[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <4EBD63EC.3000408@hushmail.com>
Date: Fri, 11 Nov 2011 19:05:32 +0100
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: more targets using sse-intrinsics.S
I see... For the x86-64 builds, we *must* use .S because otherwise the
compiler will (or may?) use sse-intrinsics.c to produce
sse-intrinsics.o! But since there is no sse-intrinsics-32.c we can
indeed replace all sse-intrinsics-32.S in the targets' object lists to
sse-intrinsics-32.o.
I just tested this with the linux sse2i target and it works fine. I
suppose it's easier to "follow" if you change them too.
magnum
2011-11-11 18:47, jfoug wrote:
> There is a problem building under cygwin. I get this:
>
> [clip]
> gcc -c -Wall -O2 -fomit-frame-pointer -I/usr/local/include -Wall -mpreferred-stack
> gcc -c -DUNDERSCORES x86.S
> gcc -c -DUNDERSCORES x86-sse.S
> gcc sse-intrinsics-32.S DES_fmt.o DES_std.o ...... unique.o x86.o x86-sse.o
> -s -L/usr/local/lib -L/usr/local/ssl/lib -lcrypto -lssl -lm -lz -lkernel32 -o ../run/john.exe
> /tmp/ccWTQyvz.o:sse-intrinsics.c:(.text+0x126): undefined reference to `memcpy'
> /tmp/ccWTQyvz.o:sse-intrinsics.c:(.text+0x1cb): undefined reference to `memcpy'
> /tmp/ccWTQyvz.o:sse-intrinsics.c:(.text+0x22c): undefined reference to `memcpy'
> /tmp/ccWTQyvz.o:sse-intrinsics.c:(.text+0x291): undefined reference to `memcpy'
> [clip]
>
> I clipped a lot to avoid unneed linker error messages.
>
> This is easy to fix. In cygwin (and mingw) sse2i, the build is this:
>
> JOHN_OBJS="sse-intrinsics-32.S $(JOHN_OBJS) x86.o x86-sse.o " \
>
> If this is changed to this:
>
> JOHN_OBJS="sse-intrinsics-32.o $(JOHN_OBJS) x86.o x86-sse.o " \
>
> Then the behavior works.
>
> Here is the build 'view' of things:
>
> [clip]
> gcc -c -Wall -O2 -fomit-frame-pointer -I/usr/local/include -Wall -mpreferred-stack
> gcc -c -DUNDERSCORES x86.S
> gcc -c -DUNDERSCORES x86-sse.S
> gcc -c -DUNDERSCORES sse-intrinsics-32.S
> gcc sse-intrinsics-32.o DES_fmt.o DES_std.o ...... unique.o x86.o x86-sse.o
> -s -L/usr/local/lib -L/usr/local/ssl/lib -lcrypto -lssl -lm -lz -lkernel32 -o ../run/john.exe
> [clip but build worked just fine]
>
> So at least for the build items needing -DUNDERSCORES, it appears to be very important to have sse-intrinsics-32.o in the OBJS, and not sse-intrinsics-32.S Actually, I am not sure why you have been adding the sse-intrinsics-32.S to the OBJ items in the makefile.
>
> I have a patch which would work for mingw/cygwin, however, before making that, I wonder if we should simply replace all of the sse-intrinsics-32.S in the OBJ's, to sse-intrinsics-32.o?
>
> Jim.
>
>
>> From: magnum [mailto:john.magnum@...hmail.com]
>>
>> 2011-11-10 21:37, jfoug wrote:
>>> Patch 0005 has been placed on the wiki. It is a perl script, placed
>>> in ./run, and a couple of changes to Makefile.
>>
>> Patch 0007 unifies them - I moved your perl script to "make intrinsics"
>> and reverted the target changes - the same, windows-ready, .S file is
>> now used for windows and non-windows. Seems to work fine, I hope I did
>> not screw anything up.
>
>
>
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
