[Bug middle-end/30784] New: ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[Bug middle-end/30784] New: ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[fxcoudert at gcc dot gnu dot org]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1726 bytes --]

$ cat a.c
void foo(void)
{
  static int irows[50000];
  int i;

  for (i = 1; i < 1000; i++)
    irows[i-1] = i;
}

$ gcc -c -O1 -march=athlon-xp -ftree-vectorize a.c
a.c: In function ‘foo’:
a.c:8: internal compiler error: in expand_simple_binop, at optabs.c:1192
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

$ gcc -v                      
Using built-in specs.
Target: i386-pc-linux-gnu
Configured with: /home/fxcoudert/gfortran_nightbuild/trunk/configure
--prefix=/home/fxcoudert/gfortran_nightbuild/irun-20070209
--enable-languages=c,fortran --host=i386-pc-linux-gnu --enable-checking=release
--with-gmp=/home/fxcoudert/gfortran_nightbuild/software
Thread model: posix
gcc version 4.3.0 20070209 (experimental)


The original bug report (http://gcc.gnu.org/ml/fortran/2007-02/msg00288.html)
was about the follow Fortran code:

      subroutine gaussj2_cvb(n)
      dimension irows(n)
      do 100 i=1,n
100   irows(i)=i
      end

When someone fixes this, can you verify that the ICE is also gone on the
Fortran code (same optimization options)?


-- 
           Summary: ICE on loop vectorization (-O1 -march=athlon-xp -ftree-
                    vectorize)
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: fxcoudert at gcc dot gnu dot org
 GCC build triplet: i386-pc-linux-gnu
  GCC host triplet: i386-pc-linux-gnu
GCC target triplet: i386-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug middle-end/30784] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[fxcoudert at gcc dot gnu dot org]

-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2007-02-13 10:27:36
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2007-02-13 10:33 -------
This works fine with me on powerpc-darwin with -O1 -ftree-vectorize -maltivec 
plus the loop was vectorized also.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |target


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2007-02-13 12:48 -------
I get

./cc1 -quiet -O -march=athlon-xp -m32 -ftree-vectorize x.i        
x.i: In function 'foo':
x.i:2: error: invalid reference prefix
{4, 4, 4, 4}

x.i:2: error: invalid reference prefix
{4, 4, 4, 4}

x.i:2: error: invalid reference prefix
{4, 4, 4, 4}

x.i:2: error: invalid reference prefix
{4, 4, 4, 4}

x.i:2: internal compiler error: verify_stmts failed
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

After vectorization:

<bb 2>:
  stmp_var_.29_22 = 2;
  stmp_var_.30_23 = stmp_var_.29_22 + 1;
  stmp_var_.31_24 = stmp_var_.30_23 + 1;
  vect_cst_.32_25 = {1, stmp_var_.29_22, stmp_var_.30_23, stmp_var_.31_24};
  vect_cst_.33_26 = {4, 4, 4, 4};
  vect_pirows.39_29 = (vector int *) &irows;
  vect_pirows.35_30 = vect_pirows.39_29;

  # ivtmp.41_33 = PHI <ivtmp.41_34(7), 0(2)>
  # ivtmp.40_31 = PHI <ivtmp.40_32(7), vect_pirows.35_30(2)>
  # vect_vec_iv_.34_27 = PHI <vect_vec_iv_.34_28(7), vect_cst_.32_25(2)>
  # ivtmp.25_1 = PHI <ivtmp.25_5(7), 999(2)>
  # i_10 = PHI <i_4(7), 1(2)>
<L0>:;
  D.1842_3 = i_10 + -1;
  vect_vec_iv_.34_28 = vect_vec_iv_.34_27 + vect_cst_.33_26;
  *ivtmp.40_31 = vect_vec_iv_.34_27;
  i_4 = i_10 + 1;
  ivtmp.25_5 = ivtmp.25_1 - 1;
  ivtmp.40_32 = ivtmp.40_31 + 16B;
  ivtmp.41_34 = ivtmp.41_33 + 1;
  if (ivtmp.41_34 < 249) goto <L5>; else goto <L10>;

