[Bug target/113156] New: AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[Bug target/113156] New: AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[dimitar at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

            Bug ID: 113156
           Summary: AVR build broken due to ICE while compiling libgcc,
                    started with r14-6201-gf0a90c7d7333fc
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dimitar at gcc dot gnu.org
  Target Milestone: ---

AVR target build is broken after r14-6201-gf0a90c7d7333fc :

/mnt/nvme/dinux/local-workspace/gcc/libgcc/strub.c:91:1: internal compiler
error: 'global_options' are modified in local context
   91 | {
      | ^
0xda0b43 cl_optimization_compare(gcc_options*, gcc_options*)
        /mnt/nvme/dinux/local-workspace/avr-gcc-build/gcc/options-save.cc:13414
0x94230c handle_optimize_attribute
        /mnt/nvme/dinux/local-workspace/gcc/gcc/c-family/c-attribs.cc:5972
0x8224a4 decl_attributes(tree_node**, tree_node*, int, tree_node*)
        /mnt/nvme/dinux/local-workspace/gcc/gcc/attribs.cc:891
0x828bad c_decl_attributes
        /mnt/nvme/dinux/local-workspace/gcc/gcc/c/c-decl.cc:5425
0x840acb start_function(c_declspecs*, c_declarator*, tree_node*)
        /mnt/nvme/dinux/local-workspace/gcc/gcc/c/c-decl.cc:10255
0x8ab355 c_parser_declaration_or_fndef
        /mnt/nvme/dinux/local-workspace/gcc/gcc/c/c-parser.cc:2912
0x8b663b c_parser_external_declaration
        /mnt/nvme/dinux/local-workspace/gcc/gcc/c/c-parser.cc:2046
0x8b7023 c_parser_translation_unit
        /mnt/nvme/dinux/local-workspace/gcc/gcc/c/c-parser.cc:1900
0x8b7023 c_parse_file()
        /mnt/nvme/dinux/local-workspace/gcc/gcc/c/c-parser.cc:26713
0x929101 c_common_parse_file()
        /mnt/nvme/dinux/local-workspace/gcc/gcc/c-family/c-opts.cc:1301
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.
make[4]: *** [/mnt/nvme/dinux/local-workspace/gcc/libgcc/static-object.mk:17:
strub.o] Error 1

[Bug target/113156] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Most like __optimize__ ("-fno-tree-loop-distribute-patterns") was not working
for avr before.

[Bug target/113156] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=105069

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Related to the sh issue pr 105069.

[Bug target/113156] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[paulbendixen at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

Paul M. Bendixen <paulbendixen at gmail dot com> changed:

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

--- Comment #3 from Paul M. Bendixen <paulbendixen at gmail dot com> ---
I can confirm this also happens for me.

I have found the same commit to be the problem and have tried with both gcc 11
and 12 
    gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
    gcc-12 (Ubuntu 12.3.0-1ubuntu1~22.04) 12.3.0

For reproducability it was configured using:
../configure --prefix=$PREFIX --target=avr --enable-languages=c,c++
--disable-nls --disable-libssp --with-dwarf2

The error only seems to happen when compiling the long64 part of the multilib
(or at least first), the parts compiled for avrXX and xmegaX targets run fine.

[Bug target/113156] [14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |14.0
                 CC|                            |aoliva at gcc dot gnu.org
           Keywords|                            |build
            Summary|AVR build broken due to ICE |[14 Regression] AVR build
                   |while compiling libgcc,     |broken due to ICE while
                   |started with                |compiling libgcc, started
                   |r14-6201-gf0a90c7d7333fc    |with
                   |                            |r14-6201-gf0a90c7d7333fc
           Priority|P3                          |P1

[Bug target/113156] [14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 57020
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57020&action=edit
preprocessed source

