[Bug middle-end/33662] New: [4.3 Regression] Wrong register allocation on SH

[Bug middle-end/33662] New: [4.3 Regression] Wrong register allocation on SH

[kkojima at gcc dot gnu dot org]

Several libffi testcases and gcc.c-torture/execute/pr23135.c
fail with the execution error for sh4-unknown-linux-gnu after
the patch:

r128957 | zadeck | 2007-10-02 22:10:07 +0900 (Tue, 02 Oct 2007) | 70 lines

A reduced testcase is

typedef struct A {
  unsigned long long a;
  unsigned char b;
} A;

typedef struct B {
  struct A x;
  unsigned char y;
} B;

B B_fn(struct A b0, struct B b1)
{
  struct B result;

  result.x.a = b0.a + b1.x.a;
  result.x.b = b0.b + b1.x.b + b1.y;
  result.y = b0.b + b1.x.b;

  printf("%d %d %d %d %d: %d %d %d\n", (int)b0.a, b0.b,
         (int)b1.x.a, b1.x.b, b1.y,
         (int)result.x.a, result.x.b, result.y);

  return result;
}

which is compiled to

B_fn:
        mov.l   r8,@-r15
        clrt
        mov.l   r9,@-r15
        mov.l   r10,@-r15
        mov.l   r11,@-r15
        mov.l   r12,@-r15
        mov.l   r13,@-r15
        mov     r2,r13
        mov.l   r14,@-r15
        sts.l   pr,@-r15
        add     #-28,r15
        mov     r15,r14
        mov     r14,r1
        add     #60,r1
        mov.l   @r1+,r7
        mov.l   @r1+,r8
        add     #-20,r15
        mov.b   @r1,r0
        extu.b  r6,r8

with -O2 -ml -m4 on sh-elf.  The last extu.b instruction
clobbers r8 whose content was loaded from a memory with
"mov.l  @r1+,r8".

The corresponding .lreg dump is:

(insn:HI 20 68 23 3 str3.c:12 (set (reg:DI 164 [ b1$x$a ])
        (mem/s/c:DI (post_inc:SI (reg/f:SI 178)) [5 b1.x.a+0 S8 A32])) 186
{*movdi_i} (expr_list:REG_INC (reg/f:SI 178)
        (nil)))

(insn:HI 23 20 19 3 str3.c:12 (set (reg:QI 179 [ b1.x.b ])
        (mem/s/c:QI (reg/f:SI 178) [0 b1.x.b+0 S1 A32])) 182 {movqi_i}
(expr_list:REG_DEAD (reg/f:SI 178)
        (nil)))

(insn:HI 19 23 76 3 str3.c:12 (set (reg:SI 165 [ b0$b ])
        (zero_extend:SI (reg:QI 6 r6))) 153 {*zero_extendqisi2_compact} (nil))

and the .greg dump is:

(insn:HI 20 68 23 2 str3.c:12 (set (reg:DI 7 r7 [orig:164 b1$x$a ] [164])
        (mem/s/c:DI (post_inc:SI (reg/f:SI 1 r1 [178])) [5 b1.x.a+0 S8 A32]))
186 {*movdi_i} (expr_list:REG_INC (reg/f:SI 1 r1 [178])
        (nil)))

(insn:HI 23 20 19 2 str3.c:12 (set (reg:QI 0 r0 [orig:179 b1.x.b ] [179])
        (mem/s/c:QI (reg/f:SI 1 r1 [178]) [0 b1.x.b+0 S1 A32])) 182 {movqi_i}
(nil))

(insn:HI 19 23 79 2 str3.c:12 (set (reg:SI 8 r8 [orig:165 b0$b ] [165])
        (zero_extend:SI (reg:QI 6 r6))) 153 {*zero_extendqisi2_compact} (nil))

It looks that the register allocater selects r8 for pseudo
165, but r8 is already allocated as a part of the register
pair r7..r8 which is allocated for pseudo 164.


