UNUSED (scratch:SI)
            (expr_list:REG_DEAD (reg:SI 4 r4 [ sizet ])
                (nil))))
    (expr_list (use (reg:SI 4 r4 [ sizet ]))
        (expr_list (use (reg:SI 3 r3 [ caddr ]))
            (nil))))

Since on Darwin extern (library) calls need to go through a stub, this is wrong.
One way of fixing this would be rewrite Darwin's back-end so that the decission about the 
stubs only happen at the last moment when compiling aka when writting out the asm 
from the RTL.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-03-15 16:09:44
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[roger at eyesopen dot com]

------- Additional Comments From roger at eyesopen dot com  2004-03-15 16:17 -------
Could you also post the instruction containing the REG_EQUAL note that's
being used by combine?  I suspect that the real problem is that the symbol
held in the REG_EQUAL note is not infact "equal" to that in the SET_SRC.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-15 17:27 -------
Here is the RTL you are asking for, I still spectect that the darwin's back-end is doing 
something wrong.
(insn 14 44 15 0 (use (symbol_ref:SI ("!t__munmap") [flags 0x41] <function_decl 
0x6c5600 munmap>)) -1 (nil)
    (nil))

(insn 15 14 16 0 (set:SI (reg:SI 119)
        (plus:SI (reg:SI 31 r31)
            (high:SI (const:SI (minus:SI (symbol_ref:SI ("&L_munmap$non_lazy_ptr"))
                        (symbol_ref:SI ("<pic base>"))))))) 37 {addsi3_high} (nil)
    (nil))

(insn 16 15 17 0 (set (reg/v/u/f:SI 118 [ munmap ])
        (mem/u:SI (lo_sum:SI (reg:SI 119)
                (const:SI (minus:SI (symbol_ref:SI ("&L_munmap$non_lazy_ptr"))
                        (symbol_ref:SI ("<pic base>"))))) [0 S4 A8])) 303 {movsi_low} (insn_list 15 
(nil))
    (expr_list:REG_DEAD (reg:SI 119)
        (expr_list:REG_EQUAL (symbol_ref:SI ("!t__munmap") [flags 0x41] <function_decl 
0x6c5600 munmap>)
            (nil))))

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[roger at eyesopen dot com]

------- Additional Comments From roger at eyesopen dot com  2004-03-16 17:48 -------
Well, it looks like combine is doing "the right thing" as far as the RTL-level
transformation is concerned.  It's turning

(set (reg119) (...)) (REG_EQUAL symbol_ref (!t__unmap)))
(set (reg3) (call (mem (reg119))))

into

(set (reg3) (call (mem (symbol_ref (!t__unmap)))))

i.e. it's optimized an indirect jump into a direct jump.  I suspect that
the problem with the rs6000.md backend may be that the constraints for the
*call_value_nonlocal_sysv pattern shouldn't be accepting this replacement,
recognizing "mem (symbol_ref (!t__unmap))" as a invalid operand on darwin.

I think rs6000's use of REG_EQUAL notes is reasonable in this case, as they
can be used in other optimizations.  So I believe that the correct fix is to
inform the backend that library functions must be called indirectly via their
thunks.  I'm also surprised that this hasn't bitten folks in CSE and/or GCSE
before now.

Of course, I'm assuming that the RTL is accurately describing the intended
semantics here.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-16 22:09 -------
*** Bug 14612 has been marked as a duplicate of this bug. ***

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rwgk at yahoo dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-21 17:41 -------
Created an attachment (id=5963)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=5963&action=view)
Patch which fixes it

Mine.  Here is a patch which should workaround the problem.  The correct fix is
have the back-end only determain when it the symbol needs to be a stub at the
last momement but that is not going to be done for a while so this just rejects
the references who is not local (stubs are considered local).

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |pinskia at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-21 17:49 -------
Patch here: <http://gcc.gnu.org/ml/gcc-patches/2004-03/msg01775.html>.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch
   Last reconfirmed|2004-03-15 16:09:44         |2004-03-21 17:49:51
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-03-22 15:05 -------
Subject: Bug 14580

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	pinskia@gcc.gnu.org	2004-03-22 15:05:15

Modified files:
	gcc            : ChangeLog 

Log message:
	2004-03-22  Andrew Pinski  <pinskia@physics.uc.edu>
	
	PR target/14580
	* config/rs6000/rs6000.c (symbol_ref_operand): Reject symbols
	who are not local for Darwin PIC.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.3254&r2=2.3255



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580

[Bug target/14580] [3.5 Regression] symbol emitted instead of a non_lazy_symbol

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-22 15:08 -------
Fixed.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14580