Exact command line:
```
/home/apinski/src/upstream-full-cross/gcc/objdir-avrr/./gcc/xgcc
-B/home/apinski/src/upstream-full-cross/gcc/objdir-avrr/./gcc/
-B/home/apinski/src/upstream-full-cross/gcc/objdir-avrr/../../install-avr/avr/bin/
-B/home/apinski/src/upstream-full-cross/gcc/objdir-avrr/../../install-avr/avr/lib/
-isystem
/home/apinski/src/upstream-full-cross/gcc/objdir-avrr/../../install-avr/avr/include
-isystem
/home/apinski/src/upstream-full-cross/gcc/objdir-avrr/../../install-avr/avr/sys-include
   -g -O2 -mdouble=64 -O2  -g -O2 -DIN_GCC -DCROSS_DIRECTORY_STRUCTURE  -W
-Wall -Wno-narrowing -Wwrite-strings -Wcast-qual -Wstrict-prototypes
-Wmissing-prototypes -Wold-style-definition  -isystem ./include  -Dinhibit_libc
-mcall-prologues -Os -g -DIN_LIBGCC2 -fbuilding-libgcc -fno-stack-protector
-Dinhibit_libc -Dinhibit_libc -mcall-prologues -Os -I. -I. -I../../.././gcc
-I../../../../libgcc -I../../../../libgcc/. -I../../../../libgcc/../gcc
-I../../../../libgcc/../include  -DHAVE_CC_TLS -DUSE_EMUTLS  -o strub.o -MT
strub.o -MD -MP -MF strub.dep  -c ../../../../libgcc/strub.c
-fvisibility=hidden -DHIDE_EXPORTS

```

[Bug target/113156] [14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2024-01-10
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
           Keywords|                            |ice-checking
      Known to fail|                            |13.1.0

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Reduced testcase:
```
void
__attribute__ ((__optimize__ ("-fno-tree-loop-distribute-patterns")))
a (void) {}
```

Just need `-mdouble=64` to cause the ICE.

Also ICEs in GCC 13 too; just the source in libgcc is new.

Confirmed.

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |11.1.0
      Known to work|                            |10.1.0
            Summary|[14 Regression] AVR build   |[11/12/13/14 Regression]
                   |broken due to ICE while     |AVR build broken due to ICE
                   |compiling libgcc, started   |while compiling libgcc,
                   |with                        |started with
                   |r14-6201-gf0a90c7d7333fc    |r14-6201-gf0a90c7d7333fc

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The testcase fails with GCC 11+ with -fchecking .

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 57021
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57021&action=edit
Patch which seems to fix it

I cannot test this patch fully but this seems to fix it.


Signed-off-by: Andrew Pinski <quic_apinski@quicinc.com>

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[gjl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #8 from Georg-Johann Lay <gjl at gcc dot gnu.org> ---
Is there as comprehensible explanation when option property "Save" is needed?

The internals just state

> Build the cl_target_option structure to hold a copy of the option,
> add the functions cl_target_option_save and cl_target_option_restore
> to save and restore the options.

Which explains what is does, but not when and why it is needed.

For example, -mdouble= and -mlong-double= are multilib options, and not
optimizatoin options, and they whould never change during a program.

Why is "Save" needed for some multilib options and not for others?

Isn't it even a bug to allow multilib options to be changed due to optimization
flags?  A code that uses -mdouble=32 in some places and -mdouble=64 is others
is bogus.

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Maybe the case is mdouble=64 is not being set correctly after the __optimize__
attribute.  If I get some more time, I will look into what other targets do for
these kind of options.

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[gjl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #10 from Georg-Johann Lay <gjl at gcc dot gnu.org> ---
-mdouble and -mlong-double are more complicated than other options because they
depend on each other in way that (to my knowledge) cannot be described by
option properties. For example, in

-mdouble=64 -mlong-double=32

the second option will trigger -mdouble=32 in order to maintain double <= long
double.

This logic is implemented in avr-common.cc::avr_handle_option(), and maybe
that's the reason "Save" is needed?

Anyway, when the actions of avr_handle_option are changed by some attribue
handling, then the result will be wong code.  For example, when the above
options, for whatever reasons, would lead to double=64 longdouble=32 at some
point, wrong code will be generated.

