Merge branch 'develop' into feature/scidac-wp1

systems/Aurora/config-command

@@ -1,16 +1,16 @@
-TOOLS=$HOME/tools
+
 ../../configure \
 	--enable-simd=GPU \
 	--enable-gen-simd-width=64 \
 	--enable-comms=mpi-auto \
-	--enable-accelerator-cshift \
 	--disable-gparity \
 	--disable-fermion-reps \
 	--enable-shm=nvlink \
 	--enable-accelerator=sycl \
+	--enable-accelerator-aware-mpi=no\
 	--enable-unified=no \
 	MPICXX=mpicxx \
 	CXX=icpx \
-	LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$TOOLS/lib64/ -L${MKLROOT}/lib -qmkl=parallel " \
-	CXXFLAGS="-fiopenmp -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-compare -I$HOME/ -I$TOOLS/include -qmkl=parallel"
+	LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L${MKLROOT}/lib -qmkl=parallel -lsycl" \
+	CXXFLAGS="-fiopenmp -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-compare -I$HOME/ -qmkl=parallel"
 

systems/Aurora/sourceme-sunspot-deterministic.sh

@@ -0,0 +1,2 @@
+module load oneapi/eng-compiler/2023.05.15.003
+module load mpich/51.2/icc-all-deterministic-pmix-gpu

systems/Aurora/tests/repro128.pbs

@@ -0,0 +1,41 @@
+#!/bin/bash
+
+## qsub -q EarlyAppAccess -A Aurora_Deployment -I -l select=1 -l walltime=60:00
+
+#PBS -q EarlyAppAccess
+#PBS -l select=128
+#PBS -l walltime=02:00:00
+#PBS -A LatticeQCD_aesp_CNDA
+
+#export OMP_PROC_BIND=spread
+#unset OMP_PLACES
+
+cd $PBS_O_WORKDIR
+
+source ../sourceme.sh
+
+cat $PBS_NODEFILE
+
+export OMP_NUM_THREADS=3
+export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
+
+#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE
+#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE
+#unset MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST
+
+export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0
+export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0
+export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1
+export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_BUFFER_SZ=1048576
+export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_THRESHOLD=131072
+export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16
+export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16
+export MPICH_OFI_NIC_POLICY=GPU
+
+# 12 ppn, 16 nodes, 192 ranks
+# 12 ppn, 128 nodes, 1536 ranks
+CMD="mpiexec -np 1536 -ppn 12  -envall \
+	     ./gpu_tile_compact.sh \
+	     ./Test_dwf_mixedcg_prec --mpi 4.4.4.24 --grid 128.128.128.384 \
+		--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 7000 --comms-overlap "
+$CMD 

systems/Aurora/tests/repro16.pbs

@@ -2,26 +2,39 @@
 
 ## qsub -q EarlyAppAccess -A Aurora_Deployment -I -l select=1 -l walltime=60:00
 
-#PBS -q EarlyAppAccess
-#PBS -l select=16
-#PBS -l walltime=01:00:00
+#PBS -l select=16:system=sunspot,place=scatter
 #PBS -A LatticeQCD_aesp_CNDA
+#PBS -l walltime=01:00:00
+#PBS -N dwf
+#PBS -k doe
 
 #export OMP_PROC_BIND=spread
 #unset OMP_PLACES
 
 cd $PBS_O_WORKDIR
 
-source ../sourceme.sh
+#source ../sourceme.sh
 
 cat $PBS_NODEFILE
 
+#export MPICH_COLL_SYNC=1
+#export MPICH_ENV_DISPLAY=1
+export MPICH_
 export OMP_NUM_THREADS=3
 export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
+module load oneapi/eng-compiler/2023.05.15.003
+module load mpich/51.2/icc-all-deterministic-pmix-gpu
+#export LD_LIBRARY_PATH=/soft/restricted/CNDA/updates/2023.05.15.001/oneapi/compiler/eng-20230512/compiler/linux/lib/:$LD_LIBRARY_PATH
 
 #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_ALLREDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_REDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_ALLREDUCE_INTRA_ALGORITHM=recursive_doubling
+unset MPIR_CVAR_CH4_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_CH4_POSIX_COLL_SELECTION_TUNING_JSON_FILE
 
 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0
 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0
@@ -32,9 +45,17 @@ export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16
 export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16
 export MPICH_OFI_NIC_POLICY=GPU
 
