It is possible to decrease compilation time on fpga_compile node more then 3 times.
During compilation aoc make a lot of read/writes on project and tmp folders. Usually these folders are on users login virtual machine and compile node are on separate computer in cluster. These operations is quite slow. But there is temporary folder resides on compile node which can be used locally on node and which can help to avoid a lot of transfers through nfs. This folder is being created for each job on a node and is being deleted after job finished. You can find it in $PBS_SCRATCHDIR
So, simple script to use this feature (for "vector add" example) is:
#!/bin/bash
date
source /opt/intel/inteloneapi/setvars.sh
xTMPDIR=$TMPDIR
export TMPDIR=$PBS_SCRATCHDIR/tmp
U_SRC_DIR=$PWD
cd $PBS_SCRATCHDIR
mkdir tmp
cp -rf $U_SRC_DIR/* .
CPU_QTY=`nproc`
echo "Number of CPUs: $CPU_QTY"
make -j $CPU_QTY hw -f Makefile.fpga
cp -rf $PBS_SCRATCHDIR/tmp $xTMPDIR
rm -Rf $PBS_SCRATCHDIR/tmp
cp -rf $PBS_SCRATCHDIR $U_SRC_DIR
export TMPDIR=$xTMPDIR
cd $U_SRC_DIR
date
exit 0
to run this script on compile node:
qsub -l nodes=1:fpga_compile:ppn=2,walltime=24:00:00 -d . build_fpga_hw.sh
Instead of more than 4 hours it takes 1.2 hour.
