Re: LTO with cross compiler and static linking

[Andrew Pinski <pinskia at gmail dot com>]

On Tue, Mar 19, 2013 at 3:30 AM, Sebastian Huber
<sebastian.huber@embedded-brains.de> wrote:
> Hello,
>
> I test currently the GCC 4.8 release candidate.  Some GCC LTO tests fail for
> me since the introduction of LTO in GCC.  I use Binutils 2.23.1 for PowerPC
> EABI (powerpc-rtems4.11 target) configured with:
>
> /scratch/archive/binutils-2.23.1/configure --target=powerpc-rtems4.11
> --prefix=/scratch/install-4.8.0-RC-20130316 --disable-werror --enable-lto
>
> For GCC I use:
>
> /scratch/archive/gcc-4.8.0-RC-20130316/configure --disable-werror
> --enable-threads=rtems --with-gnu-as --disable-multilib --enable-lto
> --enable-newlib-mb --enable-newlib-iconv --with-gnu-ld --with-newlib
> --verbose --with-system-zlib --disable-nls
> --enable-version-specific-runtime-libs --enable-languages=c,c++
> --target=powerpc-rtems4.11 --prefix=/scratch/install-4.8.0-RC-20130316


The easiest fix is inside the GCC testsuite itself.  In
testsuite/gcc.c-torture/execute/builtins/builtins.exp change:
if { [istarget *-*-eabi*]
     || [istarget *-*-elf] } {
   lappend additional_flags "-Wl,--allow-multiple-definition"
}
To:
if { [istarget *-*-eabi*]
     || [istarget *-*-elf]
     || [istarget *-*-rtems*]} {
   lappend additional_flags "-Wl,--allow-multiple-definition"
}

