[Bug target/101737] New: SH4 -Os causes internal compiler error when building pixman

[Bug target/101737] New: SH4 -Os causes internal compiler error when building pixman

[thomas.petazzoni@free-electrons.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

            Bug ID: 101737
           Summary: SH4 -Os causes internal compiler error when building
                    pixman
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: thomas.petazzoni@free-electrons.com
  Target Milestone: ---

Created attachment 51247
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51247&action=edit
source file that exhibits the issue

The attached source file test.c, which comes from the pixman open-source
project, when compiled for SuperH 4 using gcc 9.3.0 causes an internal compiler
error when built at -Os. Building at -O0 or -O2 doesn't cause any problem.

$ sh4-linux-gcc -c -Os test.c 
during RTL pass: split1
pixman-fast-path.c: In function
‘fast_composite_scaled_nearest_8888_565_normal_OVER’:
pixman-fast-path.c:1204:1: internal compiler error: Segmentation fault
 1204 | FAST_NEAREST (8888_565_normal, 8888, 0565, uint32_t, uint16_t, OVER,
NORMAL)
      | ^~~~~~
unrecognized DWARF version in .debug_info at 6
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://bugs.buildroot.net/> for instructions.
$ sh4-linux-gcc -c -O2 test.c 
$ sh4-linux-gcc -c -O0 test.c

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[thomas.petazzoni@free-electrons.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

--- Comment #1 from Thomas Petazzoni <thomas.petazzoni@free-electrons.com> ---
This still happens with gcc 11.1.0, with the exact same conditions.

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
                 CC|                            |marxin at gcc dot gnu.org
   Last reconfirmed|                            |2021-08-03

--- Comment #2 from Martin Liška <marxin at gcc dot gnu.org> ---
Confirmed, reduced test-case:

$ cat pr101737.c
int fast_composite_scaled_nearest_8888_8888_cover_SRC_info_1,
    fast_composite_scaled_nearest_8888_8888_cover_SRC_info_0,
    fast_composite_scaled_nearest_8888_8888_cover_SRC_dst;

void
fast_composite_scaled_nearest_8888_8888_cover_SRC() {
  int width =
      fast_composite_scaled_nearest_8888_8888_cover_SRC_info_0;
  while (fast_composite_scaled_nearest_8888_8888_cover_SRC_info_1) {
    int w = fast_composite_scaled_nearest_8888_8888_cover_SRC_info_0;
    while ((w -= 2) >= 0)
      fast_composite_scaled_nearest_8888_8888_cover_SRC_dst++;
    if (w & 1)
      fast_composite_scaled_nearest_8888_8888_cover_SRC_dst = 0;
  }
}

crash happens for the following RTX instruction:
(code_label 40 31 87 4 5 (nil) [1 uses])

bool
sh_is_nott_insn (const rtx_insn* i)
{
  return i != NULL && GET_CODE (PATTERN (i)) == SET
         && t_reg_operand (XEXP (PATTERN (i), 0), VOIDmode)
         && negt_reg_operand (XEXP (PATTERN (i), 1), VOIDmode);
}

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[nospam-abuse at ilyaz dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

Ilya Zakharevich <nospam-abuse at ilyaz dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |nospam-abuse at ilyaz dot org

--- Comment #3 from Ilya Zakharevich <nospam-abuse at ilyaz dot org> ---
Another occurrence:

cc -c -I pari-2.3.5/src -I pari-2.3.5/src/headers -I pari-2.3.5/src -I
./libPARI  -I ../pari-2.3.5/src -I ../pari-2.3.5/src/headers -I
../pari-2.3.5/src/graph -I . -fwrapv -fno-strict-aliasing -pipe
-fstack-protector-strong -I/usr/local/include -D_LARGEFILE_SOURCE
-D_FILE_OFFSET_BITS=64  -O2   -DVERSION=\"\" -DXS_VERSION=\"\" -fPIC
-DPLOT_IS_TUNABLE -DHAS_STAT -DHAS_OPENDIR -DASMINLINE -DGCC_INLINE
-DDYNAMIC_PLOTTING -o trans2.o ../pari-2.3.5/src/basemath/trans2.c
during RTL pass: split1
../pari-2.3.5/src/basemath/trans2.c: In function âÂÂgasinâÂÂ:
../pari-2.3.5/src/basemath/trans2.c:214:1: internal compiler error:
Segmentation fault
  214 | }
      | ^

This was extracted from the remote testing report:
  https://www.cpantesters.org/cpan/report/cdb99a16-b09c-11ec-abe2-d569c5ce77ec
(This is “remote testing”, so this report is “as good as it gets”; I cannot
provide more details than in the report above.)

