[Bug target/101697] New: [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[Bug target/101697] New: [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

            Bug ID: 101697
           Summary: [11/12 regression] ICE compiling uClibc-ng for
                    h8300-linux
           Product: gcc
           Version: 11.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: mikpelinux at gmail dot com
  Target Milestone: ---

Created attachment 51226
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51226&action=edit
pre-processed source of uClibc's getaddrinfo.c

Attempting to build uClibc-ng-1.0.38 with a gcc-11.2.0 cross to
h8300-unknown-linux-uclibc fails with an ICE:

libc/inet/getaddrinfo.c: In function 'gaih_inet':
libc/inet/getaddrinfo.c:801:1: error: unrecognizable insn:
  801 | }
      | ^
(insn 1444 1443 164 31 (parallel [
            (set (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [3  S4 A32])
                (reg/f:SI 7 sp))
            (clobber (reg:CC 12 cc))
        ]) "libc/inet/getaddrinfo.c":466:11 -1
     (expr_list:REG_ARGS_SIZE (const_int 4 [0x4])
        (nil)))
during RTL pass: split2
libc/inet/getaddrinfo.c:801:1: internal compiler error: in extract_insn, at
recog.c:2770
0x41a6f0 _fatal_insn(char const*, rtx_def const*, char const*, int, char
const*)
        /mnt/scratch/cross/sources/gcc-11.2.0/gcc/rtl-error.c:108
0x41a70c _fatal_insn_not_found(rtx_def const*, char const*, int, char const*)
        /mnt/scratch/cross/sources/gcc-11.2.0/gcc/rtl-error.c:116
0x4199e8 extract_insn(rtx_insn*)
        /mnt/scratch/cross/sources/gcc-11.2.0/gcc/recog.c:2770
0x8cf1c4 extract_insn_cached(rtx_insn*)
        /mnt/scratch/cross/sources/gcc-11.2.0/gcc/recog.c:2659
0x67c782 cleanup_subreg_operands(rtx_insn*)
        /mnt/scratch/cross/sources/gcc-11.2.0/gcc/final.c:3330
0x8cd42f split_insn
        /mnt/scratch/cross/sources/gcc-11.2.0/gcc/recog.c:3399
0x8d2407 split_all_insns()
        /mnt/scratch/cross/sources/gcc-11.2.0/gcc/recog.c:3467
0x8d2498 execute
        /mnt/scratch/cross/sources/gcc-11.2.0/gcc/recog.c:4385

A gcc-12 cross also ICEs on the same location, but with a different message:

libc/inet/getaddrinfo.c: In function 'gaih_inet':
libc/inet/getaddrinfo.c:801:1: error: could not split insn
  801 | }
      | ^
(insn 163 1443 1444 (set (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [3  S4 A32])
        (reg/f:SI 7 sp)) "libc/inet/getaddrinfo.c":466:11 19 {*movsi}
     (expr_list:REG_ARGS_SIZE (const_int 4 [0x4])
        (nil)))
during RTL pass: final
libc/inet/getaddrinfo.c:801:1: internal compiler error: in final_scan_insn_1,
at final.c:2861
0x41abf2 _fatal_insn(char const*, rtx_def const*, char const*, int, char
const*)
        /mnt/scratch/cross/sources/gcc-12-20210725/gcc/rtl-error.c:108
0x40ded8 final_scan_insn_1
        /mnt/scratch/cross/sources/gcc-12-20210725/gcc/final.c:2861
0x687d9b final_scan_insn(rtx_insn*, _IO_FILE*, int, int, int*)
        /mnt/scratch/cross/sources/gcc-12-20210725/gcc/final.c:2940
0x688064 final_1
        /mnt/scratch/cross/sources/gcc-12-20210725/gcc/final.c:1997
0x688842 rest_of_handle_final
        /mnt/scratch/cross/sources/gcc-12-20210725/gcc/final.c:4285
0x688842 execute,
        /mnt/scratch/cross/sources/gcc-12-20210725/gcc/final.c:4363

gcc-10.3.0 has no problems building uClibc-1.0.38 for the h8300.

Attaching the preprocessed getaddrinfo.i. Compile with -fno-builtin and -Os,
-O2, or -O3 to trigger the ICE. Removing -fno-builtin or reducing optimization
to -Og, -O1, or -O0 stops the ICE.

The cross-compilers were configured (yes there's some redundancies there):
--target=h8300-unknown-linux-uclibc --build=x86_64-pc-linux-gnu
--prefix=/mnt/scratch/cross/cross-h8300
--with-sysroot=/mnt/scratch/cross/cross-h8300
--with-gmp=/mnt/scratch/cross/host-tools/gmp-6.2.1
--with-mpfr=/mnt/scratch/cross/host-tools/mpfr-4.1.0
--with-mpc=/mnt/scratch/cross/host-tools/mpc-1.2.1 --without-cloog
--without-isl --disable-libatomic --disable-libgomp --disable-libitm
--disable-libmpx --disable-libmudflap --disable-libquadmath
--disable-libsanitizer --disable-lto --disable-nls --disable-plugin
--disable-shared --enable-checking=release --enable-languages=c
--enable-multilib --disable-threads --disable-libssp

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |11.3
           Keywords|                            |ice-on-valid-code

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

--- Comment #1 from Mikael Pettersson <mikpelinux at gmail dot com> ---
The ICE in gcc-11 started with:
[f16897cb4b1468374d63b1a6b12d8b7be845874a] H8 cc0 conversion

It changed from "unrecognizable insn" to "could not split insn" in gcc-12 with:
[549d7f4310f6f8c2c64efcb6f3efcee99c9d9f4f] Fix split conditions in H8/300 port

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

--- Comment #2 from Jeffrey A. Law <law at gcc dot gnu.org> ---
I can't reproduce this on the trunk or with the referenced git hashes.

