[Bug target/16482] New: gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[Bug target/16482] New: gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[kkojima at gcc dot gnu dot org]

The following part of gcc.c-torture/unsorted/SFset.c:

int glob1;
 
adrreg0limm1_set (float *p0)
{
  p0[10000000] = (float) glob1;
}

causes a spill failure for sh-elf on 3.5 with -m4 -O2:

/exp/ldroot/dodes/LOCAL/gcc/gcc/testsuite/gcc.c-torture/unsorted/SFset.c: In
function `adrreg0limm1_set':
/exp/ldroot/dodes/LOCAL/gcc/gcc/testsuite/gcc.c-torture/unsorted/SFset.c:135:
error: unable to find a register to spill in class `R0_REGS'
/exp/ldroot/dodes/LOCAL/gcc/gcc/testsuite/gcc.c-torture/unsorted/SFset.c:135:
error: this is the insn:
(insn:HI 25 44 40 0 (parallel [
            (set (mem:SF (plus:SI (reg:SI 6 r6 [ p0 ])
                        (reg:SI 2 r2 [171])) [2 S4 A32])
                (reg:SF 65 fr1 [173]))
            (use (reg/v:PSI 151 ))
            (clobber (scratch:SI))
        ]) 150 {movsf_ie} (insn_list 24 (insn_list 22 (insn_list 42 (nil))))
    (expr_list:REG_DEAD (reg/v:PSI 151 )
        (expr_list:REG_DEAD (reg:SF 65 fr1 [173])
            (expr_list:REG_DEAD (reg:SI 6 r6 [ p0 ])
                (expr_list:REG_DEAD (reg:SI 2 r2 [171])
                    (expr_list:REG_UNUSED (scratch:SI)
                        (nil)))))))

This test passed on 3.3 and 3.4 successfully.

-- 
           Summary: gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: kkojima at gcc dot gnu dot org
                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: sh-unknown-elf


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[kkojima at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |3.5.0
      Known to work|                            |3.3.4 3.4.1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [3.5 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|gcc.c-                      |[3.5 Regression] gcc.c-
                   |torture/unsorted/SFset.c    |torture/unsorted/SFset.c
                   |fails with "-O2 -m4"        |fails with "-O2 -m4"
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-10-13 02:19 -------
Confirmed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
      Known to fail|4.0                         |4.0.0
   Last reconfirmed|0000-00-00 00:00:00         |2004-10-13 02:19:55
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2004-10-15 13:35 -------
Your examples return an undefined value, which raises some extra issues, but
that is really a red herring;  I see the same failure with:

int glob1;

int
adrreg01limm1_set (float *p0)
{
  int r = 5;
  p0[10] = (float) glob1;
  return r;
}

compiled with -m4 -O2 -fPIC .

In particular, sched1 moves instruction past the return value copy and use.

after regmove:
(insn:HI 27 24 33 0 (set (reg/i:SI 0 r0 [ <result> ])
        (const_int 5 [0x5])) 129 {movsi_ie} (nil)
    (nil))

(insn:HI 33 27 0 0 (use (reg/i:SI 0 r0 [ <result> ])) -1 (insn_list:REG_DEP_TRUE
27 (nil))
    (nil))
;; End of basic block 0, registers live:
 0 [r0] 12 [r12] 14 [r14] 15 [r15] 145 [ap] 151 []

after sched1:
(insn:HI 27 14 47 1 (set (reg/i:SI 0 r0 [ <result> ])
        (const_int 5 [0x5])) 129 {movsi_ie} (nil)
    (nil))

(insn 47 27 40 1 (set (reg:SI 175)
        (plus:SI (reg:SI 174)
            (reg:SI 12 r12))) 23 {*addsi3_compact} (insn_list:REG_DEP_TRUE 46 (nil))
    (expr_list:REG_DEAD (reg:SI 174)
        (nil)))

(insn 40 47 48 1 (set (reg:SI 170)
        (const:SI (unspec [
                    (symbol_ref:SI ("__fpscr_values"))
                ] 7))) 129 {movsi_ie} (nil)
    (nil))

(insn 48 40 33 1 (set (reg:SI 173)
        (mem/u:SI (reg:SI 175) [0 S4 A32])) 129 {movsi_ie}
(insn_list:REG_DEP_TRUE 47 (nil))
    (expr_list:REG_DEAD (reg:SI 175)
        (expr_list:REG_EQUAL (symbol_ref:SI ("__fpscr_values"))
            (nil))))

(insn:HI 33 48 16 1 (use (reg/i:SI 0 r0 [ <result> ])) -1
(insn_list:REG_DEP_TRUE 27 (nil))
    (nil))

(insn:HI 16 33 41 1 (set (reg/f:SI 162)
        (mem/u:SI (plus:SI (reg:SI 12 r12)
                (reg/f:SI 163)) [0 S4 A32])) 129 {movsi_ie}
(insn_list:REG_DEP_TRUE 14 (nil))
    (expr_list:REG_DEAD (reg/f:SI 163)
        (expr_list:REG_EQUAL (symbol_ref:SI ("glob1") <var_decl 0x401e415c glob1>)
            (nil))))

The return value use schould remain at the end of the basic block, and the
preceding return value copy should be in the same scheduling group.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amylaar at gcc dot gnu dot
                   |                            |org
   Last reconfirmed|2004-10-13 02:19:55         |2004-10-15 13:35:41
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[kkojima at gcc dot gnu dot org]

------- Additional Comments From kkojima at gcc dot gnu dot org  2004-10-15 14:23 -------
I asked a question about this issue:

http://gcc.gnu.org/ml/gcc/2004-10/msg00176.html

and Toshi replied:

http://gcc.gnu.org/ml/gcc/2004-10/msg00595.html

Now I'm trying a patch in

http://gcc.gnu.org/ml/gcc/2004-10/msg00600.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2004-10-15 15:49 -------
Looking closer, I see that sched-deps.c already has code to keep branches, calls
and USEs at the end of the last basic block.
The problem is that lcm.c:optimize_mode_switching has put some
instructions after the USE before sched1 is run.