To reproduce: one should get and untar 
 
https://cpan.metacpan.org/authors/id/I/IL/ILYAZ/modules/Math-Pari-2.030522a.tar.gz

then run (in the extracted directory)
  perl Makefile.PL
  make

⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜

In fact, I can also produce the text of the subroutine, and how it is
preprocessed (on MY machine, with gcc 4.9.3):

GEN
gasin(GEN x, long prec)
{
  long sx;
  pari_sp av;
  GEN a, y, p1;

  switch(typ(x))
  {
    case t_REAL: sx = signe(x);
      if (!sx) return real_0_bit(expo(x));
      if (absrnz_egal1(x)) { /* |x| = 1 */
        if (sx > 0) return Pi2n(-1, lg(x)); /* 1 */
        y = Pi2n(-1, lg(x)); setsigne(y, -1); return y; /* -1 */
      }
      if (expo(x) < 0) return mpasin(x);
      y = cgetg(3,t_COMPLEX);
      gel(y,1) = Pi2n(-1, lg(x));
      gel(y,2) = mpach(x);
      if (sx < 0)
      {
        setsigne(y[1],-signe(y[1]));
        setsigne(y[2],-signe(y[2]));
      }
      return y;

    case t_COMPLEX:
      av = avma;
      return gerepilecopy(av, mulcxmI(gash(mulcxI(x), prec)));

    case t_INTMOD: case t_PADIC: pari_err(typeer,"gasin");
    default:
      av = avma; if (!(y = toser_i(x))) break;
      if (gcmp0(y)) return gcopy(y);
      /* lg(y) > 2*/
      if (valp(y) < 0) pari_err(negexper,"gasin");
      p1 = gsubsg(1,gsqr(y));
      if (gcmp0(p1))
      {
        GEN t = Pi2n(-1,prec);
        if (gsigne(gel(y,2)) < 0) setsigne(t, -1);
        return gerepileupto(av, scalarser(t, varn(y), valp(p1)>>1));
      }
      p1 = gdiv(derivser(y), gsqrt(p1,prec));
      a = integ(p1,varn(y));
      if (!valp(y)) a = gadd(a, gasin(gel(y,2),prec));
      return gerepileupto(av, a);
  }
  return transc(gasin,x,prec);
}

⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜⁜  Proprocessed (with gcc v4.9.3)

GEN
gasin(GEN x, long prec)
{
  long sx;
  pari_sp av;
  GEN a, y, p1;

  switch(((long)(((pari_ulong)((x)[0])) >> (32 - 7))))
  {
    case t_REAL: sx = (((long)((x)[1])) >> (32 - 2));
      if (!sx) return real_0_bit(((long) ((((pari_ulong)((x)[1])) & ((1UL<<(32
- 2))-1)) - (1UL<<((32 - 2)-1)))));
      if (absrnz_equal1(x)) {
        if (sx > 0) return Pi2n(-1, ((long)(((pari_ulong)((x)[0])) & ((1UL<<(32
- 1 - 7))-1))));
        y = Pi2n(-1, ((long)(((pari_ulong)((x)[0])) & ((1UL<<(32 - 1 -
7))-1)))); (((pari_ulong*)(y))[1]= (((pari_ulong*)(y))[1]&(~(~((1UL<<(32 - 2))
- 1)))) | (pari_ulong)(((pari_ulong)(-1)) << (32 - 2))); return y;
      }
      if (((long) ((((pari_ulong)((x)[1])) & ((1UL<<(32 - 2))-1)) - (1UL<<((32
- 2)-1)))) < 0) return mpasin(x);
      y = cgetg(3,t_COMPLEX);
      (((GEN*) (y))[1]) = Pi2n(-1, ((long)(((pari_ulong)((x)[0])) & ((1UL<<(32
- 1 - 7))-1))));
      (((GEN*) (y))[2]) = mpacosh(x);
      if (sx < 0) togglesign((((GEN*) (y))[1])); else togglesign((((GEN*)
(y))[2]));
      return y;

    case t_COMPLEX:
      if (ismpzero((((GEN*) (x))[2]))) return gasin((((GEN*) (x))[1]), prec);
      av = avma;
      return gerepilecopy(av, mulcxmI(gasinh(mulcxI(x), prec)));
    default:
      av = avma; if (!(y = toser_i(x))) break;
      if (gequal0(y)) return gerepilecopy(av, y);

