X-Authentication-Warning: delorie.com: mail set sender to djgpp-workers-bounces using -f
X-Recipient: djgpp-workers AT delorie DOT com
Message-ID: <529E1885.9020804@iki.fi>
Date: Tue, 03 Dec 2013 19:44:37 +0200
From: Andris Pavenis <andris DOT pavenis AT iki DOT fi>
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.1.0
MIME-Version: 1.0
To: djgpp-workers AT delorie DOT com
Subject: Re: Binutils and LTO in GCC
References: <5290666B DOT 6030906 AT iki DOT fi> <5290B287 DOT 9070607 AT gmx DOT de> <5290BB33 DOT 1080806 AT iki DOT fi> <529DE517 DOT 103 AT gmx DOT de>
In-Reply-To: <529DE517.103@gmx.de>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Reply-To: djgpp-workers AT delorie DOT com

On 12/03/2013 04:05 PM, Juan Manuel Guerrero wrote:
> Am 23.11.2013 15:26, schrieb Andris Pavenis:
>> On 11/23/2013 03:49 PM, Juan Manuel Guerrero wrote:
>>> Am 23.11.2013 09:25, schrieb Andris Pavenis:
>>>> We should discard .gnu.lto_* sections in linker similarly as it
>>>> is done for Linux
>>>>
>>>> /DISCARD/: { *(gnu.lto_*) }
>>>>
>>>> Now that at least for current development version of GCC (4.9)
>>>> one can with some additional modifications enable LTO it would
>>>> be nice to have support for it also in binutils
>>>>
>>>> The related source file is ld/scripttempl/i386go32.sc
>>>>
>>>> Andris
>>>>
>>>
>>> Do you need a new release of bnu2232 with the linker script that has the
>>> DISCARD line added?
>>>
>>>
>> I guess, it is not urgent. We do not yet have released version of GCC
>> for DJGPP which supports LTO. It is not critical problem like broken
>> debugging info support, so it can wait. gcc-4.8.3 or gcc-4.9.0 release
>> is not expected very soon also.
>>
>> I could also build binutils myself if needed (although I would perhaps
>> do it under Linux using cross-compiling instead of native build for DJGPP)
>>
>> Also if one likes to check out, then build of snapshot with LTO support
>> enabled can be found at:
>>
>> http://ap1.pp.fi/djgpp/gcc/test/4.9.0-20131121/
>>
>> I have not done much testing with it. Build bash-2.0.5b using it and
>> specified CFLAGS="-O2 -g -flto" for building. All seemed to work.
>>
>> Andris
>>
>>
>>
>>
>
> In a couple of days I would like to upload the port of binutils 2.24.
> Apart from the usual changes to i386go32.sc that are not showed in the
> patch below, do you need some other LTO specific changes that must go
> into the port?  I will look at the sources anyway but you may be more
> familiar with the LTO stuff.
>
>

I only tried whether I can get it to work for DJGPP and specified -flto when
building bash-2.05b for DJGPP. Seemed to work.

I tested linker script update by providing modified linker script with one
line added (as below) and provided it using GCC command line option -T
without actually rebuilding binutils.

>
>
> diff -aprNU10 binutils-2.24.orig/ld/scripttempl/i386go32.sc binutils-2.24/ld/scripttempl/i386go32.sc
> --- binutils-2.24.orig/ld/scripttempl/i386go32.sc    2013-11-04 15:33:38 +0100
> +++ binutils-2.24/ld/scripttempl/i386go32.sc    2013-12-03 14:33:56 +0100
> @@ -63,12 +63,13 @@ SECTIONS
>      ${RELOCATING+ . = ALIGN(${SEGMENT_SIZE});}
>    }
>    /* Stabs debugging sections.  */
>    .stab 0 : { *(.stab) }
>    .stabstr 0 : { *(.stabstr) }
>  EOF
>
>  . $srcdir/scripttempl/DWARF.sc
>
>  cat <<EOF
> +  /DISCARD/: { *(gnu.lto_*) }
>  }
>  EOF
>
>
>

You can test it using mu build of gcc snapshot I mentioned before. Look whether
LTO related sections are discarded in executable. Even simple hello world style
program is sufficient for this test

Andris