In this code:
    /* Split the edge from the entry block and the fallthrough edge to the
       exit block, so that we can note that there NORMAL_MODE is supplied /
       required.  */
    edge eg;
    edge_iterator ei;
    post_entry = split_edge (EDGE_SUCC (ENTRY_BLOCK_PTR, 0));
    /* The only non-call predecessor at this stage is a block with a
       fallthrough edge; there can be at most one, but there could be
       none at all, e.g. when exit is called.  */
    pre_exit = 0;
    FOR_EACH_EDGE (eg, ei, EXIT_BLOCK_PTR->preds)
      if (eg->flags & EDGE_FALLTHRU)
        {
          regset live_at_end = eg->src->global_live_at_end;

          gcc_assert (!pre_exit);
          pre_exit = split_edge (eg);
          COPY_REG_SET (pre_exit->global_live_at_start, live_at_end);
          COPY_REG_SET (pre_exit->global_live_at_end, live_at_end);
          break;
        }
  }

We create a new block at the end to put the final mode switch into.
If the fall-through predecessor of the exit block is the only predecessor
of the exit block, and it ends with a USE, then instead of split_edge
we should use split_block, to place the USE and the preceding copy from
the result to the return value register into the new block.

I would write a patch for this, but the copyright assignment for STM is not
finished yet.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[naveen dot sharma at noida dot hcltech dot com]

------- Additional Comments From naveen dot sharma at noida dot hcltech dot com  2004-10-18 12:03 -------
The patch submitted by Kaz looks OK to me. 
There are some heuristics in sh.c that were written to disable 
ready queue reordering when R0 is under pressure. But (mem (pseudo) (pseudo)) 
is an indirect implicit reference to R0.
I am wondering why this was missed? Was R0 a fixed_reg sometime during pre-3.3 
releases, so as to generate (mem (r0) (pesudo)) instead of (mem (pseudo) 
(pseudo) form?

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |naveens at noida dot hcltech
                   |                            |dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[joern dot rennecke at superh-support dot com]

------- Additional Comments From joern dot rennecke at superh-support dot com  2004-10-18 15:16 -------
Subject: Re:  [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

> The patch submitted by Kaz looks OK to me.

The bug is in lcm.c, so it should be fixed there.

> I am wondering why this was missed? Was R0 a
> fixed_reg sometime during pre-3.3 
> releases, so as to generate (mem (r0) (pesudo))
> instead of (mem (pseudo) 
> (pseudo) form?

The first scheduler has has long been disabled for
SH.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[naveen dot sharma at noida dot hcltech dot com]

------- Additional Comments From naveen dot sharma at noida dot hcltech dot com  2004-10-19 05:57 -------
(In reply to comment #6)
> > I am wondering why this was missed? Was R0 a
> > fixed_reg sometime during pre-3.3 
> > releases, so as to generate (mem (r0) (pesudo))
> > instead of (mem (pseudo) 
> > (pseudo) form?
> The first scheduler has has long been disabled for
> SH.

Thats OK, but the intent of my question was to ask
if internal RTL form has changed?

I will try to confirm your analysis for bug to be in 
lcm.c.

Thanks,
Naveen Sharma.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[sanjivkumargupta at yahoo dot com]

------- Additional Comments From sanjivkumargupta at yahoo dot com  2004-10-19 06:06 -------
Anyways, Kaz's method of determining pressure on R0 looks
more correct than what I had written. So, I think we
should keep his patch besides waiting for a fix in lcm.c

Thanks
--Sanjiv

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[pinskia at gcc dot gnu dot org]

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[joern dot rennecke at st dot com]

------- Additional Comments From joern dot rennecke at st dot com  2004-12-01 18:37 -------
Subject: Re:  [4.0 Regression] gcc.c-torture/unsorted/SFset.c
 fails with "-O2 -m4"

This patch is still waiting for review: 
http://gcc.gnu.org/ml/gcc-patches/2004-11/msg01477.html



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-01-05 14:41 -------
Subject: Bug 16482

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	kkojima@gcc.gnu.org	2005-01-05 14:41:31

Modified files:
	gcc            : ChangeLog lcm.c 

Log message:
	PR target/16482
	* lcm.c (create_pre_exit): New.
	(optimize_mode_switching): In MODE_ENTRY / MODE_EXIT case, set
	ENTRY_EXIT_EXTRA to 3.  Use create_pre_exit.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7031&r2=2.7032
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/lcm.c.diff?cvsroot=gcc&r1=1.69&r2=1.70



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482

[Bug target/16482] [4.0 Regression] gcc.c-torture/unsorted/SFset.c fails with "-O2 -m4"

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-05 14:50 -------
Fixed.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16482