Compiling ECM 5.0.3 for windows

[BotXXX]

I am trying to compile the ECM 5.0.3 package on a windows system to create a windows binary. I am currently using OpenWatcom but for one reason it won't go with the supplied make file. Besides that it also looks to me that the makefile is desinged for Linux systems instead of Windows systems.

I would like to build some binaries for testing and working. Can anybody give me a help/guide or a push in the right direction. Would it matter to use Cygwin for compiling instead of OpenWatcom? Or does someone have experience in compiling the 5.0.3 as a windows binary.

Source at http://webloria.loria.fr/~zimmerma/r...ecm-5.0.3.html

This is the message i get when compiling:

Open Watcom C/C++ NMAKE Clone for 386 Version 1.3
Portions Copyright (c) 1995-2002 Sybase, Inc. All Rights Reserved.
Source code is available under the Sybase Open Watcom Public License.
See http://www.openwatcom.org/ for details.
makefile(62): Error(E23): Extensions reversed in implicit rule
Error(E02): Make execution terminated

The lines of code where it goes wrong:

.SUFFIXES: .c .o

ecm: $(ALLOBJS) ecm.h ecm-gmp.h
$(LD) -L$(GMP)/lib $(ALLOBJS) -o $@ $(LDFLAGS)

ecm_with_ntl:
make ecm GMP=$(GMP) NTL=$(NTL) CXX=g++ EXTRAOBJS="ntl.o polyz.o" LDFLAGS="-L$(NTL)/lib -lntl $(LDFLAGS)" CFLAGS="$(CFLAGS) -DPOLYGCD"

tune: mpmod.o ecm.h tune.o auxi.o mul_lo.o ecm-gmp.h
$(CC) $(CFLAGS) -L$(GMP)/lib tune.o mpmod.o auxi.o mul_lo.o -o $@ -lgmp

ntl.o: ntl.c
$(CXX) $(CFLAGS) -c -I$(GMP)/include -I$(NTL)/include $<

.c.o: ecm.h ecm-gmp.h
$(CXX) $(CFLAGS) -I$(GMP)/include -c $<

The 62th line is starting with ".c.o:"

[xilman]

Originally Posted by BotXXX

I am trying to compile the ECM 5.0.3 package on a windows system to create a windows binary. I am currently using OpenWatcom but for one reason it won't go with the supplied make file. Besides that it also looks to me that the makefile is desinged for Linux systems instead of Windows systems.

I would like to build some binaries for testing and working. Can anybody give me a help/guide or a push in the right direction. Would it matter to use Cygwin for compiling instead of OpenWatcom? Or does someone have experience in compiling the 5.0.3 as a windows binary.

Source at http://webloria.loria.fr/~zimmerma/r...ecm-5.0.3.html

This is the message i get when compiling:
The lines of code where it goes wrong:

The 62th line is starting with ".c.o:"

Here's one I built earlier. It assumes a p6 cpu, i.e., a PPro or later, and will not work on vanilla Pentium or earlier.


Paul

[BotXXX]

Paul, thank you. I see that you compiled it with Cygwin. I will have a go with that. In the mean time i will use your compiled version.

[akruppa]

Paul,

can you give us hints on what it takes to get gmp-ecm to compile under Windows? Neither PaulZ or I can do much in this regards as neither of us has Windows running anywhere. It'd be great if we could include instructions in future gmp-ecm releases.

Alex

[JHansen]

Originally Posted by akruppa

Paul,

can you give us hints on what it takes to get gmp-ecm to compile under Windows? Neither PaulZ or I can do much in this regards as neither of us has Windows running anywhere. It'd be great if we could include instructions in future gmp-ecm releases.

Alex

Well, if you just install Cygwin (and remembers to include the gcc tools and make), and then install GMP, GMP-ECM installs without any trouble at all.

The only thing you have to make sure, is that you manually select the gcc tools and the make tool during installation of cygwin.


Jes

[dleclair]

Both GMP and GMP-ECM 5.0.3 compile easily using Mingw which gives you a Windows binary without the dependency on the Cygwin DLL.

Don Leclair

[xilman]

Originally Posted by akruppa

Paul,

can you give us hints on what it takes to get gmp-ecm to compile under Windows? Neither PaulZ or I can do much in this regards as neither of us has Windows running anywhere. It'd be great if we could include instructions in future gmp-ecm releases.

Alex

There are two main ways.

