search

CUFCTL / face-recognition

A GPU-accelerated real-time face recognition system based on classical machine learning algorithms
MIT License
23 stars 11 forks source link

Segfaults in MAGMA LAPACK calls #20

Closed bentsherman closed 7 years ago

bentsherman commented 7 years ago

I've added MAGMA to the matrix library, and it includes wrappers for cuBLAS functions so we don't actually need cuBLAS explicitly anymore. However, when I try to run ./test-matrix I get segfaults for m_eigen and m_inverse, each time in an LAPACK call.

Here are the instructions to build MAGMA on Palmetto:

# request a node
qsub -I -l select=1:ncpus=8:mem=16gb;walltime=02:00:00

# the Makefile looks in ../magma-2.2.0
cd face-rec/../

# download MAGMA
wget http://icl.cs.utk.edu/projectsfiles/magma/downloads/magma-2.2.0.tar.gz
tar -xvf magma-2.2.0.tar.gz magma-2.2.0

# select the right make.inc for Intel MKL
cp make.inc-examples/make.inc.mkl-icc make.inc

# build (set num-jobs to the number of cores in your CPU)
make -j [num-jobs]
bentsherman commented 7 years ago

My latest commit resolved these segfaults, the problem was that some of the workspace variables were allocated on the GPU when they should be allocated in host memory.