(insn 1444 1443 164 31 (parallel [
            (set (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [3  S4 A32])
                (reg/f:SI 7 sp))
            (clobber (reg:CC 12 cc))
        ]) "libc/inet/getaddrinfo.c":466:11 -1
     (expr_list:REG_ARGS_SIZE (const_int 4 [0x4])
        (nil)))

I'm pretty sure this is ill-formed though.  The register inside the
auto-increment addressing mode isn't allowed to be used elsewhere in the same
insn.  In particular note that we have (pre_dec (reg sp)) and another reference
to (reg sp) in the same insn.

I would probably start by identifying when that happens in the dumps.

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

--- Comment #3 from Mikael Pettersson <mikpelinux at gmail dot com> ---
As far as I can tell the problem is introduced by reload.

With a gcc-11.2.0 cross, getaddrinfo.i.290r.ira has

(insn 161 159 163 31 (set (reg/f:SI 185)
        (reg/f:SI 7 sp)) "libc/inet/getaddrinfo.c":465:13 13 {*movsi}
     (nil))
(insn 163 161 164 31 (set (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [3  S4 A32])
        (reg/f:SI 185)) "libc/inet/getaddrinfo.c":466:11 13 {*movsi}
     (expr_list:REG_ARGS_SIZE (const_int 4 [0x4])
        (nil)))

these are adjacent.

Then getaddrinfo.i.291r.reload has

Reloads for insn # 161
Reload 0: reload_out (SI) = (reg/f:SI 185)
        NO_REGS, RELOAD_FOR_OUTPUT (opnum = 0), optional
        reload_out_reg: (reg/f:SI 185)

Reloads for insn # 163
Reload 0: reload_out (SI) = (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [3  S4 A32])
        NO_REGS, RELOAD_FOR_OUTPUT (opnum = 0), optional
        reload_out_reg: (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [3  S4 A32])
Reload 1: reload_in (SI) = (reg/f:SI 185)
        GENERAL_REGS, RELOAD_FOR_INPUT (opnum = 1)
        reload_in_reg: (reg/f:SI 185)
        reload_reg_rtx: (reg/f:SI 7 sp)

and later

(insn 161 159 995 31 (set (mem/c:SI (plus:SI (reg/f:SI 6 r6)
                (const_int -140 [0xffffffffffffff74])) [30 %sfp+-132 S4 A32])
        (reg/f:SI 7 sp)) "libc/inet/getaddrinfo.c":465:13 13 {*movsi}
     (nil))
(insn 995 161 163 31 (set (reg/f:SI 7 sp)
        (mem/c:SI (plus:SI (reg/f:SI 6 r6)
                (const_int -140 [0xffffffffffffff74])) [30 %sfp+-132 S4 A32]))
"libc/inet/getaddrinfo.c":466:11 13 {*movsi}
     (expr_list:REG_ARGS_SIZE (const_int 0 [0])
        (nil)))
(insn 163 995 164 31 (set (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [3  S4 A32])
        (reg/f:SI 7 sp)) "libc/inet/getaddrinfo.c":466:11 13 {*movsi}
     (expr_list:REG_ARGS_SIZE (const_int 4 [0x4])
        (nil)))

It's strange you can't reproduce. My current build is literally:

> mkdir objdir-binutils; cd objdir-binutils
> /tmp/binutils-2.37/configure --target=h8300-unknown-linux-uclibc --prefix=/tmp/cross-h8300 --with-sysroot=/tmp/cross-h8300
> make -jN
> make install
> cd ..
> mkdir objdir-gcc; cd objdir-gcc
> PATH=/tmp/cross-h8300/bin:/usr/bin /tmp/gcc-11.2.0/configure --target=h8300-unknown-linux-uclibc --prefix=/tmp/cross-h8300 --with-sysroot=/tmp/cross-h8300 --with-newlib --disable-libquadmath --disable-libssp --disable-threads --disable-libatomic --disable-libgomp --enable-languages=c --enable-checking=release
> PATH=/tmp/cross-h8300/bin:/usr/bin make -jN
> gcc/xgcc -Bgcc -fno-builtin -Os -S /tmp/getaddrinfo.i

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

Jeffrey A. Law <law at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |law at gcc dot gnu.org
   Last reconfirmed|                            |2021-08-17

--- Comment #4 from Jeffrey A. Law <law at gcc dot gnu.org> ---
I've managed to reproduce.  It's possible I was missing the -fno-builtin on my
command line.

