Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
-
fletchjp
- Posts: 203
- Joined: Mon Dec 27, 2010 7:29 pm
Post
by fletchjp » Thu Oct 02, 2014 1:18 pm
I am returning to use Magma after some time away and have downloaded 1.5.0. When I copy my makefile which works for 1.3.0 and run make I get the following error message:
Code: Select all
#error -- unsupported GNU version! gcc 4.6 and up are not supported!
This implies that gcc 4.6 and higher versions are not supported.
I am running Ubuntu 12.04 with gcc 4.6 as the default compiler. I also have gcc 4.8.2 and gcc 4.9.0 and could use them, although I may need to install the fortran as I have not needed that before.
Is there some documentation somewhere about the changes to the compiler requirements?
Similarly, are there any changes to the requirements for CUDA?
Thank you
John
-
fletchjp
- Posts: 203
- Joined: Mon Dec 27, 2010 7:29 pm
Post
by fletchjp » Thu Oct 02, 2014 2:03 pm
My current CUDA version is 4.1 and that may be the problem. Please advise me what to do. Update CUDA?
John
-
mgates3
- Posts: 918
- Joined: Fri Jan 06, 2012 2:13 pm
Post
by mgates3 » Thu Oct 02, 2014 3:23 pm
This is not a MAGMA error. It may be a CUDA nvcc error. Yes, I would recommend updating CUDA.
If that doesn't solve the problem, please include more details such as your make.inc file and the complete output of make when the error occurs. (Not in parallel, make -j1.) It's hard to debug from just the one line error message.
-mark
-
fletchjp
- Posts: 203
- Joined: Mon Dec 27, 2010 7:29 pm
Post
by fletchjp » Thu Oct 02, 2014 3:24 pm
I have installed gcc/g++/gfortran 4.5 and set this in the makefile. CUDA is still accessing the default compiler (4.6) and stopping. Is there a define for CUDA to tell it to look at the other compiler?
-
fletchjp
- Posts: 203
- Joined: Mon Dec 27, 2010 7:29 pm
Post
by fletchjp » Thu Oct 02, 2014 3:33 pm
Apologies, the last post was before I had read your reply. Makefile is attached as make.inc.txt. The error message is below.
Code: Select all
fletcher@fletcher-desktop:~/magma-1.5.0$ make
compile for CUDA arch 2.x (Fermi)
======================================== magmablas
( cd magmablas && make )
make[1]: Entering directory `/home/fletcher/magma-1.5.0/magmablas'
/usr/local/cuda/bin/nvcc -O3 -DADD_ --compiler-options -fno-strict-aliasing -DUNIX -gencode arch=compute_20,code=sm_20 -gencode arch=compute_20,code=compute_20 -I/usr/local/cuda/include -I../include -I../control -I../sparse-iter/include -c zcaxpycp.cu -o zcaxpycp.o
In file included from /usr/local/cuda/include/cuda_runtime.h:59:0,
from <command-line>:0:
/usr/local/cuda/include/host_config.h:82:2: error: #error -- unsupported GNU version! gcc 4.6 and up are not supported!
make[1]: *** [zcaxpycp.o] Error 1
make[1]: Leaving directory `/home/fletcher/magma-1.5.0/magmablas'
make: *** [libmagma] Error 2
John
-
Attachments
-
- make.inc.txt
- Makefile for magma-1.5.0 with .txt added.
- (1.81 KiB) Downloaded 114 times
-
fletchjp
- Posts: 203
- Joined: Mon Dec 27, 2010 7:29 pm
Post
by fletchjp » Thu Oct 02, 2014 3:45 pm
I have found a way ahead without (yet) updating cuda. After loading gfortran-4.5 I can get nvcc to work by adding -ccbin /usr/bin/gfortran-4.5 to the NVCC line in make.inc.
That is currently building.
Thanks
John