(looks good)

but then veclower does

  # ivtmp.41_33 = PHI <ivtmp.41_34(7), 0(2)>
  # ivtmp.40_31 = PHI <ivtmp.40_32(7), vect_pirows.35_30(2)>
  # vect_vec_iv_.34_27 = PHI <vect_vec_iv_.34_28(7), vect_cst_.32_25(2)>
  # ivtmp.25_1 = PHI <ivtmp.25_5(7), 999(2)>
  # i_10 = PHI <i_4(7), 1(2)> 
<L0>:;
  D.1842_3 = i_10 + -1;
  D.1899_20 = BIT_FIELD_REF <vect_vec_iv_.34_27, 32, 0>;
  D.1900_19 = BIT_FIELD_REF <vect_cst_.33_26, 32, 0>;
  D.1901_17 = D.1899_20 + D.1900_19;
  D.1902_35 = BIT_FIELD_REF <vect_vec_iv_.34_27, 32, 32>;
  D.1903_36 = BIT_FIELD_REF <vect_cst_.33_26, 32, 32>;
  D.1904_37 = D.1902_35 + D.1903_36;
  D.1905_38 = BIT_FIELD_REF <vect_vec_iv_.34_27, 32, 64>;
  D.1906_39 = BIT_FIELD_REF <vect_cst_.33_26, 32, 64>;
  D.1907_40 = D.1905_38 + D.1906_39;
  D.1908_41 = BIT_FIELD_REF <vect_vec_iv_.34_27, 32, 96>;
  D.1909_42 = BIT_FIELD_REF <vect_cst_.33_26, 32, 96>;
  D.1910_43 = D.1908_41 + D.1909_42;
  vect_vec_iv_.34_28 = {D.1901_17, D.1904_37, D.1907_40, D.1910_43};
  *ivtmp.40_31 = vect_vec_iv_.34_27;
  i_4 = i_10 + 1;
  ivtmp.25_5 = ivtmp.25_1 - 1;
  ivtmp.40_32 = ivtmp.40_31 + 16B;
  ivtmp.41_34 = ivtmp.41_33 + 1;
  if (ivtmp.41_34 < 249) goto <L5>; else goto <L10>;

but then DOM const-props into the BIT_FIELD_REFs which confuses us:

  # ivtmp.44_8 = PHI <ivtmp.44_45(3), 0(2)>
  # vect_vec_iv_.34_27 = PHI <vect_vec_iv_.34_28(3), vect_cst_.32_25(2)>
<L0>:;
  D.1899_20 = BIT_FIELD_REF <vect_vec_iv_.34_27, 32, 0>;
  D.1900_19 = BIT_FIELD_REF <{4, 4, 4, 4}, 32, 0>;
  D.1901_17 = D.1899_20 + D.1900_19;
  D.1902_35 = BIT_FIELD_REF <vect_vec_iv_.34_27, 32, 32>;
  D.1903_36 = BIT_FIELD_REF <{4, 4, 4, 4}, 32, 32>;
  D.1904_37 = D.1902_35 + D.1903_36;
  D.1905_38 = BIT_FIELD_REF <vect_vec_iv_.34_27, 32, 64>;
  D.1906_39 = BIT_FIELD_REF <{4, 4, 4, 4}, 32, 64>;
  D.1907_40 = D.1905_38 + D.1906_39;
  D.1908_41 = BIT_FIELD_REF <vect_vec_iv_.34_27, 32, 96>;
  D.1909_42 = BIT_FIELD_REF <{4, 4, 4, 4}, 32, 96>;
  D.1910_43 = D.1908_41 + D.1909_42;
  vect_vec_iv_.34_28 = {D.1901_17, D.1904_37, D.1907_40, D.1910_43};
  MEM[symbol: irows, index: ivtmp.44_8] = vect_vec_iv_.34_27;
  ivtmp.44_45 = ivtmp.44_8 + 16;
  if (ivtmp.44_45 != 3984) goto <L0>; else goto <L15>;

so we fail to fold BIT_FIELD_REF <{4, 4, 4, 4}, 32, 96> to a constant for
example.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com,
                   |                            |rguenth at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[fxcoudert at gcc dot gnu dot org]