      if (((long) ((((pari_ulong)((y)[1])) & ((1UL<<(32 - 2 - 14))-1)) -
(1UL<<((32 - 2 - 14)-1)))) < 0) pari_err_DOMAIN("asin","valuation", "<", gen_0,
x);
      p1 = gsubsg(1,gsqr(y));
      if (gequal0(p1))
      {
        GEN t = Pi2n(-1,prec);
        if (gsigne((((GEN*) (y))[2])) < 0) (((pari_ulong*)(t))[1]=
(((pari_ulong*)(t))[1]&(~(~((1UL<<(32 - 2)) - 1)))) |
(pari_ulong)(((pari_ulong)(-1)) << (32 - 2)));
        return gerepileupto(av, scalarser(t,
((long)((((pari_ulong)((y)[1]))&(((1UL<<14)-1)<<(32 - 2 - 14))) >> (32 - 2 -
14))), ((long) ((((pari_ulong)((p1)[1])) & ((1UL<<(32 - 2 - 14))-1)) -
(1UL<<((32 - 2 - 14)-1))))>>1));
      }
      p1 = gdiv(derivser(y), gsqrt(p1,prec));
      a = integser(p1);
      if (!((long) ((((pari_ulong)((y)[1])) & ((1UL<<(32 - 2 - 14))-1)) -
(1UL<<((32 - 2 - 14)-1))))) a = gadd(a, gasin((((GEN*) (y))[2]),prec));
      return gerepileupto(av, a);
  }
  return trans_eval("asin",gasin,x,prec);
}

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[pietro.gcc at sociotechnical dot xyz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

pietro <pietro.gcc at sociotechnical dot xyz> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pietro.gcc at sociotechnical dot x
                   |                            |yz

--- Comment #4 from pietro <pietro.gcc at sociotechnical dot xyz> ---
Using the sh4-linux cross compiler from debian unstable:

$ sh4-linux-gnu-gcc-14 -v
Using built-in specs.
COLLECT_GCC=sh4-linux-gnu-gcc-14
COLLECT_LTO_WRAPPER=/usr/libexec/gcc-cross/sh4-linux-gnu/14/lto-wrapper
Target: sh4-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Debian 14-20240201-3'
--with-bugurl=file:///usr/share/doc/gcc-14/README.Bugs
--enable-languages=c,ada,c++,fortran,objc,obj-c++,rust --prefix=/usr
--with-gcc-major-version-only --program-suffix=-14 --enable-shared
--enable-linker-build-id --libexecdir=/usr/libexec --without-included-gettext
--enable-threads=posix --libdir=/usr/lib --enable-nls --with-sysroot=/
--enable-clocale=gnu --enable-libstdcxx-debug --enable-libstdcxx-time=yes
--with-default-libstdcxx-abi=new --enable-libstdcxx-backtrace
--enable-gnu-unique-object --disable-libsanitizer --disable-libquadmath
--disable-libquadmath-support --enable-plugin --with-system-zlib
--enable-multiarch --disable-werror --with-cpu=sh4
--with-multilib-list=m4,m4-nofpu --enable-checking=yes
--build=aarch64-linux-gnu --host=aarch64-linux-gnu --target=sh4-linux-gnu
--program-prefix=sh4-linux-gnu- --includedir=/usr/sh4-linux-gnu/include
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 14.0.1 20240131 (experimental) [master r14-8680-g2f14c0dbb78]
(Debian 14-20240201-3)

I'm able to reproduce the issue with the attached test source file:

# sh4-linux-gnu-gcc-14 -c -Os test.c
during RTL pass: split1
pixman-fast-path.c: In function
'fast_composite_scaled_nearest_8888_565_cover_OVER':
pixman-fast-path.c:1201: internal compiler error: Segmentation fault
0xde6a7b crash_signal
        ../../src/gcc/toplev.cc:317
0x11ed95c sh_is_nott_insn(rtx_insn const*)
        ../../src/gcc/config/sh/sh.cc:11770
0x16f82df gen_split_13(rtx_insn*, rtx_def**)
        ../../src/gcc/config/sh/sh.md:951
0x946e5b try_split(rtx_def*, rtx_insn*, int)
        ../../src/gcc/emit-rtl.cc:3941
0xd3223f split_insn
        ../../src/gcc/recog.cc:3405
0xd37a97 split_all_insns()
        ../../src/gcc/recog.cc:3509
0xd37b8b execute
        ../../src/gcc/recog.cc:4433

