CompileNotes.txt

First, get the right build directory:

        GIT_SSL_NO_VERIFY=1 git clone https://usqcd.lns.mit.edu/git/LHPC/Public/build2.git; cd build2; git checkout build2-20170804

This is the version that has a tested QUDA interface. Then, get the source code of the
dependencies

        GIT_SSL_NO_VERIFY=1 make TARGET=fnal-pi0g update

We must then edit the edit the file configs/fnal-pi0g to ensure
it knows where your CUDA build is, and also that we wish to use the best
sm_XX availible.

        # CUDA installation
        CUDA.root = /usr/local/cuda-7.0

will need to change, as will

        -DQUDA_GPU_ARCH=sm_35

to refect the change to the CUDA version and sm_XX. We must also edit the script to allow HDF5 to compile.

Next obstacle is the GSL version that is downloaded is custom. It has a set of complex
QR decomposition routines written by Andrew, and the version he wrote them in does
not compile on Sierra. The solution is to download a different version of gsl that has
the complex QR routines and compiles on Sierra. Navigate to

        build2/parts/gsl

and execute

        git clone https://github.com/mangroveck/gsl-qr-complex-devel.git

you then move (or delete) the original 'tree' directory, and move the new gsl
directory to 'tree'.

        mv tree tree_old; mv gsl-qr-complex-devel tree

You must then generate the configure script.

        mv tree; autoreconf -i

Last, you must edit the HDF5.config so that the libdl library is included

        HDF5.config = CC="$(XCC)" \
                      CFLAGS="$(XCFLAGS)" \
                      LD="$(XCC) -ldl" \
                      --prefix="$(HDF5.dest)" \
                      --enable-parallel

and also in the qlua.libs line

         QLUA.libs   = z cudart dl

This should be sufficient to compile a QUDA enables Qlua. We may then make.

         make PREFIX=/g/g90/howarth1/clean_QLUA/QLUA TARGET=fnal-pi0g

Once you are satisfied that QLUA has compiled correctly, copy the source code
files over and recompile.