[Bug target/99706] New: [11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[Bug target/99706] New: [11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[asolokha at gmx dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

            Bug ID: 99706
           Summary: [11 Regression] ICE: maximum number of generated
                    reload insns per insn achieved (90)
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: asolokha at gmx dot com
  Target Milestone: ---
            Target: aarch64-linux-gnu

gcc-11.0.1-alpha20210321 snapshot (g:fc24ea2374259d401a46ce3526688b7e79d4cc13)
ICEs when compiling the following testcase w/ -O2 -ftree-loop-vectorize
-fno-tree-pta --param scev-max-expr-size=5:

int kk;

void
ii (int BUFSIZE)
{
  char buf[BUFSIZE], *ptr = buf;
  int i;

  for (i = 0; i < BUFSIZE; i++)
    *ptr++ = kk = 0;
}

% aarch64-linux-gnu-gcc-11.0.1 -O2 -ftree-loop-vectorize -fno-tree-pta --param
scev-max-expr-size=5 -c oeplxogh.c
during RTL pass: reload
oeplxogh.c: In function 'ii':
oeplxogh.c:11:1: internal compiler error: maximum number of generated reload
insns per insn achieved (90)
   11 | }
      | ^
0xd42ab0 lra_constraints(bool)
       
/var/tmp/portage/cross-aarch64-linux-gnu/gcc-11.0.1_alpha20210321/work/gcc-11-20210321/gcc/lra-constraints.c:5091
0xd2b3e5 lra(_IO_FILE*)
       
/var/tmp/portage/cross-aarch64-linux-gnu/gcc-11.0.1_alpha20210321/work/gcc-11-20210321/gcc/lra.c:2336
0xce15c4 do_reload
       
/var/tmp/portage/cross-aarch64-linux-gnu/gcc-11.0.1_alpha20210321/work/gcc-11-20210321/gcc/ira.c:5834
0xce15c4 execute
       
/var/tmp/portage/cross-aarch64-linux-gnu/gcc-11.0.1_alpha20210321/work/gcc-11-20210321/gcc/ira.c:6020

I believe it is different from PR97141.

[Bug target/99706] [11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |11.0
                 CC|                            |pinskia at gcc dot gnu.org

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Thos are some interesting options.  How did you decide to use them?

Note the options used here are unrelated to the ICE though.

[Bug target/99706] [11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-03-22
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
It is this RTL which is causing the issue:
(insn 31 27 29 4 (set (reg:V2DI 125 [ ivtmp.27 ])
        (subreg:V2DI (reg/f:DI 31 sp) 0)) 1163 {*aarch64_simd_movv2di}
     (expr_list:REG_DEAD (reg:TI 32 v0)
        (nil)))

Which was created while expanding (well sp is not there right away, it is done
later on):
  _42 = (sizetype) buf.1_10;
  _43 = _42 + 1;
  _44 = {_42, _43};

[Bug target/99706] [11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[asolokha at gmx dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

--- Comment #3 from Arseny Solokha <asolokha at gmx dot com> ---
(In reply to Andrew Pinski from comment #1)
> Thos are some interesting options.  How did you decide to use them?

TL;DR: they were randomly chosen.

I run (somewhat) automated gcc fuzz testing for several years by now. I briefly
explained it in the first paragraph of PR92115 comment 4 (and likely
elsewhere), but can discuss it in more details if so desired.

[Bug target/99706] [11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Likely an "RTL expansion testcase" using the GIMPLE FE with IL as seen around
RTL expansion could more reliably produce the bad IL.

