For reference, it looks like all my CPUs are
Intel(R) Xeon(R) CPU E5-2665 0 @ 2.40GHz
so maybe I should try core-avx2 (though I am not sure if this is the right
instruction, either...)
is
-march=native
the accepted solution right now? I've also tried this and am running into
similar errors.
My g++ is 5.2
On Tuesday, January 24, 2017 at 3:07:29 PM UTC-5, Jack Hessel wrote:
>
> I am trying to use theano on a redhat linux cluster, and I am running into
> a few problems.
>
> Specifically, I am getting an error about an undefined instruction. This
> error happens only sporadically (even on the same machine), which makes
> debugging it even harder.
>
> Here's the exception:
>
> Exception: ('The following error happened while compiling the node',
> Gemm{no_inplace}(dense_1_target\
> , TensorConstant{1.0}, dense_input_1, dense_1_W, TensorConstant{-1.0}),
> '\n', "Compilation failed (r\
> eturn status=1): /var/opt/condor/execute/dir_122882/cc4LC9Rp.s: Assembler
> messages:. /var/opt/condor\
> /execute/dir_122882/cc4LC9Rp.s:2579: Error: no such instruction: `shrx
> %rcx,(%r9),%rdi'. /var/opt/co\
> ndor/execute/dir_122882/cc4LC9Rp.s:2581: Error: no such instruction: `shrx
> %rcx,8(%r9),%r10'. /var/o\
> pt/condor/execute/dir_122882/cc4LC9Rp.s:2604: Error: no such instruction:
> `shrx %rdx,(%rcx),%rdx'. /\
> var/opt/condor/execute/dir_122882/cc4LC9Rp.s:2785: Error: no such
> instruction: `shrx %rcx,(%r9),%rdi\
> '. /var/opt/condor/execute/dir_122882/cc4LC9Rp.s:2787: Error: no such
> instruction: `shrx %rcx,8(%r9)\
> ,%r10'. /var/opt/condor/execute/dir_122882/cc4LC9Rp.s:2810: Error: no such
> instruction: `shrx %rdx,(\
> %rcx),%rdx'. ", '[Gemm{no_inplace}(dense_1_target, TensorConstant{1.0},
> dense_input_1, dense_1_W, Te\
> nsorConstant{-1.0})]')
>
> Here's the full compilation line (forgive my ugly paths -- It's a long
> story!)
>
> /home/jmh563/gcc-5.2.0/bin/g++ -shared -g -O3 -fno-math-errno
> -Wno-unused-label -Wno-unused-variable -Wno-write-strings
> -Wl,-rpath,/home/jmh563/otherInstalls/OpenBLAS/lib -march=haswell -mmmx
> -mno-3dnow \
> -msse -msse2 -msse3 -mssse3 -mno-sse4a -mcx16 -msahf -mmovbe -maes
> -mno-sha -mpclmul -mpopcnt -mabm -mno-lwp -mfma -mno-fma4 -mno-xop -mbmi
> -mbmi2 -mno-tbm -mavx -mavx2 -msse4.2 -msse4.1 -mlzcnt -mno-rtm\
> -mno-hle -mrdrnd -mf16c -mfsgsbase -mno-rdseed -mno-prfchw -mno-adx
> -mfxsr -mxsave -mxsaveopt -mno-avx512f -mno-avx512er -mno-avx512cd
> -mno-avx512pf -mno-prefetchwt1 -mno-clflushopt -mno-xsavec -mno-xsa\
> ves -mno-avx512dq -mno-avx512bw -mno-avx512vl -mno-avx512ifma
> -mno-avx512vbmi -mno-clwb -mno-pcommit -mno-mwaitx --param l1-cache-size=32
> --param l1-cache-line-size=64 --param l2-cache-size=20480 -mtune=\
> haswell -DNPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION -m64 -fPIC
> -I/home/jmh563/curPythonInstall/lib/python2.7/site-packages/numpy-1.13.0.dev0+1c8ecc7-py2.7-linux-x86_64.egg/numpy/core/include
>
> -I/home/jmh5\
> 63/curPythonInstall/include/python2.7
> -I/home/jmh563/curPythonInstall/lib/python2.7/site-packages/Theano-0.9.0.dev5-py2.7.egg/theano/gof
>
> -L/home/jmh563/otherInstalls/OpenBLAS/lib -L/home/jmh563/curPython\
> Install/lib -fvisibility=hidden -o
> /var/tmp/theano-360142/compiledir_Linux-2.6-el6.x86_64-x86_64-with-redhat-6.6-Santiago-x86_64-2.7.13-64/tmpnoStZa/1ef31f0e43a0cb34a58a9d31cef1eb8d.so
>
> /var/tmp/theano-36\
> 0142/compiledir_Linux-2.6-el6.x86_64-x86_64-with-redhat-6.6-Santiago-x86_64-2.7.13-64/tmpnoStZa/mod.cpp
>
> -lopenblas -lpython2.7
>
> and here's my theanorc
>
> [global]
> device = cpu
> floatX = float32
> cxx = /home/jmh563/gcc-5.2.0/bin/g++
> gcc.cxxflags = '-march=core2'
>
> (I tried to add -march=core2 because I read in a PR somewhere that that
> would work)
>
> I have done some googling and this seems to be a relatively common problem
> for folks. Is there an accepted set of steps to address this sort of thing?
> Or just try out a bunch of different settings of gcc.cxxflags?
>
> Jack
>
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
