[Bug rtl-optimization/96998] New: GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[Bug rtl-optimization/96998] New: GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[mkuvyrkov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

            Bug ID: 96998
           Summary: GCC ICEs in on building AArch64 Linux kernel after
                    basepoints/gcc-11-2903-g6b3034eaba83
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: mkuvyrkov at gcc dot gnu.org
  Target Milestone: ---

GCC ICEs on building Linux kernel for AArch64 target after
basepoints/gcc-11-2903-g6b3034eaba83:
===
commit 6b3034eaba83935d9f6dfb20d2efbdb34b5b00bf
Author: Alex Coplan <alex.coplan@arm.com>
Date:   Thu Aug 27 09:49:57 2020 +0100

    lra: Canonicalize mult to shift in address reloads
===

The error message is:
===
00:01:02 during RTL pass: reload
00:01:02 mm/backing-dev.c: In function ‘clear_bdi_congested’:
00:01:02 mm/backing-dev.c:930:1: internal compiler error: in
lra_set_insn_recog_data, at lra.c:1006
00:01:02   930 | }
00:01:02       | ^
00:01:03 0xb4f991 lra_set_insn_recog_data(rtx_insn*)
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra.c:1004
00:01:03 0xb4feb7 lra_get_insn_recog_data
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra-int.h:488
00:01:03 0xb4feb7 lra_update_insn_regno_info(rtx_insn*)
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra.c:1625
00:01:03 0xb50027 lra_update_insn_regno_info(rtx_insn*)
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra.c:1774
00:01:03 0xb50027 lra_push_insn_1
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra.c:1780
00:01:03 0xb50027 lra_push_insn(rtx_insn*)
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra.c:1788
00:01:03 0xb50027 push_insns
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra.c:1831
00:01:03 0xb51c0c push_insns
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/rtl.h:886
00:01:03 0xb51c0c lra_process_new_insns(rtx_insn*, rtx_insn*, rtx_insn*, char
const*)
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra.c:1879
00:01:03 0xb6981e curr_insn_transform
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra-constraints.c:4572
00:01:03 0xb6b7cd lra_constraints(bool)
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra-constraints.c:5066
00:01:03 0xb52c4c lra(_IO_FILE*)
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/lra.c:2415
00:01:03 0xb032a9 do_reload
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/ira.c:5525
00:01:03 0xb032a9 execute
00:01:03       
/home/tcwg-buildslave/workspace/tcwg_kernel_0/abe/snapshots/gcc.git~master/gcc/ira.c:5711
00:01:03 Please submit a full bug report,
00:01:03 with preprocessed source if appropriate.
00:01:03 Please include the complete backtrace with any bug report.
00:01:03 See <https://gcc.gnu.org/bugs/> for instructions.
00:01:03 scripts/Makefile.build:283: recipe for target 'mm/backing-dev.o'
failed
00:01:03 make[1]: *** [mm/backing-dev.o] Error 1
===

Toolchain configurations is a no-frills default aarch64-linux-gnu toolchain,
but only linux-next tree seems to be affected.  It's possible that the ICE is
due to a wrong register constraint in a poorly-tested patch in linux-next. 
Nevertheless, the ICE is worth investigating.

The linux-next tree as tested was "7a6956579ce69 Add linux-next specific files
for 20200903".

More details on the toolchain build:
https://ci.linaro.org/view/tcwg_kernel-gnu/job/tcwg_kernel-gnu-bisect-gnu-master-aarch64-next-allnoconfig/4/artifact/artifacts/build-first_bad/04-build_abe-stage1/console.log/*view*/
More details on the linux build:
https://ci.linaro.org/view/tcwg_kernel-gnu/job/tcwg_kernel-gnu-bisect-gnu-master-aarch64-next-allnoconfig/4/artifact/artifacts/build-first_bad/05-build_linux/console.log/*view*/
.

[Bug rtl-optimization/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[mkuvyrkov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

Maxim Kuvyrkov <mkuvyrkov at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |acoplan at gcc dot gnu.org,
                   |                            |mkuvyrkov at gcc dot gnu.org
             Target|                            |aarch64-linux-gnu

--- Comment #1 from Maxim Kuvyrkov <mkuvyrkov at gcc dot gnu.org> ---
Alex, would you please look into this?

Let me know if you have problems reproducing the crash and I'll make a smaller
reproducer.

[Bug rtl-optimization/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

--- Comment #2 from Alex Coplan <acoplan at gcc dot gnu.org> ---
I'll take a look, thanks.

[Bug rtl-optimization/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

Alex Coplan <acoplan at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-09-09
     Ever confirmed|0                           |1
           Assignee|unassigned at gcc dot gnu.org      |acoplan at gcc dot gnu.org
             Status|UNCONFIRMED                 |ASSIGNED

--- Comment #3 from Alex Coplan <acoplan at gcc dot gnu.org> ---
Confirmed. GCC trunk ICEs on AArch64 when compiling kernel/cgroup/cgroup.c.
Seems to happen when compiling multiple kernel source files.

I'll reduce a testcase and take a look.

[Bug rtl-optimization/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

--- Comment #4 from Alex Coplan <acoplan at gcc dot gnu.org> ---
Ok, the following reduced testcase ICEs at -O2:

int h(void);
struct c d;
struct c {
  int e[1];
};

void f(void) {
  int g;
  for (;; g = h()) {
    int *i = &d.e[g];
    asm("" : "=Q"(*i));
  }
}

[Bug rtl-optimization/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

--- Comment #5 from Alex Coplan <acoplan at gcc dot gnu.org> ---
This just looks like a missing pattern which we've started noticing since the
new canonicalisation was introduced.

