02-beginners/dot-product/.config/run-once.sh

 PROGRAM=dot-product-p
 
-export NX_SMP_WORKERS=1
+export NX_SMP_WORKERS=4
 
 ./$PROGRAM 8192 128
 

02-beginners/matmul/.config/run-once.sh

 PROGRAM=matmul-p
 
-export NX_SMP_WORKERS=1
+export NX_SMP_WORKERS=4
 
 ./$PROGRAM 16
 

02-beginners/multisort/.config/run-once.sh

 PROGRAM=multisort-p
 
-export NX_SMP_WORKERS=1
+export NX_SMP_WORKERS=4
 
 ./$PROGRAM 65536 256 512
 

03-gpu-devices/nbody-cuda/.config/multirun.sh

 PROGRAM=nbody-p
 
 export NX_SMP_WORKERS=1
-export NX_GPUS=2 #change this in order to use more GPUs
 
+for gpus in 1 2; do
+  export NX_GPUS=$gpus
   NX_ARGS="--cache-policy writethrough --gpu-max-memory 90" ./$PROGRAM nbody_input-16384.in
+done
 

03-gpu-devices/saxpy-cuda/.config/multirun.sh

 PROGRAM=saxpy-p
 
 export NX_GPUMAXMEM=90
-export NX_GPUS=2 #change this in order to use more GPUs
 
+for gpus in 1 2; do
+  export NX_GPUS=$gpus
   ./$PROGRAM
+done
 

04-mpi+ompss/heat/.config/multirun.sh

 PROGRAM=heat-mpi-ompss-p
 
-export IFS=";"
-
-THREADS="01;02;03;04;05;06;07;08;09;10;11;12"
+THREADS=(1 2 3 4 5 6)
 INPUT=test.dat
 
-for thread in $THREADS; do
-  NX_SMP_WORKERS=$thread mpirun ./$PROGRAM $INPUT test.ppm
+for thread in ${THREADS[@]}; do
+  NX_SMP_WORKERS=$thread ${MPIRUN_COMMAND} ./$PROGRAM $INPUT test.ppm
 done

04-mpi+ompss/heat/.config/run-once.sh

-PROGRAM=heat-mpi-ompss-i
+PROGRAM=heat-mpi-ompss-p
 
-# Run with 2 threads per MPI process in the same node
-
-export SMP_NUM_WORKERS=2
+# Run with 6 threads per MPI process in the same node
+export NX_SMP_WORKERS=6
 
 # Uncomment to instrument
 #export INST=./graph.sh
 #export INST=./trace.sh
 
-mpirun $INST ./$PROGRAM test.dat test.ppm
+${MPIRUN_COMMAND} $INST ./$PROGRAM test.dat test.ppm
+
+# Generate the trace if needed
+if [[ "$INST" == *"trace"* ]]; then
+	mpi2prv -f TRACE.mpits -o myTrace.prv
+fi

04-mpi+ompss/heat/Makefile

@@ -4,6 +4,7 @@ PREFIX=.
 TARGETS=$(PROGRAM)-p $(PROGRAM)-i $(PROGRAM)-d
 
 JOB_SCHED_VERSION=-mpi
+TRACE_VERSION=-mpi
 
 BASE_DIR=../..
 include $(BASE_DIR)/common-files/Makefile

04-mpi+ompss/heat/graph.sh

-#!/bin/bash
-
-# Uncomment following line to trace MPI programs
-
-export NX_INSTRUMENTATION=new-graph
-
-$*

04-mpi+ompss/matmul/.config/multirun.sh

 PROGRAM=matmul-p
 
-export NX_ARGS="--pes 2 --disable-cuda=yes"
-
-echo "NX_ARGS is $NX_ARGS"
-echo "LD_PRELOAD is $LD_PRELOAD"
-echo "LD_LIBRARY_PATH is $LD_LIBRARY_PATH"
-echo "##################################################"
-
-srun ./$PROGRAM
+THREADS=(1 2 3 4 5 6)
 
+for thread in ${THREADS[@]}; do
+	NX_SMP_WORKERS=$thread ${MPIRUN_COMMAND} ./$PROGRAM
+done