------- Comment #3 from fxcoudert at gcc dot gnu dot org  2007-03-14 10:55 -------
(In reply to comment #2)
> ./cc1 -quiet -O -march=athlon-xp -m32 -ftree-vectorize x.i        
> x.i: In function 'foo':
> x.i:2: error: invalid reference prefix
> {4, 4, 4, 4}
> 
> x.i:2: error: invalid reference prefix
> {4, 4, 4, 4}
> 
> x.i:2: error: invalid reference prefix
> {4, 4, 4, 4}
> 
> x.i:2: error: invalid reference prefix
> {4, 4, 4, 4}
> 
> x.i:2: internal compiler error: verify_stmts failed

It's still present on trunk, and it's a regression from the 4.2 branch.


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |4.3.0
      Known to work|                            |4.2.0 4.1.2
   Last reconfirmed|2007-02-13 10:27:36         |2007-03-14 10:55:20
               date|                            |
            Summary|ICE on loop vectorization (-|[4.3 regression] ICE on loop
                   |O1 -march=athlon-xp -ftree- |vectorization (-O1 -
                   |vectorize)                  |march=athlon-xp -ftree-
                   |                            |vectorize)
   Target Milestone|---                         |4.3.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[dorit at il dot ibm dot com]

------- Comment #4 from dorit at il dot ibm dot com  2007-03-14 12:13 -------
I also saw this on powerpc64, on a different testcase (vectorizing longs with
-m64). seems like constant propagation during dom3 propagates the vector
initializer into a BIT_FIELD_EXPR, which results in invalid gimple?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[dorit at il dot ibm dot com]

------- Comment #5 from dorit at il dot ibm dot com  2007-03-14 12:29 -------
this is the testcase I have ICE-ing on powerpc64-yellowdog, when compiled with
-ftree-vectorize -maltivec -m64 -O2:

long stack_vars_sorted[32];
void partition_stack_vars (long stack_vars_num)
{
  long si, n = stack_vars_num;
  for (si = 0; si < n; ++si)
    stack_vars_sorted[si] = si;
}

(extracted from cfgexpand.c which ICEs during bootstrap with vectorization)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[pinskia at gcc dot gnu dot org]

------- Comment #6 from pinskia at gcc dot gnu dot org  2007-03-14 19:01 -------
I think this was introduced by me, by making vector based CONSTRUCTOR with
CONSTANT set invariant.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[dorit at il dot ibm dot com]

------- Comment #7 from dorit at il dot ibm dot com  2007-03-24 08:00 -------
patch: http://gcc.gnu.org/ml/gcc/2007-03/msg00918.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[patchapp at dberlin dot org]

------- Comment #8 from patchapp at dberlin dot org  2007-03-24 17:10 -------
Subject: Bug number PR30784

A patch for this bug has been added to the patch tracker.
The mailing list url for the patch is
http://gcc.gnu.org/ml/gcc-patches/2007-03/msg01607.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug target/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[dorit at gcc dot gnu dot org]

------- Comment #9 from dorit at gcc dot gnu dot org  2007-03-25 13:08 -------
Subject: Bug 30784

Author: dorit
Date: Sun Mar 25 12:08:29 2007
New Revision: 123197

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=123197
Log:
        PR tree-optimization/30784
        * fold-const.c (fold_ternary): Handle CONSTRUCTOR in case
        BIT_FIELD_REF.


Added:
    trunk/gcc/testsuite/gcc.dg/vect/pr30784.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/fold-const.c
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug middle-end/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[pinskia at gcc dot gnu dot org]

------- Comment #10 from pinskia at gcc dot gnu dot org  2007-03-25 19:40 -------
Fixed.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
          Component|target                      |middle-end
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784

[Bug middle-end/30784] [4.3 regression] ICE on loop vectorization (-O1 -march=athlon-xp -ftree-vectorize)

[pinskia at gcc dot gnu dot org]

------- Comment #11 from pinskia at gcc dot gnu dot org  2007-06-18 06:16 -------
*** Bug 30958 has been marked as a duplicate of this bug. ***


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dcb314 at hotmail dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30784