I imagine reinstating the *add_<optab><mode>_multp2 pattern and rewriting it to
use shifts instead will fix the issue.

I should have a patch to fix this tomorrow.

[Bug target/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

--- Comment #6 from Alex Coplan <acoplan at gcc dot gnu.org> ---
I have a patch, just testing it now.

[Bug target/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

--- Comment #7 from Alex Coplan <acoplan at gcc dot gnu.org> ---
Patch up for review:
https://gcc.gnu.org/pipermail/gcc-patches/2020-September/553605.html

[Bug target/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

Alex Coplan <acoplan at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dr.duncan.p.simpson at gmail dot c
                   |                            |om

--- Comment #8 from Alex Coplan <acoplan at gcc dot gnu.org> ---
*** Bug 97275 has been marked as a duplicate of this bug. ***

[Bug target/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

--- Comment #9 from Alex Coplan <acoplan at gcc dot gnu.org> ---
So the plan is to fix this with a patch to combine. Waiting on a review from
Segher for https://gcc.gnu.org/pipermail/gcc-patches/2020-September/555158.html

[Bug target/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

Alex Coplan <acoplan at gcc dot gnu.org> changed:

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

--- Comment #10 from Alex Coplan <acoplan at gcc dot gnu.org> ---
*** Bug 97526 has been marked as a duplicate of this bug. ***

[Bug target/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

--- Comment #11 from Alex Coplan <acoplan at gcc dot gnu.org> ---
The patch series has been restructured to first fix the AArch64 bug, and then
restore code quality with a patch to combine.

The AArch64 patch
(https://gcc.gnu.org/pipermail/gcc-patches/2020-October/556237.html) has been
approved, we're waiting for a review of the 2/2 combine patch.

[Bug target/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

--- Comment #12 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Alex Coplan <acoplan@gcc.gnu.org>:

https://gcc.gnu.org/g:7de23b8c536397117bbea04a722fa1b86564dd7c

commit r11-4563-g7de23b8c536397117bbea04a722fa1b86564dd7c
Author: Alex Coplan <alex.coplan@arm.com>
Date:   Fri Oct 30 09:21:31 2020 +0000

    aarch64: Fix PR96998 and restore code quality in combine

    This change fixes a bug in the AArch64 backend. Currently, we accept an
    odd sign_extract representation of addresses, but don't accept that same
    odd form of address as an LEA.

    This is the cause of PR96998. In the testcase given in the PR, combine
    produces:

    (insn 9 8 10 3 (set (mem:SI (plus:DI (sign_extract:DI (mult:DI (subreg:DI
(reg/v:SI 92 [ g ]) 0)
                            (const_int 4 [0x4]))
                        (const_int 34 [0x22])
                        (const_int 0 [0]))
                    (reg/f:DI 96)) [3 *i_5+0 S4 A32])
            (asm_operands:SI ("") ("=Q") 0 []
                 []
                 [] test.c:11)) "test.c":11:5 -1
         (expr_list:REG_DEAD (reg/v:SI 92 [ g ])
            (nil)))

    Then LRA reloads the address and we ICE because we fail to recognize the
    sign_extract outside the mem:

    (insn 33 8 34 3 (set (reg:DI 100)
            (sign_extract:DI (ashift:DI (subreg:DI (reg/v:SI 92 [ g ]) 0)
                    (const_int 2 [0x2]))
                (const_int 34 [0x22])
                (const_int 0 [0]))) "test.c":11:5 -1
         (nil))

    The aarch64 changes here remove the support for this sign_extract
    representation of addresses, fixing PR96998. Now this by itself would
    regress code quality, so this change is paired with an improvement to
    combine which prevents an extract rtx from being emitted in this case:
    we now write the rtx above as a shift of an extend, which allows the
    combination to go ahead.

    Prior to this, combine.c:make_extraction() identified where we can emit
    an ashift of an extend in place of an extraction, but failed to make the
    corresponding canonicalization/simplification when presented with a mult
    by a power of two. Such a representation is canonical when representing
    a left-shifted address inside a mem.

    This change remedies this situation. For rtxes such as:

    (mult:DI (subreg:DI (reg:SI r) 0) (const_int 2^n))

    where the bottom 32 + n bits are valid (the higher-order bits are
    undefined) and make_extraction() is being asked to sign_extract the
    lower (valid) bits, after the patch, we rewrite this as:

    (mult:DI (sign_extend:DI (reg:SI r)) (const_int 2^n))

    instead of using a sign_extract.

    gcc/ChangeLog:

            PR target/96998
            * combine.c (make_extraction): Also handle shifts written as
            (mult x 2^n), avoid creating an extract rtx for these.
            * config/aarch64/aarch64.c (aarch64_is_extend_from_extract):
Delete.
            (aarch64_classify_index): Remove extract-based address handling.
            (aarch64_strip_extend): Likewise.
            (aarch64_rtx_arith_op_extract_p): Likewise, remove now-unused
parameter.
            Update callers...
            (aarch64_rtx_costs): ... here.

    gcc/testsuite/ChangeLog:

            PR target/96998
            * gcc.c-torture/compile/pr96998.c: New test.

[Bug target/96998] GCC ICEs in on building AArch64 Linux kernel after basepoints/gcc-11-2903-g6b3034eaba83

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96998

Alex Coplan <acoplan at gcc dot gnu.org> changed:

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

--- Comment #13 from Alex Coplan <acoplan at gcc dot gnu.org> ---
Fixed. After the above commit I can now build a defconfig linux-next on AArch64
with trunk GCC.