[Bug ada/103785] New: [12 Regression] Ada LTO+PGO bootstrap ICEs on i586

[Bug ada/103785] New: [12 Regression] Ada LTO+PGO bootstrap ICEs on i586

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

            Bug ID: 103785
           Summary: [12 Regression] Ada LTO+PGO bootstrap ICEs on i586
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: ada
          Assignee: unassigned at gcc dot gnu.org
          Reporter: marxin at gcc dot gnu.org
  Target Milestone: ---

The bootstrap ends with:

[ 4274s] +===========================GNAT BUG
DETECTED==============================+
[ 4274s] | 12.0.0 20211220 (experimental) [revision
b3f58f87d78b958e35e4a44f5fdb4b7721cb2837] (i586-suse-linux) |
[ 4274s] | Storage_Error stack overflow or erroneous memory access             
    |
[ 4274s] | Error detected at system.ads:106:30                                 
    |
[ 4274s] | Compiling ../../gcc/ada/ali.adb                                     
    |
[ 4274s] | Please submit a bug report; see https://gcc.gnu.org/bugs/ .         
    |
[ 4274s] | Use a subject line meaningful to you and us to track the bug.       
    |
[ 4274s] | Include the entire contents of this bug box in the report.          
    |
[ 4274s] | Include the exact command that you entered.                         
    |
[ 4274s] | Also include sources listed below.                                  
    |
[ 4274s]
+==========================================================================+

g:8d1e342b4afbad77cb92f8057cf6d1e996bfb077 is fine, while
g:b3f58f87d78b958e35e4a44f5fdb4b7721cb2837 fails.

[Bug ada/103785] [12 Regression] Ada LTO+PGO bootstrap ICEs on i586

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-12-21
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
                 CC|                            |roger at nextmovesoftware dot com
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=103773

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
So very likely caused by r12-6077-gc9c466ea33dd2110fd9b73b05c30dacabff8f85a.

[Bug target/103785] [12 Regression] Ada LTO+PGO bootstrap ICEs on i586

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |build, ice-on-valid-code,
                   |                            |lto
          Component|ada                         |target
   Target Milestone|---                         |12.0
           Severity|normal                      |blocker

[Bug target/103785] [12 Regression] Ada LTO+PGO bootstrap ICEs on i586

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

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

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

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 103787 has been marked as a duplicate of this bug. ***

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|lto                         |

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
As reported in PR 103787, PGO and LTO is not needed.

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[roger at nextmovesoftware dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

--- Comment #4 from Roger Sayle <roger at nextmovesoftware dot com> ---
This is either (i) related to PR 103773, where GNAT is somehow setting
optimize_size to a value greater than 1 (i.e. -Oz) and getting hit by the known
memory corruption or (ii) somehow related to the highpart multiplication
changes (which on double checking look sound) or (iii) caused by someone elses
changes.
If we could identify which patch to revert to resolve the issue, or even which
hunk, that would help.  I'm attempting various gnat bootstraps myself.

Sorry for the inconvenience.

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hjl.tools at gmail dot com

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 103820 has been marked as a duplicate of this bug. ***

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ubizjak at gmail dot com

--- Comment #6 from H.J. Lu <hjl.tools at gmail dot com> ---
A patch is posted at

https://gcc.gnu.org/pipermail/gcc-patches/2021-December/587326.html

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

--- Comment #7 from H.J. Lu <hjl.tools at gmail dot com> ---
sem_type.adb is miscompiled.

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

--- Comment #8 from H.J. Lu <hjl.tools at gmail dot com> ---
Created attachment 52055
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=52055&action=edit
Good and bad asm

--- good.s      2021-12-24 11:24:26.531365375 -0800
+++ bad.s       2021-12-24 11:24:30.769344666 -0800
@@ -10,7 +10,6 @@ sem_type__interp_map__setX:
        .cfi_def_cfa_offset 8
        .cfi_offset 7, -8
        movl    %edx, %edi
-       movl    $714200473, %edx
        pushl   %esi
        .cfi_def_cfa_offset 12
        .cfi_offset 6, -12
@@ -18,7 +17,8 @@ sem_type__interp_map__setX:
        .cfi_def_cfa_offset 16
        .cfi_offset 3, -16
        movl    %eax, %ebx
-       imull   %edx
+       movl    $714200473, %eax
+       imull   %eax
        movl    %ebx, %eax
        sarl    $31, %eax
        sarl    $9, %edx
@@ -2161,9 +2161,8 @@ sem_type__get_first_interp:
        cmpb    $71, %al
        je      .L242
 .L226:
-       movl    %ebx, %eax
-       movl    $714200473, %edx
-       imull   %edx
+       movl    $714200473, %eax
+       imull   %ebx
        movl    %ebx, %eax
        sarl    $31, %eax
        sarl    $9, %edx
@@ -10929,9 +10928,8 @@ sem_type__save_interps:
        cmpb    $71, %al
        je      .L1375
 .L1358:
-       movl    %ebx, %eax
-       movl    $714200473, %edx
-       imull   %edx
+       movl    $714200473, %eax
+       imull   %ebx
        movl    %ebx, %eax
        sarl    $31, %eax
        sarl    $9, %edx

