Cannot build latest Julia

[milktrader]

[julia (master)]
✈  make
f951: error: unrecognized command line option "-frecursive"
f951: error: unrecognized command line option "-frecursive"
make[5]: *** [sgbcon.o] Error 1
make[5]: *** Waiting for unfinished jobs....
make[5]: *** [sgbbrd.o] Error 1
make[4]: *** [lapacklib] Error 2
make[3]: *** [netlib] Error 2
*** Clean the OpenBLAS build with 'make -C deps clean-openblas'. Rebuild with 'make OPENBLAS_USE_THREAD=0 if OpenBLAS had trouble linking libpthread.so, and
 with 'make OPENBLAS_TARGET_ARCH=NEHALEM' if there were errors building SandyBridge support. Both these options can also be used simultaneously. ***
make[2]: *** [openblas-v0.2.9/libopenblas.dylib] Error 1
make[1]: *** [julia-release] Error 2
make: *** [release] Error 2

Commit:

8293dc614cb34847389bf8efca9fc1d4c201c6d6 Remove -lz from OSLIBS. Should be handled automatically by llvm-config in src/Makefile

[milktrader]

I tried 'make -C deps clean-openblas followed by make OPENBLAS_USE_THREAD=0 (which errored out) and then again with make OPENBLAS_TARGET_ARCH=NEHALEM which also errored out.

[milktrader]

I did get this to compile with echo override USE_SYSTEM_BLAS = 1 >> Make.user cc @vtjnash

[davidanthoff]

Same here, was able to compile yesterday, now get the same error message as @milktrader. On Windows using msys2 setup.

[ViralBShah]

What is your gfortran version?

[davidanthoff]

$ gfortran --version
GNU Fortran (rev5, Built by MinGW-W64 project) 4.8.1
Copyright (C) 2013 Free Software Foundation, Inc.

I just followed the instructions README.windows.md to set everything up, including mingw-builds.

[davidanthoff]

Actually, I get a different error message than @milktrader:

../libopenblasp-r0.2.9.a(sgesvd.o):sgesvd.f:(.text+0x40e): undefined reference to `_gfortran_concat_string'
../libopenblasp-r0.2.9.a(sgesvd.o):sgesvd.f:(.text+0xe7d): undefined reference to `_gfortran_concat_string'
c:/mingw-builds/x64-4.8.1-win32-seh-rev5/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/4.8.1/../../../../x86_64-w64-mingw32/bin/ld.exe: ../libopenblasp-r0.2.9.a(sgesvd.o): bad reloc address 0x0 in section `.pdata'
collect2.exe: error: ld returned 1 exit status
Makefile:85: recipe for target '../libopenblas.dll' failed
make[4]: *** [../libopenblas.dll] Error 1
Makefile:100: recipe for target 'shared' failed
make[3]: *** [shared] Error 2
*** Clean the OpenBLAS build with 'make -C deps clean-openblas'. Rebuild with 'make OPENBLAS_USE_THREAD=0 if OpenBLAS had trouble linking libpthread.so, and with 'make OPENBLAS_TARGET_ARCH=NEHALEM' if there were errors building SandyBridge support. Both these options can also be used simultaneously. ***
Makefile:804: recipe for target 'openblas-v0.2.9/libopenblas.dll' failed
make[2]: *** [openblas-v0.2.9/libopenblas.dll] Error 1
Makefile:43: recipe for target 'julia-release' failed
make[1]: *** [julia-release] Error 2
Makefile:32: recipe for target 'release' failed
make: *** [release] Error 2

[Luthaf]

I get a slightly different error message, but it seems to be related to OpenBLAS too.

The traceback is pretty long, so I putted it here

My gfortran version, in the case it may be interesting:

$ gfortran --version
GNU Fortran (GCC) 4.4.7 20120313 (Red Hat 4.4.7-4)
Copyright (C) 2010 Free Software Foundation, Inc.

[milktrader]

✈  gfortran --version
GNU Fortran (GCC) 4.2.1 (Apple Inc. build 5666) (dot 3)
Copyright (C) 2007 Free Software Foundation, Inc.

[stevengj]

The -frecursive option seems to have been added in gfortran 4.3.x.

[davidanthoff]

Skimming the logs posted here, it actually looks like the build breaks for three different reasons for @milktrader, @Luthaf and me... I'm starting a fresh clean build off master on my machine now and will post the full log once it is done.

[davidanthoff]

I'm preparing a PR that fixes my issue.

[milktrader]

I'll try the gfortran update route, but it appears brew has bundled it with gcc

[milktrader]

I commented out the Make.user override USE_SYSTEM_BLAS = 1 trick and the compile worked with a brew gcc, but when I try to get a version for my gfortran I get this:

✈  gfortran --version
gfortran: error trying to exec '/usr/local/bin/../Cellar/gcc/4.8.3_1/bin/i686-apple-darwin11-gfortran-4.2.1': execvp: No such file or directory

Is that because gfortran is bundled with gcc now? In any case, I'll close this issue since it seems resolved with updating gfortran

[tkelman]

@Luthaf do you have a Haswell processor? I saw the same problem you did with Error: no such instruction: vpermpd in the Haswell kernels, and was able to work around it by putting override OPENBLAS_DYNAMIC_ARCH=0 in my Make.user.

[Luthaf]

@Luthaf do you have a Haswell processor? I saw the same problem you did with Error: no such instruction: vpermpd in the Haswell kernels, and was able to work around it by putting override OPENBLAS_DYNAMIC_ARCH=0 in my Make.user.

I don't know. I'll check this on Monday. But I have already managed to build Julia every week for the two last months.

[Luthaf]

@Luthaf do you have a Haswell processor?

No, it is a Core2 Duo. cat /proc/cpuinfo gives me Intel(R) Core(TM)2 Duo CPU E6750 @ 2.66GHz.

A fully fresh buil :

git clone https://github.com/JuliaLang/julia julia
cd julia
make

produces the same error.

[tkelman]

Try

echo "OPENBLAS_DYNAMIC_ARCH=0" > Make.user
make

[Luthaf]

Well, it finally works with this setting. But after three make clean && make pass, which is weird.

Thank you !

[timholy]

I just got bit by the same compilation error as @Luthaf. To make it easier for others to search for the issue, one of the final errors is

../kernel/x86_64/dgemm_kernel_4x4_haswell.S:1743: Error: no such instruction: `vpermpd $ 0xb1,%ymm3,%ymm3'

I did a make -C deps clean-openblas and then used @tkelman's solution 2 comments above. Worked for me.

[tkelman]

If this ends up being really common, should we try to come up with a little configure-style probe and do this automatically or suggest the fix within the makefile?