#183 Make check fails with LAPACK 3.4.0

[Anonymous]

This may be related to the other problem I experienced, but after the build, issuing a 'make check' fails. Here is the output:
sqrtst.o: In function `qrtest':
qrtst.c:(.text+0x687): undefined reference to `ATL_C2Fsormqr'
sqrtst.o: In function `rqtest':
qrtst.c:(.text+0xc24): undefined reference to `ATL_C2Fsormrq'
sqrtst.o: In function `lqtest':
qrtst.c:(.text+0x11da): undefined reference to `ATL_C2Fsormlq'
sqrtst.o: In function `qltest':
qrtst.c:(.text+0x17b6): undefined reference to `ATL_C2Fsormql'
collect2: ld returned 1 exit status
make[3]: *** [xsqrtst] Error 1
make[3]: Leaving directory `/home/melrobin/octave/ATLAS/build/bin'
make[2]: *** [sanity_test] Error 2
make[2]: Leaving directory `/home/melrobin/octave/ATLAS/build/bin'
make[1]: *** [sanity_test] Error 2
make[1]: Leaving directory `/home/melrobin/octave/ATLAS/build'
make: *** [test] Error 2
[melrobin@melrobin build]$

Any ideas on how I can fix this?

[R. Clint Whaley]

Yeah, I would need an error report, and this is probably a symptom of another error if you experienced other problems. Can you point me at your other error report, and we'll start there and see if that clears this problem up? If it already has, just close this tracker item.

Thanks!
Clint

[R. Clint Whaley]

In order to diagnose the problem better, I need the error report. Read about that at:
http://math-atlas.sourceforge.net/faq.html#help

Thanks,
Clint

[Anonymous]

Error report for this ticket.

[Anonymous]

OK, I did a make error_report and attached it. The problem is still occurring with 3.9.55 too.

[R. Clint Whaley]

Hi,

OK, let's try to simply. In the future, always post the error report from the most recent developer (you posted .53, but said you had tried with .55). The most recent release will be the easiest for me to grapple with.

Now, what is the configure command you are using? I notice that you are building the shared libraries. Let's eliminate this and see if the static build also has problems.

In the meantime, I have started an install on the same architecture (etl-core28), and will update when I see if that reproduces your error.

Thanks,
Clint

[R. Clint Whaley]

Works fine for me building static libs. Trying again with dynamic build, will update when install complete (core28).

[Anonymous]

OK, I tried to just create static libraries and it failed again with the same messages. Here is my configure statement:
../configure -b 64 --with-netlib-lapack-tarfile=../lapack-3.4.0.tgz

I am trying this with .55 and am uploading an error report.

[Anonymous]

Error report using just static libraries on .55

[R. Clint Whaley]

My dynamic build also works fine.

[R. Clint Whaley]

The new error report is still 3.9.53, not 55, can you delete and upload the report for 55.

Also, please specify the exact configure command you are using. What it almost looks like is that you have told ATLAS to build the full lapack, but not given it access to a compilable lapack tarfile . . .

Thanks,
Clint

[JohnF]

Found same problem. Linked to change of compilation of lapack-3.4.0. The PLAT variable is no longer used so the compile wont create the lapack_ARCH.a library which is looked for by atlas.

The main problem is that the build does not pick this up as a reportable error.

I hacked the make.inc (lapack) to set
LAPACKLIB = lapack_Corei164SSE3.a

Not the correct way to do it but it provided the right behaviour.

[Anonymous]

ATLAS builds the full LAPACK 3.4.0 fine, but just fails make check. In other words, make build works perfectly and creates the shared libraries. make check fails and calls out those object files. It seems as though it is having some troubles just building the tests.

This was the exact command that I issued from the Linux prompt:
../configure -b 64 --with-netlib-lapack-tarfile=../lapack-3.4.0.tgz

The GZIP'd tar file is in the directory above the build directory.

[JohnF]

Found same problem. Linked to change of compilation of lapack-3.4.0. The PLAT variable is no longer used so the compile wont create the lapack_ARCH.a library which is looked for by atlas.

The main problem is that the build does not pick this up as a reportable error.

I hacked the make.inc (lapack) to set
LAPACKLIB = lapack_Corei164SSE3.a

Not the correct way to do it but it provided the right behaviour.

[Anonymous]

@jfloyd99: Yes, but this leaves me unsure of what I should now do. ATLAS is completely built with the LAPACK file being GZIP'd and tarred. Are you saying that for the make check I should gunzip, untar, edit the make.inc file and then make check?

[R. Clint Whaley]

Floyd: thanks for pointing this out! I will get lapack 3.4 and see what has changed.

In the meantime google-account dude can try building with an older lapack. The last build I used lapack 3.2.1 . . .

Cheers,

[JohnF]

Found same problem. Linked to change of compilation of lapack-3.4.0. The PLAT variable is no longer used so the compile wont create the lapack_ARCH.a library which is looked for by atlas.

The main problem is that the build does not pick this up as a reportable error.

I hacked the make.inc (lapack) to set
LAPACKLIB = lapack_Corei164SSE3.a

Not the correct way to do it but it provided the right behaviour.

[Anonymous]

I just confirmed that make check does work using .55 and lapack-3.3.1.

[R. Clint Whaley]

Confirm the bug. It is indeed due to a change in lapack 3.4.0. Am presently testing a fix.

[R. Clint Whaley]

OK, this should be fixed in 3.9.56 which I just posted. Reopen if the problem reoccurs with that release.

Thanks!
Clint