[Bug bootstrap/62005] New: [4.10 regression] with --enable-checking=rtl : loop-unroll.c:2095:10: error: function may return address of local variable

[Bug bootstrap/62005] New: [4.10 regression] with --enable-checking=rtl : loop-unroll.c:2095:10: error: function may return address of local variable

[dimhen at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62005

            Bug ID: 62005
           Summary: [4.10 regression] with --enable-checking=rtl :
                    loop-unroll.c:2095:10: error: function may return
                    address of local variable
           Product: gcc
           Version: 4.10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dimhen at gmail dot com

r213523, r213529 FAIL
r213316 PASS

~/src/gcc_r213529/configure --enable-checking=rtl
make
[...]
make[3]: Entering directory `/home/dimhen/build/gcc_r213529_rtl/gcc'
/home/dimhen/build/gcc_r213529_rtl/./prev-gcc/xg++
-B/home/dimhen/build/gcc_r213529_rtl/./prev-gcc/
-B/usr/local/x86_64-unknown-linux-gnu/bin/ -nostdinc++
-B/home/dimhen/build/gcc_r213529_rtl/prev-x86_64-unknown-linux-gnu/libstdc++-v3/src/.libs
-B/home/dimhen/build/gcc_r213529_rtl/prev-x86_64-unknown-linux-gnu/libstdc++-v3/libsupc++/.libs

-I/home/dimhen/build/gcc_r213529_rtl/prev-x86_64-unknown-linux-gnu/libstdc++-v3/include/x86_64-unknown-linux-gnu

-I/home/dimhen/build/gcc_r213529_rtl/prev-x86_64-unknown-linux-gnu/libstdc++-v3/include
 -I/home/dimhen/src/gcc_r213529/libstdc++-v3/libsupc++
-L/home/dimhen/build/gcc_r213529_rtl/prev-x86_64-unknown-linux-gnu/libstdc++-v3/src/.libs
-L/home/dimhen/build/gcc_r213529_rtl/prev-x86_64-unknown-linux-gnu/libstdc++-v3/libsupc++/.libs
-c   -g -O2 -gtoggle -DIN_GCC    -fno-exceptions -fno-rtti
-fasynchronous-unwind-tables -W -Wall -Wno-narrowing -Wwrite-strings
-Wcast-qual -Wmissing-format-attribute -Woverloaded-virtual -pedantic
-Wno-long-long -Wno-variadic-macros -Wno-overlength-strings -Werror  
-DHAVE_CONFIG_H -I. -I. -I/home/dimhen/src/gcc_r213529/gcc
-I/home/dimhen/src/gcc_r213529/gcc/.
-I/home/dimhen/src/gcc_r213529/gcc/../include
-I/home/dimhen/src/gcc_r213529/gcc/../libcpp/include
-I/home/dimhen/build/gcc_r213529_rtl/./gmp -I/home/dimhen/src/gcc_r213529/gmp
-I/home/dimhen/build/gcc_r213529_rtl/./mpfr -I/home/dimhen/src/gcc_r213529/mpfr
-I/home/dimhen/src/gcc_r213529/mpc/src 
-I/home/dimhen/src/gcc_r213529/gcc/../libdecnumber
-I/home/dimhen/src/gcc_r213529/gcc/../libdecnumber/bid -I../libdecnumber
-I/home/dimhen/src/gcc_r213529/gcc/../libbacktrace -DCLOOG_INT_GMP
-I/home/dimhen/build/gcc_r213529_rtl/./cloog/include
-I/home/dimhen/src/gcc_r213529/cloog/include
-I/home/dimhen/build/gcc_r213529_rtl/./isl/include
-I/home/dimhen/src/gcc_r213529/isl/include  -o loop-unroll.o -MT loop-unroll.o
-MMD -MP -MF ./.deps/loop-unroll.TPo
/home/dimhen/src/gcc_r213529/gcc/loop-unroll.c
/home/dimhen/src/gcc_r213529/gcc/loop-unroll.c: In function 'rtx_def**
get_ivts_expr(rtx, iv_to_split*)':
/home/dimhen/src/gcc_r213529/gcc/loop-unroll.c:2095:10: error: function may
return address of local variable [-Werror=return-local-addr]
   return ret;
          ^
/home/dimhen/src/gcc_r213529/gcc/loop-unroll.c:2087:20: note: declared here
 get_ivts_expr (rtx expr, struct iv_to_split *ivts)
                    ^
cc1plus: all warnings being treated as errors
make[3]: *** [loop-unroll.o] Error 1
make[3]: Leaving directory `/home/dimhen/build/gcc_r213529_rtl/gcc'