-- 
           Summary: [4.3 Regression] Wrong register allocation on SH
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: kkojima at gcc dot gnu dot org
GCC target triplet: sh-elf


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662

[Bug middle-end/33662] [4.3 Regression] Wrong register allocation on SH

[kkojima at gcc dot gnu dot org]

------- Comment #1 from kkojima at gcc dot gnu dot org  2007-10-08 02:22 -------
I've confirmed that the patch in the trail of pr33669 fixes this too.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662

[Bug middle-end/33662] [4.3 Regression] Wrong register allocation on SH

[zadeck at naturalbridge dot com]

------- Comment #2 from zadeck at naturalbridge dot com  2007-10-08 03:53 -------


*** This bug has been marked as a duplicate of 33669 ***


-- 

zadeck at naturalbridge dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |DUPLICATE


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662

[Bug middle-end/33662] [4.3 Regression] Wrong register allocation on SH

[kkojima at gcc dot gnu dot org]

------- Comment #3 from kkojima at gcc dot gnu dot org  2007-10-10 13:28 -------
Not fixed by r129192.  I see

FAIL: gcc.c-torture/execute/pr33669.c execution,  -O1
FAIL: gcc.c-torture/execute/pr33669.c execution,  -O2
FAIL: gcc.c-torture/execute/pr33669.c execution,  -Os

on sh4-unknown-linux-gnu with r129192.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662

[Bug middle-end/33662] [4.3 Regression] Wrong register allocation on SH

[zadeck at naturalbridge dot com]

------- Comment #4 from zadeck at naturalbridge dot com  2007-10-10 13:33 -------
Subject: Re:  [4.3 Regression] Wrong register allocation
 on SH

kkojima at gcc dot gnu dot org wrote:
> ------- Comment #3 from kkojima at gcc dot gnu dot org  2007-10-10 13:28 -------
> Not fixed by r129192.  I see
>
> FAIL: gcc.c-torture/execute/pr33669.c execution,  -O1
> FAIL: gcc.c-torture/execute/pr33669.c execution,  -O2
> FAIL: gcc.c-torture/execute/pr33669.c execution,  -Os
>
> on sh4-unknown-linux-gnu with r129192.
>
>
>   
i am so embarrassed.

kenny


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662

[Bug middle-end/33662] [4.3 Regression] Wrong register allocation on SH

[spark at gcc dot gnu dot org]

------- Comment #5 from spark at gcc dot gnu dot org  2007-10-11 00:51 -------
What configure option should I use to enable -m1 option support ?
Looks like plain-vanilla configured sh4-elf compiler doesn't support -m1 flag.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662

[Bug middle-end/33662] [4.3 Regression] Wrong register allocation on SH

[spark at gcc dot gnu dot org]

------- Comment #6 from spark at gcc dot gnu dot org  2007-10-11 03:07 -------
Never mind. I should pay attention to 1 vs l :( 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662

[Bug middle-end/33662] [4.3 Regression] Wrong register allocation on SH

[zadeck at naturalbridge dot com]

------- Comment #7 from zadeck at naturalbridge dot com  2007-10-11 21:50 -------
kazumoto, 

there was a set of miscommunications associated with the final patch for
pr33669.

hj had checked in an earlier version of the patch and that testcase and i asked
him to revert it because there were issues with it.  He only reverted the code
and left the testcase in.  You tested against version 129192 and i checked in
the corrected patch as 129193.

given that, pr33669.c should have failed.  seongbae has verified that pr33669.c
and the testcase here no longer fails on the current truck with sh-elf.

I am going to assume that this is closed unless you find some other issue. 
Sorry for the mess up.

Kenny


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662

[Bug middle-end/33662] [4.3 Regression] Wrong register allocation on SH

[kkojima at gcc dot gnu dot org]

------- Comment #8 from kkojima at gcc dot gnu dot org  2007-10-12 02:06 -------
> You tested against version 129192 and i checked in the corrected patch as 129193.

Oh, my tester had fallen into that narrow pitfall :-)
I've confirmed that the errors go away with the current trunk.
Thanks both of you and seongbae!


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33662