[Bug target/26255] New: [3.3 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[Bug target/26255] New: [3.3 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[tausq at debian dot org]

Ref: http://bugs.debian.org/352529

The attached testcase, reduced from the Debian BTS bug report, causes an ICE

bug.c: In function 'foo':
bug.c:31: error: insn does not satisfy its constraints:
(insn 156 35 37 0 (set (reg:HI 70 %fr23 [122])
        (reg:HI 20 %r20)) 53 {*pa.md:2926} (nil)
    (nil))
bug.c:31: internal compiler error: in reload_cse_simplify_operands, at
postreload.c:391
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.
For Debian GNU/Linux specific bug reporting instructions,
see <URL:file:///usr/share/doc/gcc-4.0/README.Bugs>.

------------ bug.c -------------
typedef short int16_t;

int round_sample(int *sum);

#define MULS(ra, rb) ((ra) * (rb))

#define SUM8(sum, op, w, p) \
{                                               \
    sum op MULS((w)[0 * 64], p[0 * 64]);\
    sum op MULS((w)[1 * 64], p[1 * 64]);\
    sum op MULS((w)[2 * 64], p[2 * 64]);\
    sum op MULS((w)[3 * 64], p[3 * 64]);\
    sum op MULS((w)[4 * 64], p[4 * 64]);\
    sum op MULS((w)[5 * 64], p[5 * 64]);\
    sum op MULS((w)[6 * 64], p[6 * 64]);\
    sum op MULS((w)[7 * 64], p[7 * 64]);\
}

void foo(int *dither_state, int *samples)
{
    int16_t *synth_buf;
    const int16_t *w, *p;
    int sum;

    sum = *dither_state;
    p = synth_buf + 16;
    SUM8(sum, +=, w, p);
    p = synth_buf + 48;
    SUM8(sum, -=, w + 32, p);
    *samples = round_sample(&sum);
}
------------ bug.c -------------
$ cc -O2 -c bug.c
bug.c: In function 'foo':
bug.c:31: error: insn does not satisfy its constraints:
(insn 156 35 37 0 (set (reg:HI 70 %fr23 [122])
        (reg:HI 20 %r20)) 53 {*pa.md:2926} (nil)
    (nil))
bug.c:31: internal compiler error: in reload_cse_simplify_operands, at
postreload.c:391
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.
For Debian GNU/Linux specific bug reporting instructions,
see <URL:file:///usr/share/doc/gcc-4.0/README.Bugs>.

Problem only happens with -O2.


-- 
           Summary: [3.3 regression] internal compiler error: in
                    reload_cse_simplify_operands, at postreload.c:391
           Product: gcc
           Version: 4.0.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: tausq at debian dot org
 GCC build triplet: hppa-linux
  GCC host triplet: hppa-linux
GCC target triplet: hppa-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.0.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[tausq at debian dot org]

------- Comment #1 from tausq at debian dot org  2006-02-13 05:13 -------
I forgot to mention that this bug does not happen with the 20060104 snapshot of
gcc-4.0_branch (Debian 4.0.2-6), only with 20060128 (Debian 4.0.2-8)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2006-02-13 05:18 -------
During that time there was a couple of HPPA patches:
2006-01-16  John David Anglin  <dave.anglin@nrc-cnrc.gc.ca>

        * pa.md: Disparage copies between general and floating-point registers
        in 32-bit move patterns.

2006-01-10  John David Anglin  <dave.anglin@nrc-cnrc.gc.ca>

        PR target/20754
        * pa.md: Create separate 32 and 64-bit move patterns for SI, DI, SF
        and DF modes.  Add alternatives to copy between general and floating
        point registers to the 32-bit patterns.
        * pa-64.h (SECONDARY_MEMORY_NEEDED_RTX): Delete undefine.
        * pa.h (SECONDARY_MEMORY_NEEDED_RTX): Delete define.
        (SECONDARY_MEMORY_NEEDED): Secondary memory is only needed when
        generating 64-bit code.
        * pa.c (output_move_double): Handle copies between general and
        floating registers.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[danglin at gcc dot gnu dot org]

------- Comment #3 from danglin at gcc dot gnu dot org  2006-02-13 19:08 -------
Regarding comment #2, while it's possible that removing
the SECONDARY_MEMORY_NEEDED_RTX define has had an effect,
we have never had support in pa.md to load an HImode value
into a floating-point register.  The other changes in the
patches mentioned don't affect HImode.