-# 12 ppn, 16 nodes, 192 ranks
+DIR=repro.$PBS_JOBID
+mkdir $DIR
+cd $DIR
+
 CMD="mpiexec -np 192 -ppn 12  -envall \
-	     ./gpu_tile_compact.sh \
-	     ./Test_dwf_mixedcg_prec --mpi 2.4.4.6 --grid 64.128.128.192 \
-		--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 3000"
+	     ../gpu_tile_compact.sh \
+	     ../Test_dwf_mixedcg_prec --mpi 2.4.4.6 --grid 64.128.128.192 \
+		--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 3000 --debug-stdout --log Message,Iterative"
+#--comms-overlap
 $CMD 
+
+grep Oops Grid.stderr.* > failures.$PBS_JOBID
+rm core.*
+

systems/Aurora/tests/repro1gpu.pbs

@@ -0,0 +1,82 @@
+#!/bin/bash
+
+#PBS -l select=16:system=sunspot,place=scatter
+#PBS -A LatticeQCD_aesp_CNDA
+#PBS -l walltime=02:00:00
+#PBS -N repro1gpu
+#PBS -k doe
+
+#export OMP_PROC_BIND=spread
+#unset OMP_PLACES
+
+module load oneapi/eng-compiler/2023.05.15.003
+module load mpich/51.2/icc-all-deterministic-pmix-gpu
+
+# 56 cores / 6 threads ~9
+export OMP_NUM_THREADS=6
+export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
+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
+export MPICH_OFI_NIC_POLICY=GPU
+
+export MPIR_CVAR_ALLREDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_REDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_ALLREDUCE_INTRA_ALGORITHM=recursive_doubling
+unset MPIR_CVAR_CH4_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_CH4_POSIX_COLL_SELECTION_TUNING_JSON_FILE
+export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file"
+
+cd $PBS_O_WORKDIR
+
+NN=`cat $PBS_NODEFILE | wc -l`
+echo $PBS_NODEFILE
+cat $PBS_NODEFILE
+
+echo $NN nodes in node file
+for n in `eval echo {1..$NN}`
+do
+
+THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
+echo Node $n is $THIS_NODE
+
+
+for g in {0..11}
+do
+export NUMA_MAP=(0 0 0 1 1 1 0 0 0 1 1 1 )
+export TILE_MAP=(0 0 0 0 0 0 1 1 1 1 1 1 )
+export  GPU_MAP=(0 1 2 3 4 5 0 1 2 3 4 5 )
+
+export numa=${NUMA_MAP[$g]}
+export gpu_id=${GPU_MAP[$g]}
+export tile_id=${TILE_MAP[$g]}
+export gpu=$gpu_id.$tile_id
+
+cd $PBS_O_WORKDIR
+
+DIR=repro.1gpu.$PBS_JOBID/node-$n-$THIS_NODE-GPU-$gpu
+mkdir -p $DIR
+cd $DIR
+
+echo $THIS_NODE > nodefile
+echo $gpu > gpu
+
+export ZE_AFFINITY_MASK=$gpu
+export ONEAPI_DEVICE_FILTER=gpu,level_zero
+
+CMD="mpiexec -np 1 -ppn 1  -envall --hostfile nodefile \
+	     numactl -N $numa -m $numa ../../Test_dwf_mixedcg_prec --mpi 1.1.1.1 --grid 16.16.32.32 \
+		--shm-mpi 0 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message"
+echo $CMD
+$CMD &
+
+done
+done
+
+wait
+