[Bug bootstrap/62005] [4.10 regression] with --enable-checking=rtl : loop-unroll.c:2095:10: error: function may return address of local variable

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62005

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |glisse at gcc dot gnu.org
   Target Milestone|---                         |4.10.0

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, the warning is true if ivts->n_loc == 0.  Marc?

[Bug bootstrap/62005] [4.10 regression] with --enable-checking=rtl : loop-unroll.c:2095:10: error: function may return address of local variable

[glisse at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62005

--- Comment #2 from Marc Glisse <glisse at gcc dot gnu.org> ---
Ah, I didn't test with --enable-ckecking=rtl.

We could split the "maybe" part of the warning and downgrade it to Wextra, but
I'd rather keep it at least in Wall, which means we need to change the
loop-unroll code anyway.

The code is not easy to understand. For instance, I don't see where n_loc can
be set to any value other than 1? That would make it easy to simplify this
loop...

[Bug bootstrap/62005] [4.10 regression] with --enable-checking=rtl : loop-unroll.c:2095:10: error: function may return address of local variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62005

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

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

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I've posted a loop-unroll.c fix.

That said, an interesting question is why we don't warn about this without rtl
checking, the function still returns address of the parameter.  Is that because
it is inlined in that case and so we don't warn then?

[Bug bootstrap/62005] [4.10 regression] with --enable-checking=rtl : loop-unroll.c:2095:10: error: function may return address of local variable

[glisse at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62005

--- Comment #4 from Marc Glisse <glisse at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #3)
> That said, an interesting question is why we don't warn about this without
> rtl checking, the function still returns address of the parameter.  Is that
> because it is inlined in that case and so we don't warn then?

That seems likely, RTL checking may make the function large enough not to be
inlined. Compiling without RTL checking but with -fkeep-inline-functions
probably warns as well then.

[Bug bootstrap/62005] [4.10 regression] with --enable-checking=rtl : loop-unroll.c:2095:10: error: function may return address of local variable

[glisse at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62005

--- Comment #5 from Marc Glisse <glisse at gcc dot gnu.org> ---
(In reply to Marc Glisse from comment #4)
> Compiling without RTL checking but with -fkeep-inline-functions
> probably warns as well then.

Uh, for some reason -fkeep-inline-functions does not preserve static functions,
but it does preserve static inline functions (so it warns if I add "inline" on
the definition of get_ivts_expr). I couldn't find a -fkeep-static-functions.

[Bug bootstrap/62005] [4.10 regression] with --enable-checking=rtl : loop-unroll.c:2095:10: error: function may return address of local variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62005

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

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

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Tue Aug  5 08:09:00 2014
New Revision: 213621

URL: https://gcc.gnu.org/viewcvs?rev=213621&root=gcc&view=rev
Log:
    * loop-unroll.c (struct iv_to_split): Remove n_loc and loc fields.
    (analyze_iv_to_split_insn): Don't initialize them.
    (get_ivts_expr): Removed.
    (allocate_basic_variable, insert_base_initialization): Use
    SET_SRC instead of *get_ivts_expr.
    (split_iv): Use &SET_SRC instead of get_ivts_expr.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/loop-unroll.c