But I'm not able to reproduce using the code from comment #2. I used creduce to
get a simple test case:

$ cat test2.c
int a;
void b() {
  int c;
  short *d = (short *)b;
  while (--c) {
    int e = a, f, g;
    while ((e -= 4) >= 0) {
      f += g;
      d++;
    }
    if (e & 2)
      f += g;
    if (e & 1)
      *d = f;
  }
}

$ sh4-linux-gnu-gcc-14 -c -Os test2.c
during RTL pass: split1
test2.c: In function 'b':
test2.c:16:1: internal compiler error: Segmentation fault
   16 | }
      | ^
0xde6a7b crash_signal
        ../../src/gcc/toplev.cc:317
0x11ed95c sh_is_nott_insn(rtx_insn const*)
        ../../src/gcc/config/sh/sh.cc:11770
0x16f82df gen_split_13(rtx_insn*, rtx_def**)
        ../../src/gcc/config/sh/sh.md:951
0x946e5b try_split(rtx_def*, rtx_insn*, int)
        ../../src/gcc/emit-rtl.cc:3941
0xd3223f split_insn
        ../../src/gcc/recog.cc:3405
0xd37a97 split_all_insns()
        ../../src/gcc/recog.cc:3509
0xd37b8b execute
        ../../src/gcc/recog.cc:4433

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

--- Comment #5 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Oleg Endo <olegendo@gcc.gnu.org>:

https://gcc.gnu.org/g:4ff8ffe7331cf174668cf5c729fd68ff327ab014

commit r14-9278-g4ff8ffe7331cf174668cf5c729fd68ff327ab014
Author: Oleg Endo <olegendo@gcc.gnu.org>
Date:   Sun Mar 3 14:58:58 2024 +0900

    SH: Fix 101737

    gcc/ChangeLog:
            PR target/101737
            * config/sh/sh.cc (sh_is_nott_insn): Handle case where the input
            is not an insn, but e.g. a code label.

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

--- Comment #6 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-13 branch has been updated by Oleg Endo
<olegendo@gcc.gnu.org>:

https://gcc.gnu.org/g:a38b3dfc71d6b5d07477715a3a6df7b73ebaa68d

commit r13-8402-ga38b3dfc71d6b5d07477715a3a6df7b73ebaa68d
Author: Oleg Endo <olegendo@gcc.gnu.org>
Date:   Sun Mar 3 14:58:58 2024 +0900

    SH: Fix 101737

    gcc/ChangeLog:
            PR target/101737
            * config/sh/sh.cc (sh_is_nott_insn): Handle case where the input
            is not an insn, but e.g. a code label.

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

--- Comment #7 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Oleg Endo
<olegendo@gcc.gnu.org>:

https://gcc.gnu.org/g:bbae41dc9033d6f0a9f8bc56cc6f80d90286996c

commit r12-10192-gbbae41dc9033d6f0a9f8bc56cc6f80d90286996c
Author: Oleg Endo <olegendo@gcc.gnu.org>
Date:   Sun Mar 3 14:58:58 2024 +0900

    SH: Fix 101737

    gcc/ChangeLog:
            PR target/101737
            * config/sh/sh.cc (sh_is_nott_insn): Handle case where the input
            is not an insn, but e.g. a code label.

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

--- Comment #8 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Oleg Endo
<olegendo@gcc.gnu.org>:

https://gcc.gnu.org/g:ec65cb598cc6fa126b458cf716438cc3f2404f3c

commit r11-11267-gec65cb598cc6fa126b458cf716438cc3f2404f3c
Author: Oleg Endo <olegendo@gcc.gnu.org>
Date:   Sun Mar 3 14:58:58 2024 +0900

    SH: Fix 101737

    gcc/ChangeLog:
            PR target/101737
            * config/sh/sh.c (sh_is_nott_insn): Handle case where the input
            is not an insn, but e.g. a code label.

[Bug target/101737] SH4 -Os causes internal compiler error when building pixman

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101737

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
                 CC|                            |olegendo at gcc dot gnu.org
             Status|NEW                         |RESOLVED

--- Comment #9 from Oleg Endo <olegendo at gcc dot gnu.org> ---
Thanks everyone for staying on this and re-testing.  It should be fixed now on
the open branches.  If you want to use a version older than GCC 11, please
apply the committed patch to your GCC source.