systems/Aurora/tests/reproN.pbs

@@ -0,0 +1,98 @@
+#!/bin/bash
+
+#PBS -l select=32:system=sunspot,place=scatter
+#PBS -A LatticeQCD_aesp_CNDA
+#PBS -l walltime=02:00:00
+#PBS -N reproN
+#PBS -k doe
+
+#export OMP_PROC_BIND=spread
+#unset OMP_PLACES
+
+module load oneapi/eng-compiler/2023.05.15.003
+module load mpich/51.2/icc-all-deterministic-pmix-gpu
+
+# 56 cores / 6 threads ~9
+export OMP_NUM_THREADS=6
+export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
+#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_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
+#export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1
+
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=1
+export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE_FOR_D2D_COPY=1
+export SYCL_PROGRAM_COMPILE_OPTIONS="-ze-opt-large-register-file"
+
+export GRID_PRINT_ENTIRE_LOG=0
+export GRID_CHECKSUM_RECV_BUF=0
+export GRID_CHECKSUM_SEND_BUF=0
+
+export MPICH_OFI_NIC_POLICY=GPU
+
+export MPIR_CVAR_ALLREDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_REDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_ALLREDUCE_INTRA_ALGORITHM=recursive_doubling
+unset MPIR_CVAR_CH4_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_CH4_POSIX_COLL_SELECTION_TUNING_JSON_FILE
+
+cd $PBS_O_WORKDIR
+
+NN=`cat $PBS_NODEFILE | wc -l`
+echo $PBS_NODEFILE
+cat $PBS_NODEFILE
+
+echo $NN nodes in node file
+for n in `eval echo {1..$NN}`
+do
+
+cd $PBS_O_WORKDIR
+
+THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
+echo Node $n is $THIS_NODE
+
+DIR=reproN.$PBS_JOBID/node-$n-$THIS_NODE
+
+mkdir -p $DIR
+cd $DIR
+
+echo $THIS_NODE > nodefile
+
+#CMD="mpiexec -np 12 -ppn 12  -envall --hostfile nodefile \
+#	     ../../gpu_tile_compact.sh \
+#	     ../../Test_dwf_mixedcg_prec --mpi 1.2.2.3 --grid 32.64.64.96 \
+#		--shm-mpi 0 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --comms-overlap"
+
+CMD="mpiexec -np 12 -ppn 12  -envall --hostfile nodefile \
+	     ../../gpu_tile_compact.sh \
+	     ../../Test_dwf_mixedcg_prec --mpi 1.2.2.3 --grid 32.64.64.96 \
+		--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --comms-overlap"
+
+echo $CMD > command-line
+env > environment
+$CMD &
+
+done
+
+# Suspicious wait is allowing jobs to collide and knock out
+#wait
+
+sleep 6500
+
+for n in ` eval echo {1..$NN} `
+do
+
+THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
+DIR=reproN.$PBS_JOBID/node-$n-$THIS_NODE
+
+cd $DIR
+
+grep Oops Grid.stderr.* > failures.$PBS_JOBID
+rm core.*
+
+done

systems/Aurora/tests/solver/stag16.pbs

@@ -36,5 +36,5 @@ export MPICH_OFI_NIC_POLICY=GPU
 CMD="mpiexec -np 192 -ppn 12  -envall \
 	     ./gpu_tile_compact.sh \
 	     ./Test_staggered_cg_prec --mpi 2.4.4.6 --grid 128.128.128.192 \
-	     --shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 3000"
+	     --shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 3000 --comms-overlap"
 $CMD 

systems/Sunspot/config-command

@@ -1,4 +1,4 @@
-TOOLS=$HOME/tools
+
 ../../configure \
 	--enable-simd=GPU \
 	--enable-gen-simd-width=64 \
@@ -11,6 +11,6 @@ TOOLS=$HOME/tools
 	--enable-unified=no \
 	MPICXX=mpicxx \
 	CXX=icpx \
-	LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L$TOOLS/lib64/" \
-	CXXFLAGS="-fiopenmp -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-compare -I$HOME/ -I$TOOLS/include"
+	LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -L${MKLROOT}/lib -qmkl=parallel -lsycl" \
+	CXXFLAGS="-fiopenmp -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-compare -I$HOME/ -qmkl=parallel"
 

systems/Sunspot/sourceme.sh

@@ -0,0 +1,2 @@
+module load oneapi/eng-compiler/2023.05.15.003
+module load mpich/51.2/icc-all-deterministic-pmix-gpu