The first is to install some sort of Unix emulator. By far the best, IMAO, is SFU (Services for Unix) which is a free download from microsoft.com. As well as lots of Unix goodies, it comes with a gcc and associated tool set. The site http://www.interopsystems.com/tools/warehouse.htm has lots and lots of add-ons for SFU and some replacements for the utilities shipped. I no longer run a Windows system without SFU installed.

Related is to use Cygwin and/or MIngw. Either of these work, after a fashion, but are clunky (again, IMAO). The free (as in beer) version of Cygwin has a seriously restrictive license condition attached. Any binary distributed that uses the cygwin1.dll must also distribute the corresponding source under the GPL. In the case of gmp-ecm that's not a problem so I pushed out the binary. I could not redistribute a binary built from a BSD, Artistic, Public Domain, etc, source in this way.

The other approach is to acquire (possibly by building from source) a VS-compatible gmp library and then build gmp-ecm from source. If you have such a library it is then very easy to build the executable. Unfortunately, the GMP people seem to go out of their way to make it hard for people to build GMP under Visual Studio. For instance, on the gmp-developers' mailing list this offer appeared two weeks ago:

2) Are GMP authors interested in win32 makefiles ?
I created some bakefiles (XML-like project files for
bakefile: http://bakefile.sourceforge.net) which are
able to generate makefiles for various win32 compilers:
borland, msvc, mingw, watcom....

The build system I've created does not require any change
in the code: it's a small win32 directory to add to the
main GMP folder which contains the bakefiles
(from which the makefiles can be re-generated) and
the generated makefiles ready to be run...

I think that adding these files to the official distribution
could be very interesting since it would allow a wider range of
users to take advantage of GMP.

It met with this response:

I don't want to maintain another set of makefiles,
and for a platform which I don't use. Sorry.

from Torbjorn Granlund. Here we have someone who has put in a lot of work to do something a number of people have been asking for and many people could use. He gets a peremptory dismissal in return. I was distinctly underimpressed.

Technically GMP is a superb library. I wish I could say the same about other aspects of it.

Perhaps sometime when I'm feeling less thoroughly disillusioned I may build some native Windows 5.0.3 binaries for various machine architectures.

Paul

[BotXXX]

I had a go with MinGW but encountered some problems. Maybe one can help me in solving the problem.


- Installed MinGW + MSYS.
- Copied GMP414 dir to home dir in MSYS.
- Copied ECM503 dir to home dir in MSYS.
- started MSYS and browsed to GMP414 dir.
- Typed ./configure
- Typed make
- Typed make install
- Typed make check
- Found no erros in the check.
- Browsed to ECM503 dir.
- Typed make GMP=/local

- Got the following error:

$ make GMP=/local
gcc -O2 -g -W -Wall -Wmissing-prototypes -I/local/include -c auxi.c
auxi.c:94:26: sys/resource.h: No such file or directory
auxi.c: In function `cputime':
auxi.c:99: error: storage size of `rus' isn't known
auxi.c:101: warning: implicit declaration of function `getrusage'
auxi.c:101: error: `RUSAGE_SELF' undeclared (first use in this function)
auxi.c:101: error: (Each undeclared identifier is reported only once
auxi.c:101: error: for each function it appears in.)
auxi.c:99: warning: unused variable `rus'
auxi.c: In function `get_random_ui':
auxi.c:129: warning: implicit declaration of function `gettimeofday'
make: *** [auxi.o] Error 1


Appears that resource.h is not in MinGW. How can i include this file? Which package is it connected to?

[akruppa]

On my SuSE system, sys/resource.h is in glibc-devel. A quick google revealed that this header apparantly is not available under Mingw. There are a bunch of other systems where getrusage() is not available and we fall back to clock() in that case, see auxi.c, lines 84ff. I guess all we need to do is add a test for Mingw to the #if. Please try the attached replacement auxi.c. I also #ifdef'd out gettimeofday if no sys/time.h was included.

Let me know if that works.

Alex

[BotXXX]

I Changed the defined to the following:

#if defined (ANSIONLY) || defined (USG) || defined (__SVR4) || defined (_UNICOS) || defined(__hpux) || defined(__MINGW32__)

Since it is in the windows version of MinGW called __MINGW32__

But now there is an other error.


$ make GMP=/local
gcc -O2 -g -W -Wall -Wmissing-prototypes -I/local/include -c auxi.c
auxi.c: In function `get_random_ui':
auxi.c:111: error: storage size of `tv' isn't known
auxi.c:111: warning: unused variable `tv'
make: *** [auxi.o] Error 1

[akruppa]

Oops, the

struct timeval tv;

needs to be #ifdef'd, too.

#ifdef _SYS_TIME_H
struct timeval tv;
#endif

Alex