Dear ATLAS support Team,
I am running on a Windows 7 (64 bit) with latest cygwin (32 bit) mingw compilers installed. ATLAS version is 3.11.17. I want to later use ATLAS with a 32 bit program that was compiled with MSVC 10.
I am configuring ATLAS via the line:
../configure -Si nocygwin 1 -b 32 -D c -DPentiumCPS=2890 --with-netlib-lapack-tarfile=/cygdrive/d/ATLAS-3.11.17/lapack-3.4.1.tgz
The configure results are attached. The configuration passes and upon make build
, I get the error:
i686-w64-mingw32-gcc-4.8.2: error: unrecognized command line option ‘-V’
The build output is attached as well. I am not quite sure if this has to do with these line(s) from the configure step:
make[2]: gcc-4: Command not found make[2]: *** [IRunCComp] Error 127
I am assuming that the build error (-V not recognized, see above) is related to the gcc-4 not found
, i.e. some cygwin versioning(?). Please let me know if necessary information is missing. Help is greatly appreciated.
Best Regards,
Michael
Hmm. Not at all sure what is going on here. First, try 3.10, since it was the last thing that was tested to work on Windows.
Also, are you sure that your MinGW compiler is kosher. I keep seeing this line:
: warning: SSE instruction set disabled, using 387 arithmetics [enabled by default]
If your MinGW does not support SSE, then I even if ATLAS can be installed, it is likely to be terrible performance wise.
If you have a mixture of good/bad MinGW tools, you may want to try:
http://math-atlas.sourceforge.net/atlas_install/node55.html#sec-MinGWbin
My ability to help on windows is extremely limited, but let me know what that gets you.
Cheers,
Clint
Ticket moved from /p/math-atlas/support-requests/935/
Thanks for your support.
I now tried with 3.10.1 and explicitly set the MinGW tools in a MinGW32.dat to:
The error(s) are the same as before. Is this gcc version supported? Maybe the line
contains a clue? I am not really familiar with the MinGW compiler toolchain naming in cygwin - it gives me headaches to be precise. Maybe this is the wrong compiler for what I am after?
As I said in my OP, I am on a 32 bit cygwin shell on a 64 bit Windows 7 trying to compile ATLAS with -b 32. Is there something wrong with that?
Again, thanks for your help, much appreciated.
Michael
I have the same error. I think the error:
might not be meaningful. I looked quickly in Make.top and it looks like this is just one of three possible ways of dumping version info about the compiler into an error log in the case of failure (you can also comment this line out and the rest of the error messages still appear). There are a couple other lines that I see that might be more meaningful:
and:
I found the source of the gcc-4 call in the file CONFIG/src/atlcomp.txt and changed the these lines to gcc:
After this change, I don't get any errors during configuration, but the two errors that Michael Bach gets that I mentioned still appear.
Hi, I have the same problem with the latest Cygwin/MinGW compilers. What I noticed is that the system() calls cannot find the compiler wrappers in the atlas-build dir when specified as cygwin paths (/cygdrive/d....) but it can find them when you specify them as 'D:\...' Windows style paths. I checked which compiler it was using to compile for example emit_buildinfo.c (which is the culprit in the issue mentioned in this ticket) and it is using XCC (which in my case is /usr/bin/i686-w64-mingw32-gcc.exe). I tried replacing it with /usr/bin/gcc.exe, but I still have the same issue.
Are you using cygwin Make? ATLAS will not build using MinGW alone.
Cygwin is required for build system, even if using the MinGW compilers.
Cheers,
Clint
On 04/15/2014 08:13 AM, Geert Litjens wrote:
--
R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley
Hi Clint,
I installed the latest Cygwin 32bit on Win64 to build a 32bit version of ATLAS. Within cygwin I installed GCC (C and Fortran), GNU Make, and the i686-MinGW 32bit compilers. The configure step seems to go fine, however the tune/sysinfo apps do not seem to work due the fact that they interpret the paths as Windows paths instead of Cygwin paths, even if I change the compiler (XCC) to the normal GCC (by default it was the mingw-gcc compiler). Maybe something happened in the newer versions of cygwin? Or it is somehow using the wrong includes/headers? I don't really know.
Clint,
I think this is a case sensitive thing. In my build directory, the "mgwgcc.exe" is generated during the configure step. However, when I try to run this executable with an uppercase -V, it fails. When I run with a lowercase -v, it works.
I don't see which file is setting the -V flag though... otherwise I would change it to a lowercase v and be done (I think).
This just fixes the problem with building the error report though...
Last edit: mholzel 2014-05-27
Yes, I think the '-V' flag is a distraction - it's arising and causing the error when trying to build the error report, but isn't the underlying cause of the error.
Sorry for the extreme neglect. I have overwhelmed, and have not been able to provide support for some time. In order to get back to even, I am mass closing existing old support requests.
If your support request involved the stable release, please try the new 3.10.3BETA release that you can find under the developer releases. That is the release that I have updated to work better on modern hardware, including MacOS and Windows. If you still have a problem with 3.10.3BETA, please open up a new ticket, and I will try to get it fixed before the official 3.10.3 release. Please note t
he updated ATLAS/doc/atlas_install.pdf: the online document is still for 3.10.2,
and will remain so until 3.10.3's official release.
If your support request involved the developer release, I recommend waiting for 3.11.39 before reopening, as I am currently in the middle of a major rewrite. If the problem persists into .39, I definitely want to hear about it.
Thank you, and I am very sorry about my responsiveness on support.