A much simplified testcase:
struct gaih_servtuple
{
  struct gaih_servtuple *next;
};


int
gaih_inet (const char *name,  const void *req)
{
  struct gaih_servtuple *st;
  int *tp;
  if (req)
    ++tp;

  struct gaih_servtuple **pst = &st;
  while (1)
    {
      struct gaih_servtuple *newp;


      newp = __builtin_alloca (sizeof (struct gaih_servtuple));
      frob (name, tp, req, newp);
      *pst = newp;
      pst = &(newp->next);
      tp++;
    }
}

We're fine up to and including IRA.
(insn 20 18 22 5 (set (reg/f:SI 28)
        (reg/f:SI 7 sp)) "k.c":21:14 19 {*movsi}
     (nil))
(insn 22 20 23 5 (set (mem/f:SI (pre_dec:SI (reg/f:SI 7 sp)) [1  S4 A32])
        (reg/f:SI 28)) "k.c":22:7 19 {*movsi}
     (expr_list:REG_ARGS_SIZE (const_int 4 [0x4])
        (nil)))

And IRA allocates hard reg #6 for pseudo #28, which is also fine.

Reload reports:
Reload 1: reload_in (SI) = (reg/f:SI 28)
        GENERAL_REGS, RELOAD_FOR_INPUT (opnum = 1)
        reload_in_reg: (reg/f:SI 28)
        reload_reg_rtx: (reg/f:SI 7 sp)

Which is not good as replacing (reg 28) with (reg sp) is not valid.

I haven't tried converting the H8 to LRA yet, which would be the ultimate
solution if it worked.  Ideally any port not converted to LRA ought to be
flagged as deprecated for gcc-12 and dropped if not fixed by gcc-13.

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

Jeffrey A. Law <law at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|11.3                        |11.4

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 11.3 is being released, retargeting bugs to GCC 11.4.

[Bug target/101697] [11/12/13 regression] ICE compiling uClibc-ng for h8300-linux

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

--- Comment #6 from Jeffrey A. Law <law at gcc dot gnu.org> ---
So this issue has come up again in the context of LRA conversion which happens
to trip over the same bug, but with a different testcase.

At the core of this problem is reload and LRA will both generate invalid RTL
when performing register eliminations.  Specifically, they will create an
autoinc addressing mode where the incremented/decremented register is used
elsewhere in the same insn as a source operand.  Such RTL has been considered
invalid as long as I can remember.

The H8 backend does try to prevent this behavior by checking for this scenario
and rejecting such insns in the insn condition.  But in both the reload and LRA
cases, they make substitutions in the original insn without validating the
resulting insn (which would have failed).  Even if they did try to validate the
resulting insn, neither has a code generation strategy to deal with a failed
substitution during register eliminations.

Paul K. indicated how the pdp11 port handles these cases with constraints. 
Using constraints alone was insufficient to fix this problem, but using
constraints in conjunction with the existing insn condition checks does seem to
fix this problem.  I'm currently upstreaming the various bits to make that
happen.

[Bug target/101697] [11/12/13 regression] ICE compiling uClibc-ng for h8300-linux

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

--- Comment #7 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jeff Law <law@gcc.gnu.org>:

https://gcc.gnu.org/g:4374c424a60777a7658050f0aeb1dcc9af915647

commit r13-3348-g4374c424a60777a7658050f0aeb1dcc9af915647
Author: Jeff Law <jeffreyalaw@gmail.com>
Date:   Mon Oct 17 19:52:18 2022 -0400

    Fix bogus RTL on the H8.

    This patch actually fixes the bogus RTL seen in PR101697.

    Basically we continue to use the insn condition to catch most of the
problem
    cases related to autoinc addressing modes.  This patch adds constraints
which
    can guide reload (and hopefully LRA) away from doing blind replacements
during
    register elimination that would ultimately result in bogus RTL.  The idea
is
    from Paul K. who has done something very similar on the pdp11.  I guess it
    shouldn't be a big surprise that the H8 and pdp11 need the same kind of
    handling given some of the similarities in their architectures.

    gcc/
            PR target/101697
            * config/h8300/combiner.md: Replace '<' preincment constraint with
            ZA/Z1..ZH/Z7 combinations.
            * config/h8300/movepush.md: Similarly

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

Jeffrey A. Law <law at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[11/12/13 regression] ICE   |[11/12 regression] ICE
                   |compiling uClibc-ng for     |compiling uClibc-ng for
                   |h8300-linux                 |h8300-linux

--- Comment #8 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Fixed for gcc-13.  No plans to backport.

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

--- Comment #9 from Mikael Pettersson <mikpelinux at gmail dot com> ---
I can confirm that building the latest uClibc-ng for h8300 now succeeds with
gcc master @ 4374c424a60777a7658050f0aeb1dcc9af915647.

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|11.4                        |13.0
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED

--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
.

[Bug target/101697] [11/12 regression] ICE compiling uClibc-ng for h8300-linux

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101697

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mike at mnmoran dot org

--- Comment #11 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 108913 has been marked as a duplicate of this bug. ***