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #11 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So the problem is avr_option_optimization_table sets the default
mlong-double/mdouble size but if you don't have save on the option, it changes
back to the default which then errors out.

Other targets handle setting the default like this differently, rs6000 handles
the default `long double` in some interesting fashion.

The patch is semi-correct but I need to make sure we error out if used with the
target/optimization attribute (I suspect we do).

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #12 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Georg-Johann Lay <gjl@gcc.gnu.org>:

https://gcc.gnu.org/g:f1b0b1e4454d160564090a4cc1fa649ccefdcb1e

commit r14-7240-gf1b0b1e4454d160564090a4cc1fa649ccefdcb1e
Author: Andrew Pinski <quic_apinski@quicinc.com>
Date:   Mon Jan 15 10:31:36 2024 +0100

    AVR: target/113156 - Fix ICE due to missing "Save" on -m[long-]double=
options.

            Multilib options -mdouble= and -mlong-double= are not orthogonal:
            TARGET_HANDLE_OPTION = avr-common.cc::avr_handle_option() sets them
            such that  sizeof(double) <= sizeof(long double)  is always true.

    gcc/
            PR target/113156
            * config/avr/avr.opt (-mdouble, -mlong-double): Add "Save" flag.
            (-mbranch-cost): Set "Optimization" flag.

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #13 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-13 branch has been updated by Georg-Johann Lay
<gjl@gcc.gnu.org>:

https://gcc.gnu.org/g:d73e585c72cd9c47992185a83fdd9bd28347029a

commit r13-8223-gd73e585c72cd9c47992185a83fdd9bd28347029a
Author: Andrew Pinski <quic_apinski@quicinc.com>
Date:   Mon Jan 15 10:31:36 2024 +0100

    AVR: target/113156 - Fix ICE due to missing "Save" on -m[long-]double=
options.

            Multilib options -mdouble= and -mlong-double= are not orthogonal:
            TARGET_HANDLE_OPTION = avr-common.cc::avr_handle_option() sets them
            such that  sizeof(double) <= sizeof(long double)  is always true.

    gcc/
            PR target/113156
            * config/avr/avr.opt (-mdouble, -mlong-double): Add "Save" flag.
            (-mbranch-cost): Set "Optimization" flag.

    (cherry picked from commit f1b0b1e4454d160564090a4cc1fa649ccefdcb1e)

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #14 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Georg-Johann Lay
<gjl@gcc.gnu.org>:

https://gcc.gnu.org/g:2bce77d6e17efd801b2a6fee72e511c0597dc5c2

commit r12-10097-g2bce77d6e17efd801b2a6fee72e511c0597dc5c2
Author: Andrew Pinski <quic_apinski@quicinc.com>
Date:   Mon Jan 15 10:31:36 2024 +0100

    AVR: target/113156 - Fix ICE due to missing "Save" on -m[long-]double=
options.

            Multilib options -mdouble= and -mlong-double= are not orthogonal:
            TARGET_HANDLE_OPTION = avr-common.cc::avr_handle_option() sets them
            such that  sizeof(double) <= sizeof(long double)  is always true.

    gcc/
            PR target/113156
            * config/avr/avr.opt (-mdouble, -mlong-double): Add "Save" flag.
            (-mbranch-cost): Set "Optimization" flag.

    (cherry picked from commit f1b0b1e4454d160564090a4cc1fa649ccefdcb1e)

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[gjl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

Georg-Johann Lay <gjl at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |opt-attribute

--- Comment #15 from Georg-Johann Lay <gjl at gcc dot gnu.org> ---
Should work now on v12.4+ and v13.3+

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|14.0                        |12.4
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #16 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed.

[Bug target/113156] [11/12/13/14 Regression] AVR build broken due to ICE while compiling libgcc, started with r14-6201-gf0a90c7d7333fc

[gjl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113156

--- Comment #17 from Georg-Johann Lay <gjl at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #11)
> The patch is semi-correct

Would you detail on this? What parts for a complete fix are (still) missing?