It seems as reload isn't respecting HARD_REGNO_MODE_OK
(see VALID_FP_MODE_P in pa32-regs.h).  The reason for
excluding HImode (and QImode) values from the FP
registers is you can't do anything with them, and copying
to/from the general registers has to be done thorugh memory.
Thus, it's better to use the stack.


-- 

danglin at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |danglin at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[danglin at gcc dot gnu dot org]

------- Comment #4 from danglin at gcc dot gnu dot org  2006-02-13 23:36 -------
Here's what 3.4 does for the testcase:

(insn 38 204 205 0 (set (reg:HI 19 %r19)
        (mem:HI (plus:SI (reg/v/f:SI 3 %r3 [orig:97 w ] [97])
                (const_int 256 [0x100])) [4 S2 A16])) 52 {*pa.md:2878} (nil)
    (expr_list:REG_EQUIV (mem:HI (plus:SI (reg/v/f:SI 3 %r3 [orig:97 w ] [97])
                (const_int 256 [0x100])) [4 S2 A16])
        (nil)))

(insn 205 38 206 0 (set (mem:SI (plus:SI (reg/f:SI 30 %r30)
                (const_int -16 [0xfffffff0])) [0 S4 A32])
        (reg:SI 19 %r19)) 37 {*pa.md:2265} (nil)
    (nil))

(insn 206 205 23 0 (set (reg:SI 78 %fr27)
        (mem:SI (plus:SI (reg/f:SI 30 %r30)
                (const_int -16 [0xfffffff0])) [0 S4 A32])) 37 {*pa.md:2265}
(nil
)
    (nil))

...

(insn 217 74 218 0 (set (mem:SI (plus:SI (reg/f:SI 30 %r30)
                (const_int -16 [0xfffffff0])) [0 S4 A32])
        (reg:SI 78 %fr27)) 37 {*pa.md:2265} (nil)
    (nil))

(insn 218 217 39 0 (set (reg:SI 19 %r19)
        (mem:SI (plus:SI (reg/f:SI 30 %r30)
                (const_int -16 [0xfffffff0])) [0 S4 A32])) 37 {*pa.md:2265}
(nil
)
    (nil))

(insn 39 218 219 0 (set (reg:SI 19 %r19)
        (sign_extend:SI (reg:HI 19 %r19))) 83 {extendhisi2} (insn_list 38
(nil))
    (nil))

(insn 219 39 220 0 (set (mem:SI (plus:SI (reg/f:SI 30 %r30)
                (const_int -16 [0xfffffff0])) [0 S4 A32])
        (reg:SI 19 %r19)) 37 {*pa.md:2265} (nil)
    (nil))

(insn 220 219 221 0 (set (reg:SI 78 %fr27 [118])
        (mem:SI (plus:SI (reg/f:SI 30 %r30)
                (const_int -16 [0xfffffff0])) [0 S4 A32])) 37 {*pa.md:2265}
(nil
)
    (nil))

...

(insn 42 222 83 0 (set (reg:DI 78 %fr27 [123])
        (mult:DI (zero_extend:DI (reg:SI 78 %fr27 [118]))
            (zero_extend:DI (reg:SI 69 %fr22R)))) 112 {umulsidi3} (insn_list 41
(insn_list 39 (nil)))
    (nil))

I guess the bug was instroduced by the removal of the
SECONDARY_MEMORY_NEEDED_RTX
define.  Without it reload isn't smart enough to use SImode for the
general to floating-point register copy.  Probably, adding alternatives to
do the copy will fix the problem, and just maybe if the alternatives are
disparaged reload won't use them.  Yuck.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[danglin at gcc dot gnu dot org]

------- Comment #5 from danglin at gcc dot gnu dot org  2006-02-15 02:54 -------
Subject: Bug 26255

Author: danglin
Date: Wed Feb 15 02:54:08 2006
New Revision: 111014

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=111014
Log:
        PR target/26255
        * pa.md: Create separate HI and QI move patterns for 32-bit and 64-bit
        with hardware float support, and software float support.  Add fcpy
        alternative to hardware patterns.  Add alternatives to copy between
        general and floating-point registers to the 32-bit pattern.
        * pa32-regs.h (VALID_FP_MODE_P): Allow QImode and HImode.
        * pa64-regs.h (VALID_FP_MODE_P): Likewise.