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

--- Comment #9 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to H.J. Lu from comment #8)
> Created attachment 52055 [details]
> Good and bad asm
> 
> --- good.s	2021-12-24 11:24:26.531365375 -0800
> +++ bad.s	2021-12-24 11:24:30.769344666 -0800
> @@ -10,7 +10,6 @@ sem_type__interp_map__setX:
>  	.cfi_def_cfa_offset 8
>  	.cfi_offset 7, -8
>  	movl	%edx, %edi
> -	movl	$714200473, %edx
>  	pushl	%esi
>  	.cfi_def_cfa_offset 12
>  	.cfi_offset 6, -12
> @@ -18,7 +17,8 @@ sem_type__interp_map__setX:
>  	.cfi_def_cfa_offset 16
>  	.cfi_offset 3, -16
>  	movl	%eax, %ebx
> -	imull	%edx
> +	movl	$714200473, %eax

This is wrong since EAX is an implicit input operand.

> +	imull	%eax
>  	movl	%ebx, %eax
>  	sarl	$31, %eax
>  	sarl	$9, %edx

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

--- Comment #10 from H.J. Lu <hjl.tools at gmail dot com> ---
Before

(insn 8 5 70 2 (set (reg:SI 1 dx [92])
        (const_int 714200473 [0x2a91d599]))
"/export/gnu/import/git/gitlab/x86-gcc-test/gcc/ada/sem_type.adb":2563:7 70
{*movsi_internal}
     (expr_list:REG_EQUIV (const_int 714200473 [0x2a91d599])
        (nil)))
(insn 70 8 9 2 (set (reg:SI 0 ax [105])
        (reg:SI 0 ax [orig:89 k ] [89]))
"/export/gnu/import/git/gitlab/x86-gcc-test/gcc/ada/sem_type.adb":2563:7 70
{*movsi_internal}
     (nil))
(insn 9 70 10 2 (parallel [
            (set (reg:SI 1 dx [91])
                (smul_highpart:SI (reg:SI 0 ax [105])
                    (reg:SI 1 dx [92])))
            (clobber (reg:SI 0 ax [105]))
            (clobber (reg:CC 17 flags))
        ])
"/export/gnu/import/git/gitlab/x86-gcc-test/gcc/ada/sem_type.adb":2563:7 383
{smulsi3_highpart}
     (expr_list:REG_DEAD (reg:SI 0 ax [105])
        (expr_list:REG_UNUSED (reg:CC 17 flags)
            (expr_list:REG_UNUSED (reg:SI 0 ax [105])
                (expr_list:REG_EQUAL (smul_highpart:SI (reg/v:SI 3 bx [orig:89
k ] [89])
                        (const_int 714200473 [0x2a91d599]))
                    (nil))))))

After

(insn 95 5 96 2 (set (reg:SI 0 ax [105])
        (const_int 714200473 [0x2a91d599]))
"/export/gnu/import/git/gitlab/x86-gcc-test/gcc/ada/sem_type.adb":2563:7 -1
     (nil))
(insn 96 95 10 2 (parallel [
            (set (reg:SI 1 dx [91])
                (smul_highpart:SI (reg:SI 0 ax [105])
                    (reg:SI 0 ax [orig:89 k ] [89])))
            (clobber (reg:SI 0 ax [105]))
            (clobber (reg:CC 17 flags))
        ])
"/export/gnu/import/git/gitlab/x86-gcc-test/gcc/ada/sem_type.adb":2563:7 -1
     (nil))

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

--- Comment #11 from H.J. Lu <hjl.tools at gmail dot com> ---
The v3 patch is posted at

https://gcc.gnu.org/pipermail/gcc-patches/2021-December/587364.html

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

--- Comment #12 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by H.J. Lu <hjl@gcc.gnu.org>:

https://gcc.gnu.org/g:d87483015d476a95f521f0c9dcf6988ca889063b

commit r12-6118-gd87483015d476a95f521f0c9dcf6988ca889063b
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Fri Dec 24 12:50:21 2021 -0800

    i386: Check AX input in any_mul_highpart peepholes

    When applying peephole optimization to transform

            mov imm, %reg0
            mov %reg1, %AX_REG
            imul %reg0

    to

            mov imm, %AX_REG
            imul %reg1

    disable peephole optimization if reg1 == AX_REG.

    gcc/

            PR target/103785
            * config/i386/i386.md: Swap operand order in comments and check
            AX input in any_mul_highpart peepholes.

    gcc/testsuite/

            PR target/103785
            * gcc.target/i386/pr103785.c: New test.

[Bug target/103785] [12 Regression] Ada bootstrap ICEs on i?86

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103785

H.J. Lu <hjl.tools at gmail dot com> changed:

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

--- Comment #13 from H.J. Lu <hjl.tools at gmail dot com> ---
Fixed.