MAGMA Forum

hi I've followed I believe all the advices I could find on this forum and with: ### GPU_TARGET = Fermi CC = gcc NVCC = nvcc FORT = gfortran ARCH = gcc ARCHFLAGS = -shared -o RANLIB = strip OPTS = -O3 -DADD_ -fPIC -DCUBLAS_GFORTRAN F77OPTS = -O3 -DADD_ -fPIC FOPTS = -O3 -DADD_ -fPIC -x f95-cpp-input ...

dear Magma creators/developers I'd like to ask it you maybe have your favourite or recommended GPU(s)? Or maybe you have conducted some benchmarks? Reason I wonder is that I have here a GTX580-3GB and a GT640-2GB, the former is supposedly faster according to what one can gather form Internet sources...

hi just would like to ask experts if magma 1.3.0 should build on cuda 5.5 ? my compilation fails: gcc -O3 -DADD_ -fPIC -DHAVE_CUBLAS -DGPUSHMEM=200 -I/usr/local/cuda-5.5/include -I../include -I../control -c testing_zgemm.cpp -o testing_zgemm.o ../make.inc:40: *** unterminated variable reference. Sto...

yes, apologies, I found that out later and contacted Rmagma's author
thanks

I realise I missed out path to magma libs, but how do I get hold of magmablas?
changelog reads "...Merge libmagmablas into libmagma to eliminate circular dependencies....."

is it rather R magma package that needs some work, updates?
would there be a workaround?

regards

dear all I'm trying to install magma R package but build fails, I'm hoping you can help. my MAGMA is built with gcc and ACML 5.3.1, build seems successful, I ran some tests from testing and all seem good (by the way, on AMD platforms it seems to be worth all the efforts to build with ACML, for CPU t...

thank you mgates3
can confirm I missed one -fPIC, now I have

NVOPTS = -O3 -DADD_ -Xcompiler "-fno-strict-aliasing -fPIC"

and it all builds, however, I was hoping I could use magma R package but builds fails, I should put it into a separate post
thanks

it fails for me: ======================================== libmagma.so gcc -fPIC -fopenmp -shared -o ./lib/libmagma.so \ src/*.o control/*.o \ interface_cuda/*.o magmablas/*.cu_o magmablas/*.o \ -L/opt/acml/gfortran64_fma4_mp/lib -L/usr/lib64/atlas -L/usr/local/cuda/lib64 \ -lcblas -lcublas -lcudart ...

hello
does this same apply to ACML build?
thanks.

dear experts I'm trying to help my fellow scientist to glue together all the different bits & bobs, myself am not a programmer and this is why I don't quite get it. It is simple for you I believe - since ACML (5.3.1) advertises itself as full implementation of BLAS and LAPACK - why do I still need c...