Modified:
    branches/gcc-4_1-branch/gcc/ChangeLog
    branches/gcc-4_1-branch/gcc/config/pa/pa.md
    branches/gcc-4_1-branch/gcc/config/pa/pa32-regs.h
    branches/gcc-4_1-branch/gcc/config/pa/pa64-regs.h


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[tausq at debian dot org]

------- Comment #6 from tausq at debian dot org  2006-02-15 05:50 -------
Will this go into the 4.0 branch too?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #7 from dave at hiauly1 dot hia dot nrc dot ca  2006-02-15 14:25 -------
Subject: Re:  [4.0 regression] internal compiler error: in
reload_cse_simplify_operands, at postreload.c:391

> ------- Comment #6 from tausq at debian dot org  2006-02-15 05:50 -------
> Will this go into the 4.0 branch too?

Yes.  However, I want to complete testing on on both hpux and linux.

The biggest risk in this change is that it might require support for
reloads that weren't needed before since we're now explicitly allowing
QI and HI mode values into the FP registers.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[danglin at gcc dot gnu dot org]

------- Comment #8 from danglin at gcc dot gnu dot org  2006-02-16 13:08 -------
Subject: Bug 26255

Author: danglin
Date: Thu Feb 16 13:08:01 2006
New Revision: 111136

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=111136
Log:
        PR target/26255
        * pa.md: Create separate HI and QI move patterns for 32-bit and 64-bit
        with hardware float support, and software float support.  Add fcpy
        alternative to hardware patterns.  Add alternatives to copy between
        general and floating-point registers to the 32-bit pattern.
        * pa32-regs.h (VALID_FP_MODE_P): Allow QImode and HImode.
        * pa64-regs.h (VALID_FP_MODE_P): Likewise.


Modified:
    branches/gcc-4_0-branch/gcc/ChangeLog
    branches/gcc-4_0-branch/gcc/config/pa/pa.md
    branches/gcc-4_0-branch/gcc/config/pa/pa32-regs.h
    branches/gcc-4_0-branch/gcc/config/pa/pa64-regs.h


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[danglin at gcc dot gnu dot org]

------- Comment #9 from danglin at gcc dot gnu dot org  2006-02-17 22:10 -------
Subject: Bug 26255

Author: danglin
Date: Fri Feb 17 22:09:58 2006
New Revision: 111213

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=111213
Log:
        PR target/26255
        * gcc.c-torture/compile/pr26255.c: New test case.


Added:
    trunk/gcc/testsuite/gcc.c-torture/compile/pr26255.c
Modified:
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[danglin at gcc dot gnu dot org]

------- Comment #10 from danglin at gcc dot gnu dot org  2006-02-17 22:19 -------
Subject: Bug 26255

Author: danglin
Date: Fri Feb 17 22:19:05 2006
New Revision: 111214

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=111214
Log:
        PR target/26255
        * pa.md: Create separate HI and QI move patterns for 32-bit and 64-bit
        with hardware float support, and software float support.  Add fcpy
        alternative to hardware patterns.  Add alternatives to copy between
        general and floating-point registers to the 32-bit pattern.
        * pa.c (pa_secondary_reload): Don't abort if reload tries to find a
        secondary reload to load a QI or HI mode constant into a floating
        point register. 
        * pa32-regs.h (VALID_FP_MODE_P): Allow QImode and HImode.
        * pa64-regs.h (VALID_FP_MODE_P): Likewise.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/pa/pa.c
    trunk/gcc/config/pa/pa.md
    trunk/gcc/config/pa/pa32-regs.h
    trunk/gcc/config/pa/pa64-regs.h


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255

[Bug target/26255] [4.0 regression] internal compiler error: in reload_cse_simplify_operands, at postreload.c:391

[danglin at gcc dot gnu dot org]

------- Comment #11 from danglin at gcc dot gnu dot org  2006-02-17 22:21 -------
Fixed by patches.


-- 

danglin at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26255