#!/bin/bash

##PBS -q EarlyAppAccess
#PBS -q debug
#PBS -l select=2
#PBS -l walltime=00:20:00
#PBS -A LatticeQCD_aesp_CNDA

cd $PBS_O_WORKDIR

source ../sourceme.sh

cp $PBS_NODEFILE nodefile

export OMP_NUM_THREADS=4
export MPICH_OFI_NIC_POLICY=GPU

#export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE
#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE
#unset MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0
#export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_BUFFER_SZ=1048576
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_THRESHOLD=131072
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16
#export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16

#
# Local vol 16.16.16.32
#

#VOL=32.64.64.96

for VOL in 32.32.32.96 32.64.64.96
do
for AT in 32
do
CMD="mpiexec -np 24 -ppn 12  -envall \
	     ./gpu_tile.sh ./Benchmark_dwf_fp32 --mpi 2.2.2.3 --grid $VOL \
		--shm-mpi 0 --shm 2048 --device-mem 32000 --accelerator-threads $AT --comms-overlap "

echo $CMD
$CMD
done
done