[Bug target/104014] New: [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[Bug target/104014] New: [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

            Bug ID: 104014
           Summary: [12 Regression] r12-6538 breaks bootstrap with
                    --with-arch=native --with-cpu=native
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: haochen.jiang at intel dot com
          Reporter: hjl.tools at gmail dot com
                CC: crazylht at gmail dot com, haochen.jiang at intel dot com
  Target Milestone: ---
            Target: x86-64

On x86-64, r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native:

../../src-master/libdecnumber/decContext.c: In function
??decContextRestoreStatus??:
../../src-master/libdecnumber/decContext.c:183:3: error: unrecognizable insn: 
  183 |   } /* decContextRestoreStatus */
      |   ^ 
(insn 25 24 17 2 (parallel [
            (set (mem:SI (plus:DI (reg/v/f:DI 87 [ context ])
                        (const_int 20 [0x14])) [1 context_7(D)->status+0 S4
A32])
                (ior:SI (reg:SI 97)
                    (reg:SI 98)))
            (clobber (reg:CC 17 flags))
        ]) "../../src-master/libdecnumber/decContext.c":181:18 -1
     (expr_list:REG_DEAD (reg:SI 98)
        (expr_list:REG_DEAD (reg:SI 97)
            (expr_list:REG_UNUSED (reg:CC 17 flags)
                (nil)))))
during RTL pass: ira

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.0

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2022-01-13
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

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

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

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
Is it the same as PR104001?

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Martin Liška from comment #1)
> Is it the same as PR104001?

Maybe. I don't know if the fix is complete.

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING

--- Comment #3 from Martin Liška <marxin at gcc dot gnu.org> ---
Then please provide a test-case and options.

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

David Binderman <dcb314 at hotmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dcb314 at hotmail dot com

--- Comment #4 from David Binderman <dcb314 at hotmail dot com> ---
(In reply to H.J. Lu from comment #0)
> On x86-64, r12-6538 breaks bootstrap with --with-arch=native

Perhaps it is time to introduce some additional requirements for 
candidate patches.

Anything that breaks bootstrap prevents other developers making progress.
I would have thought that bootstrap-O3 would be a minimal requirement
for any patch.

The protocol is published at:

https://gcc.gnu.org/contribute.html#testing

In practice I use this:

CC="clang -Wall" CXX="clang++ -Wall" \
../trunk.git/configure --prefix=/home/dcb/gcc/$PREFIX \
        --disable-multilib \
        --disable-werror \
        --with-pkgversion=$HASH \
        --enable-checking=df,extra,fold,rtl,yes \
        --enable-languages=c,c++,fortran

sed 's/-O2/-O3 -march=native/' < Makefile > Makefile.tmp
diff Makefile Makefile.tmp
mv Makefile.tmp Makefile

This works well for me.

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

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> ---
That makes no sense.  O2-ish bootstrap are the most commonly used ones, not
O3-ish, so mandating bootstrap-O3 for testing is wrong.
And, -march=native is a moving target, different for each developer.  So
bootstrap that works for one developer with -march=native can work fine while
on another box it will not.
--with-arch=native --with-cpu=native bootstraps aren't a good idea for
gcc-testresults posts as well because it is hard to determine what exactly that
native expands to.

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

--- Comment #6 from David Binderman <dcb314 at hotmail dot com> ---
(In reply to Jakub Jelinek from comment #5)
> That makes no sense.  

Surprising. 

bootstrap-O3 tests more of the compiler than the ordinary bootstrap-O2 does. 
That has to be a good thing for any candidate patch.

> And, -march=native is a moving target, different for each developer.  

Indeed it is. We all have different boxes. But adding -march=native
tests more of the compiler. 

> So bootstrap that works for one developer with -march=native can work fine
> while on another box it will not.

Agreed. But if a native bootstrap works for some machine,
we can be pretty sure it will work for all predecessor machines
of the same architecture.

> --with-arch=native --with-cpu=native bootstraps aren't a good idea for
> gcc-testresults posts as well because it is hard to determine what exactly
> that native expands to.

It wouldn't be impossible to require folks to specify what their native
target was.

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to David Binderman from comment #6)
> (In reply to Jakub Jelinek from comment #5)
> > That makes no sense.  
> 
> Surprising. 
> 
> bootstrap-O3 tests more of the compiler than the ordinary bootstrap-O2 does.

Not really.  It tests different code paths in the compiler. 

> > And, -march=native is a moving target, different for each developer.  
> 
> Indeed it is. We all have different boxes. But adding -march=native
> tests more of the compiler. 

Not really.

> Agreed. But if a native bootstrap works for some machine,
> we can be pretty sure it will work for all predecessor machines
> of the same architecture.

Certainly not.
> 
> > --with-arch=native --with-cpu=native bootstraps aren't a good idea for
> > gcc-testresults posts as well because it is hard to determine what exactly
> > that native expands to.
> 
> It wouldn't be impossible to require folks to specify what their native
> target was.

>From my experience, that is a total nightmare.  People usually say in such case
if they specify anything I'm using a Skylake or something similar, in better
cases Intel i9-7680X or whatever they have.  But mapping that to the actual
-march= -misa1 -misa2 etc. options that the -march=native expands to is hard,
one needs to look up what model/stepping such CPU has somewhere and exact set
of ISAs it supports and then try to repeat what driver-*.c does manually.
Sure, one can use gcc -v and paste the cc1 etc. invocation line, but people
just don't do that, so it is another step where you need to ask them.
So it is much better if people actually test --with-arch=skylake-avx512 or
whatever exact -march= they have, that isn't a moving target.

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

--- Comment #8 from Hongtao.liu <crazylht at gmail dot com> ---
Same issue as PR104001

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

--- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by hongtao Liu <liuhongt@gcc.gnu.org>:

https://gcc.gnu.org/g:b77e3b4e4589e56c01511fabdbaadb029cd47f5c

commit r12-6567-gb77e3b4e4589e56c01511fabdbaadb029cd47f5c
Author: liuhongt <hongtao.liu@intel.com>
Date:   Thu Jan 13 22:51:49 2022 +0800

    Fix ICE of unrecognizable insn. [PR target/104001]

    For define_insn_and_split "*xor2andn":

    1. Refine predicate of operands[0] from nonimmediate_operand to
    register_operand.
    2. Remove TARGET_AVX512BW from condition to avoid kmov when TARGET_BMI
    is not available.

    gcc/ChangeLog:

            PR target/104001
            PR target/94790
            PR target/104014
            * config/i386/i386.md (*xor2andn): Refine predicate of
            operands[0] from nonimmediate_operand to
            register_operand, remove TARGET_AVX512BW from condition.

    gcc/testsuite/ChangeLog:

            * gcc.target/i386/pr104001.c: New test.

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

--- Comment #10 from Hongtao.liu <crazylht at gmail dot com> ---
Fixed.

[Bug target/104014] [12 Regression] r12-6538 breaks bootstrap with --with-arch=native --with-cpu=native

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104014

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

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

--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
.