systems/Sunspot/tests/repro1gpu.pbs

@@ -0,0 +1,81 @@
+#!/bin/bash
+
+#PBS -l select=16:system=sunspot,place=scatter
+#PBS -A LatticeQCD_aesp_CNDA
+#PBS -l walltime=02:00:00
+#PBS -N repro1gpu
+#PBS -k doe
+
+#export OMP_PROC_BIND=spread
+#unset OMP_PLACES
+
+module load oneapi/eng-compiler/2023.05.15.003
+module load mpich/51.2/icc-all-deterministic-pmix-gpu
+
+# 56 cores / 6 threads ~9
+export OMP_NUM_THREADS=6
+export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
+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
+export MPICH_OFI_NIC_POLICY=GPU
+
+export MPIR_CVAR_ALLREDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_REDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_ALLREDUCE_INTRA_ALGORITHM=recursive_doubling
+unset MPIR_CVAR_CH4_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_CH4_POSIX_COLL_SELECTION_TUNING_JSON_FILE
+
+cd $PBS_O_WORKDIR
+
+NN=`cat $PBS_NODEFILE | wc -l`
+echo $PBS_NODEFILE
+cat $PBS_NODEFILE
+
+echo $NN nodes in node file
+for n in `eval echo {1..$NN}`
+do
+
+THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
+echo Node $n is $THIS_NODE
+
+
+for g in {0..11}
+do
+export NUMA_MAP=(0 0 0 1 1 1 0 0 0 1 1 1 )
+export TILE_MAP=(0 0 0 0 0 0 1 1 1 1 1 1 )
+export  GPU_MAP=(0 1 2 3 4 5 0 1 2 3 4 5 )
+
+export numa=${NUMA_MAP[$g]}
+export gpu_id=${GPU_MAP[$g]}
+export tile_id=${TILE_MAP[$g]}
+export gpu=$gpu_id.$tile_id
+
+cd $PBS_O_WORKDIR
+
+DIR=repro.1gpu.$PBS_JOBID/node-$n-$THIS_NODE-GPU-$gpu
+mkdir -p $DIR
+cd $DIR
+
+echo $THIS_NODE > nodefile
+echo $gpu > gpu
+
+export ZE_AFFINITY_MASK=$gpu
+export ONEAPI_DEVICE_FILTER=gpu,level_zero
+
+CMD="mpiexec -np 1 -ppn 1  -envall --hostfile nodefile \
+	     numactl -N $numa -m $numa ../../Test_dwf_mixedcg_prec --mpi 1.1.1.1 --grid 16.16.32.32 \
+		--shm-mpi 0 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message"
+echo $CMD
+$CMD &
+
+done
+done
+
+wait
+

systems/Sunspot/tests/reproN.pbs

@@ -0,0 +1,97 @@
+#!/bin/bash
+
+#PBS -l select=32:system=sunspot,place=scatter
+#PBS -A LatticeQCD_aesp_CNDA
+#PBS -l walltime=02:00:00
+#PBS -N reproN
+#PBS -k doe
+
+#export OMP_PROC_BIND=spread
+#unset OMP_PLACES
+
+module load oneapi/eng-compiler/2023.05.15.003
+module load mpich/51.2/icc-all-deterministic-pmix-gpu
+
+# 56 cores / 6 threads ~9
+export OMP_NUM_THREADS=6
+export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
+#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_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
+#export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1
+
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=1
+export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE_FOR_D2D_COPY=1
+
+export GRID_PRINT_ENTIRE_LOG=0
+export GRID_CHECKSUM_RECV_BUF=1
+export GRID_CHECKSUM_SEND_BUF=0
+
+export MPICH_OFI_NIC_POLICY=GPU
+
+export MPIR_CVAR_ALLREDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_REDUCE_DEVICE_COLLECTIVE=0
+export MPIR_CVAR_ALLREDUCE_INTRA_ALGORITHM=recursive_doubling
+unset MPIR_CVAR_CH4_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_COLL_SELECTION_TUNING_JSON_FILE
+unset MPIR_CVAR_CH4_POSIX_COLL_SELECTION_TUNING_JSON_FILE
+
+cd $PBS_O_WORKDIR
+
+NN=`cat $PBS_NODEFILE | wc -l`
+echo $PBS_NODEFILE
+cat $PBS_NODEFILE
+
+echo $NN nodes in node file
+for n in `eval echo {1..$NN}`
+do
+
+cd $PBS_O_WORKDIR
+
+THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
+echo Node $n is $THIS_NODE
+
+DIR=reproN.$PBS_JOBID/node-$n-$THIS_NODE
+
+mkdir -p $DIR
+cd $DIR
+
+echo $THIS_NODE > nodefile
+
+#CMD="mpiexec -np 12 -ppn 12  -envall --hostfile nodefile \
+#	     ../../gpu_tile_compact.sh \
+#	     ../../Test_dwf_mixedcg_prec --mpi 1.2.2.3 --grid 32.64.64.96 \
+#		--shm-mpi 0 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --comms-overlap"
+
+CMD="mpiexec -np 12 -ppn 12  -envall --hostfile nodefile \
+	     ../../gpu_tile_compact.sh \
+	     ../../Test_dwf_mixedcg_prec --mpi 1.2.2.3 --grid 32.64.64.96 \
+		--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --comms-overlap"
+
+echo $CMD > command-line
+env > environment
+$CMD &
+
+done
+
+# Suspicious wait is allowing jobs to collide and knock out
+#wait
+
+sleep 6500
+
+for n in ` eval echo {1..$NN} `
+do
+
+THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
+DIR=reproN.$PBS_JOBID/node-$n-$THIS_NODE
+
+cd $DIR
+
+grep Oops Grid.stderr.* > failures.$PBS_JOBID
+rm core.*
+
+done