Thanks,
Andrew


>
> Several test fail like this:
>
> /scratch/b-powerpc-gcc/gcc/xgcc -B/scratch/b-powerpc-gcc/gcc/
> /scratch/archive/gcc-4.8.0-RC-20130316/gcc/testsuite/gcc.c-torture/execute/builtins/20010124-1.c
> /scratch/archive/gcc-4.8.0-RC-20130316/gcc/testsuite/gcc.c-torture/execute/builtins/20010124-1-lib.c
> /scratch/archive/gcc-4.8.0-RC-20130316/gcc/testsuite/gcc.c-torture/execute/builtins/lib/main.c
> gcc_tg.o -fno-diagnostics-show-caret -w -O2 -flto -fuse-linker-plugin
> -fno-fat-lto-objects -fno-tree-loop-distribute-patterns -DSTACK_SIZE=2048
> -isystem /scratch/b-powerpc-gcc/powerpc-rtems4.11/./newlib/targ-include
> -isystem /scratch/archive/gcc-4.8.0-RC-20130316/newlib/libc/include
> -B/scratch/install-4.8.0-RC-20130316/powerpc-rtems4.11/psim/lib/ -specs
> bsp_specs -qrtems -B/scratch/b-powerpc-gcc/powerpc-rtems4.11/./newlib/
> -L/scratch/b-powerpc-gcc/powerpc-rtems4.11/./newlib
> /scratch/b-powerpc-gcc/rtems_gcc_main.o -Wl,-wrap,exit -Wl,-wrap,_exit
> -Wl,-wrap,main -Wl,-wrap,abort -lm -o
> /scratch/b-powerpc-gcc/gcc/testsuite/gcc/20010124-1.x8 -Wl,-Map,map.txt
> -save-temps
> [Leaving LTRANS /tmp/cc0uaa6j.args]
> [Leaving LTRANS 20010124-1.x8.ltrans.out]
> [Leaving LTRANS /tmp/cccIkbEk.args]
> [Leaving LTRANS 20010124-1.x8.ltrans0.o]
> 20010124-1.x8.ltrans0.ltrans.o: In function `memcpy':
> 20010124-1.x8.ltrans0.o:(.text+0x4): multiple definition of `memcpy'
> /scratch/b-powerpc-gcc/powerpc-rtems4.11/./newlib/libc.a(lib_a-memcpy.o):/scratch/archive/gcc-4.8.0-RC-20130316/newlib/libc/string/memcpy.c:78:
> first defined here
> collect2: error: ld returned 1 exit status
>
> The specs file (bsp_specs) is this:
>
> %rename endfile old_endfile
> %rename startfile old_startfile
> %rename link old_link
>
> *startfile:
> %{!qrtems: %(old_startfile)} \
> %{!nostdlib: %{qrtems: ecrti%O%s rtems_crti%O%s crtbegin.o%s \
>     start.o%s -e _start -u __vectors}}
>
> *link:
> %{!qrtems: %(old_link)} %{qrtems: -dp -Bstatic}
>
> *endfile:
> %{!qrtems: %(old_endfile)} %{qrtems: crtend.o%s ecrtn.o%s}
>
> The start symbol _start will pull in a lot of code from the operating system
> which uses memcpy().  The file 20010124-1-lib.c specified at the command
> line above defines a global memcpy() function.  In the linker map file I see
> this:
>
> /scratch/b-powerpc-gcc/powerpc-rtems4.11/./newlib/libc.a(lib_a-memcpy.o)
>
> /scratch/install-4.8.0-RC-20130316/powerpc-rtems4.11/psim/lib/librtemsbsp.a(exc_bspsupport.o)
> (memcpy)
>
>  .text          0x00000000000001f0       0x8c
> /scratch/install-4.8.0-RC-20130316/powerpc-rtems4.11/psim/lib/start.o
>                 0x00000000000001f4                __rtems_entry_point
>                 0x00000000000001f4                _start
>                 0x0000000000000274                bsp_reset
>  .text          0x000000000000027c       0x88 20010124-1.x8.ltrans0.ltrans.o
>  .text.startup  0x0000000000000304       0x30 20010124-1.x8.ltrans0.ltrans.o
>                 0x0000000000000304                main
>  .text          0x0000000000000334      0x33c gcc_tg.o
>                 0x000000000000041c                __wrap_exit
>                 0x0000000000000504                __wrap__exit
>                 0x00000000000005f0                __wrap_abort
>                 0x0000000000000634                __wrap_main
>
> So the memcpy from libc is used instead of the one defined in
> 20010124-1-lib.c.
>
> If I execute this command line (no LTO switches):
>
> /scratch/git-rtems-testing/gcc/b-powerpc-gcc/gcc/xgcc
> -B/scratch/git-rtems-testing/gcc/b-powerpc-gcc/gcc/
> /home/sh/archive/gcc-4.8.0-RC-20130316/gcc/testsuite/gcc.c-torture/execute/builtins/20010124-1.c
> /home/sh/archive/gcc-4.8.0-RC-20130316/gcc/testsuite/gcc.c-torture/execute/builtins/20010124-1-lib.c
> /home/sh/archive/gcc-4.8.0-RC-20130316/gcc/testsuite/gcc.c-torture/execute/builtins/lib/main.c
> gcc_tg.o -fno-diagnostics-show-caret -w -O2 -DSTACK_SIZE=2048 -isystem
> /scratch/git-rtems-testing/gcc/b-powerpc-gcc/powerpc-rtems4.11/./newlib/targ-include
> -isystem /home/sh/archive/gcc-4.8.0-RC-20130316/newlib/libc/include
> -B/scratch/git-rtems-testing/gcc/install-4.8.0-RC-20130316/powerpc-rtems4.11/psim/lib/
> -specs bsp_specs -qrtems
> -B/scratch/git-rtems-testing/gcc/b-powerpc-gcc/powerpc-rtems4.11/./newlib/
> -L/scratch/git-rtems-testing/gcc/b-powerpc-gcc/powerpc-rtems4.11/./newlib
> /scratch/git-rtems-testing/gcc/b-powerpc-gcc/rtems_gcc_main.o -Wl,-wrap,exit
> -Wl,-wrap,_exit -Wl,-wrap,main -Wl,-wrap,abort -lm -o
> /scratch/git-rtems-testing/gcc/b-powerpc-gcc/gcc/testsuite/gcc/20010124-1.x8
> -Wl,-Map,map.txt -save-temps
>
> It links fine, and in the linker map file I see this:
>
>  .text          0x00000000000001f0       0x8c
> /scratch/install-4.8.0-RC-20130316/powerpc-rtems4.11/psim/lib/start.o
>                 0x00000000000001f4                __rtems_entry_point
>                 0x00000000000001f4                _start
>                 0x0000000000000274                bsp_reset
>  .text          0x000000000000027c       0x5c 20010124-1.o
>                 0x000000000000027c                main_test
>  .text          0x00000000000002d8       0xac 20010124-1-lib.o
>                 0x00000000000002d8                f
>                 0x00000000000002fc                g
>                 0x0000000000000300                memcpy
>  .text          0x0000000000000384        0x0 main.o
>  .text.startup  0x0000000000000384       0x44 main.o
>                 0x0000000000000384                main
>  .text          0x00000000000003c8      0x33c gcc_tg.o
>                 0x00000000000004b0                __wrap_exit
>                 0x0000000000000598                __wrap__exit
>                 0x0000000000000684                __wrap_abort
>                 0x00000000000006c8                __wrap_main
>
> There is no memcpy from libc in contrast to the original GCC test.
>
> I have this problem on other targets as well.  Has someone a hint what can
> be wrong here since I am quite clueless.
>
> --
> Sebastian Huber, embedded brains GmbH
>
> Address : Dornierstr. 4, D-82178 Puchheim, Germany
> Phone   : +49 89 189 47 41-16
> Fax     : +49 89 189 47 41-09
> E-Mail  : sebastian.huber@embedded-brains.de
> PGP     : Public key available on request.
>
> Diese Nachricht ist keine geschÃftliche Mitteilung im Sinne des EHUG.