magma_types.h line 465

Open discussion for MAGMA

magma_types.h line 465

Postby patmc » Mon Nov 25, 2013 4:57 pm

Hello all,
my first post. I'm trying to get my head round Magma.
I compiled 1.4.0 for the atlas blas and lapack okay.
then tried OpenBLAS version -- still with atlas lapack.
I got the following error:

gcc -O3 -DADD_ -DGPUSHMEM=200 -DHAVE_CUBLAS -DMAGMA_WITH_PLASMA -I/usr/local/cuda/include -I/usr/local/include -I../include -I../control -c zgetrf_incpiv_gpu.cpp -o zgetrf_incpiv_gpu.o
In file included from ../include/magmablas_z.h:14:0,
from ../include/magmablas.h:12,
from ../include/magma.h:15,
from ../control/common_magma.h:58,
from zgetrf_incpiv_gpu.cpp:18:
../include/magma_types.h:465:13: error: expected primary-expression before ‘int’
../include/magma_types.h:465:13: error: expected ‘]’ before ‘int’
../include/magma_types.h:465:13: error: expected initializer before ‘int’

the line in question is:
const char* lapack_const ( int magma_const );
which looks fine to me;
I've gotten the same compile error with both 1.4.0 and with 1.4.1 which I just downloaded.
My make.inc is:
GPU_TARGET = Fermi
CC = gcc
NVCC = nvcc
FORT = gfortran
ARCH = ar
ARCHFLAGS = cr
RANLIB = ranlib
OPTS = -O3 -DADD_
F77OPTS = -O3 -DADD_
FOPTS = -O3 -DADD_ -x f95-cpp-input
NVOPTS = -O3 -DADD_ --compiler-options -fno-strict-aliasing -DUNIX
LDOPTS = -fPIC -Xlinker -zmuldefs
LIB = -lopenblas -lpthread -lcublas -lcudart -llapack -lm
CUDADIR = /usr/local/cuda
LIBDIR = -L/usr/local/OpenBLAS -L/usr/local/cuda/lib64 -L/usr/lib64 -L/usr/local/atlas/lib/
INC = -I$(CUDADIR)/include

System is an i7, running Debian, kernel 3.10-2, building with gcc 4.7 and gfortran, CUDA 5.5

Any ideas?
BTW thanks for all the work. Magma is a very cool project, I'm learning - lots!

regards
Pat
patmc
 
Posts: 3
Joined: Mon Nov 25, 2013 4:27 pm

Re: magma_types.h line 465

Postby patmc » Mon Nov 25, 2013 5:55 pm

I made some progress.
I took the atlas lapack out of the make.inc as I figured the openblas had enough functionality.
Then I used CMake to do an out of tree build on 1.4.1. Build went all the way through including the tests!
Doing mkl and acml now; picking up my 2075 tomorrow :-)

Pat
patmc
 
Posts: 3
Joined: Mon Nov 25, 2013 4:27 pm

Re: magma_types.h line 465

Postby mgates3 » Sun Dec 01, 2013 2:30 am

I think the problem with the make.inc build was a conflict with building with plasma -- both plasma and magma define lapack_const, in different ways. The quickest solution is to disable the magma-with-plasma routines. In Makefile.internal, comment out PLASMA = ..., like this:

# Use Plasma to compile zgetfl and ztstrf
# PLASMA = $(shell pkg-config --libs plasma 2> /dev/null )

I'll see if there's a sensible way for the libraries to work together.

-mark
mgates3
 
Posts: 415
Joined: Fri Jan 06, 2012 2:13 pm

Re: magma_types.h line 465

Postby patmc » Tue Dec 03, 2013 2:54 am

Thanks Mark
Been distracted building my 2075 machine.
Question about heterogeneity: I suspect I'm asking for the moon but if I were to stick both a Tesla 2075 and a Xeon Phi in the same machine can Magma handle them together?
-pat
patmc
 
Posts: 3
Joined: Mon Nov 25, 2013 4:27 pm

Re: magma_types.h line 465

Postby mgates3 » Wed Dec 04, 2013 8:29 pm

Nope, we have versions of MAGMA for CUDA, Xeon Phi (MIC), and OpenCL, but not simultaneously.
-mark
mgates3
 
Posts: 415
Joined: Fri Jan 06, 2012 2:13 pm


Return to User discussion

Who is online

Users browsing this forum: No registered users and 2 guests

cron