[Bug target/99706] [11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
ICEs since r11-272-gb6ff3ddecfa93d53867afaaa078f85fc848abbbd

[Bug target/99706] [10/11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|11.0                        |10.1.0
            Summary|[11 Regression] ICE:        |[10/11 Regression] ICE:
                   |maximum number of generated |maximum number of generated
                   |reload insns per insn       |reload insns per insn
                   |achieved (90)               |achieved (90)
      Known to work|                            |7.3.0, 7.4.0
      Known to fail|                            |10.0, 11.0
   Target Milestone|11.0                        |10.3

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Here is a testcase which fails at -O1 (without any other options) and above:
int kk;

typedef unsigned long long v2di __attribute__((vector_size(16) ));

void f(v2di);

void
ii (int BUFSIZE)
{
  char buf[BUFSIZE], *ptr = buf;
  int i;
  unsigned long t = (unsigned long)buf;
  v2di tt = {t, t+1};
  f(tt);
}
----- CUT ----
This worked in GCC 7.3.0 and 7.4.0 but fails with 10.1.0 and above.  I don't
have 9.x to test on though.

[Bug target/99706] [9/10/11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.3                        |9.4
            Summary|[10/11 Regression] ICE:     |[9/10/11 Regression] ICE:
                   |maximum number of generated |maximum number of generated
                   |reload insns per insn       |reload insns per insn
                   |achieved (90)               |achieved (90)

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The #c6 testcase started to ICE with
r9-374-gb4e2cd5b9a995e9154b1169e83a0347d257a938c

[Bug target/99706] [9/10/11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Tried
--- aarch64-simd.md.jj1 2021-03-08 23:40:33.846448416 +0100
+++ aarch64-simd.md     2021-03-23 12:40:51.325018722 +0100
@@ -133,7 +133,7 @@
   [(set (match_operand:VQMOV 0 "nonimmediate_operand"
                "=w, Umn,  m,  w, ?r, ?w, ?r, w")
        (match_operand:VQMOV 1 "general_operand"
-               "m,  Dz, w,  w,  w,  r,  r, Dn"))]
+               "m,  Dz, w,  w,  w,  rk,  rk, Dn"))]
   "TARGET_SIMD
    && (register_operand (operands[0], <MODE>mode)
        || aarch64_simd_reg_or_zero (operands[1], <MODE>mode))"
@@ -1039,7 +1039,7 @@
   [(set (match_operand:VALL_F16 0 "register_operand" "=w,w,w")
        (vec_merge:VALL_F16
            (vec_duplicate:VALL_F16
-               (match_operand:<VEL> 1 "aarch64_simd_general_operand"
"w,?r,Utv"))
+               (match_operand:<VEL> 1 "aarch64_simd_general_operand"
"w,?rk,Utv"))
            (match_operand:VALL_F16 3 "register_operand" "0,0,0")
            (match_operand:SI 2 "immediate_operand" "i,i,i")))]
   "TARGET_SIMD"
but it didn't help.

[Bug target/99706] [9/10/11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
                 CC|                            |vmakarov at gcc dot gnu.org

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
CCing Vlad if he can quickly point out what the bug is but I suspect just a
backend bug, I think the sp register isn't included in "r" and one needs "k"
for that instead.

[Bug target/99706] [9/10/11/12 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|9.4                         |9.5

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 9.4 is being released, retargeting bugs to GCC 9.5.

[Bug target/99706] [10/11/12/13 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|9.5                         |10.4

--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 9 branch is being closed

[Bug target/99706] [10/11/12/13 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.4                        |10.5

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 10.4 is being released, retargeting bugs to GCC 10.5.

[Bug target/99706] [10/11/12/13 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[asolokha at gmx dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

--- Comment #13 from Arseny Solokha <asolokha at gmx dot com> ---
I cannot reproduce it anymore w/ gcc 13.0.0 20221016 snapshot
(g:6366e3e8847af98d4728d55951534769d034d02a) and w/ gcc 12.2, though 11.3 is
still affected.

[Bug target/99706] [11/12/13/14 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.5                        |11.5

--- Comment #14 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 10 branch is being closed.

[Bug target/99706] [11 Regression] ICE: maximum number of generated reload insns per insn achieved (90)

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99706

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[11/12/13/14 Regression]    |[11 Regression] ICE:
                   |ICE: maximum number of      |maximum number of generated
                   |generated reload insns per  |reload insns per insn
                   |insn achieved (90)          |achieved (90)
                 CC|                            |law at gcc dot gnu.org

--- Comment #15 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Adjusting regression markers per c#13.