04-mpi+ompss/matmul/.config/run-once.sh

-PROGRAM=matmul-i
+PROGRAM=matmul-p
 
-export NX_ARGS="--threads 2"
+# Run with 6 threads per MPI process in the same node
+export NX_SMP_WORKERS=6
 
-echo "NX_ARGS is $NX_ARGS"
-echo "LD_PRELOAD is $LD_PRELOAD"
-echo "LD_LIBRARY_PATH is $LD_LIBRARY_PATH"
-echo "##################################################"
-
-# Uncomment to enable tracing
+# Uncomment to instrument
+#export INST=./graph.sh
 #export INST=./trace.sh
 
-mpirun --cpus-per-proc 2 $INST ./$PROGRAM
+${MPIRUN_COMMAND} $INST ./$PROGRAM
 
+# Generate the trace if needed
+if [[ "$INST" == *"trace"* ]]; then
+	mpi2prv -f TRACE.mpits -o myTrace.prv
+fi

04-mpi+ompss/matmul/Makefile

@@ -4,6 +4,7 @@ PREFIX=.
 TARGETS=$(PROGRAM)-p $(PROGRAM)-i $(PROGRAM)-d
 
 JOB_SCHED_VERSION=-mpi
+TRACE_VERSION=-mpi
 
 BASE_DIR=../..
 include $(BASE_DIR)/common-files/Makefile

common-files/Makefile

@@ -4,19 +4,23 @@ CONFIG_DIR=.config
 GFLAGS=
 
 PARAVER=extrae.xml trace.sh
+GRAPH=graph.sh
 SCRIPTS=run-once.sh multirun.sh
 
 MPI_CHECK=$(MPI_LIB_DIR)/libmpi.so
 MKL_CHECK=$(MKL_LIB_DIR)/libmkl_sequential.so
 ATLAS_CHECK=$(ATLAS_LIB_DIR)/libatlas.a
 
-all: $(TARGETS) $(SCRIPTS) $(PARAVER)
+all: $(TARGETS) $(SCRIPTS) $(PARAVER) $(GRAPH)
 
 extrae.xml:
 	cp $(COMMON_DIR)/extrae.xml .
 
 trace.sh:
-	cp $(COMMON_DIR)/trace.sh .
+	cp $(COMMON_DIR)/trace$(TRACE_VERSION).sh trace.sh
+
+graph.sh:
+	cp $(COMMON_DIR)/graph.sh .
 
 run-once.sh: $(COMMON_DIR)/run-once.sh $(CONFIG_DIR)/run-once.sh
 	cp  $(COMMON_DIR)/run-once.sh .

common-files/configure_VirtualBox

@@ -10,6 +10,7 @@ export MPI_HOME=/usr/lib/openmpi
 export MPI_LIB_DIR=$MPI_HOME/lib
 export MPI_INC_DIR=$MPI_HOME/include
 export MPI_CC=mpicc
+export MPIRUN_COMMAND="mpirun"
 export MKL_LIB_DIR=/home/user/Builds/mkl/lib/intel64
 export MKL_INC_DIR=/home/user/Builds/mkl/include
 export ATLAS_LIB_DIR=/usr/lib

common-files/configure_default

@@ -11,6 +11,7 @@ export PARAVER_HOME=
 export TEMANEJO_HOME=
 export MPI_LIB_DIR=
 export MPI_INC_DIR=
+export MPIRUN_COMMAND="mpirun"
 export MKL_LIB_DIR=
 export MKL_INC_DIR=
 export ATLAS_LIB_DIR=/home/xteruel/Applications/atlas-3.10.3/lib/

common-files/configure_mn4

@@ -14,6 +14,7 @@ export MPICC=mpiicc
 export MPICXX=mpiicpc
 export MPI_NATIVE_CC=I_MPI_CC
 export MPI_NATIVE_CXX=I_MPI_CXX
+export MPIRUN_COMMAND="mpirun"
 export MKL_LIB_DIR=/apps/INTEL/2017.4/mkl/lib/intel64
 export MKL_INC_DIR=/apps/INTEL/2017.4/mkl/include
 export ATLAS_LIB_DIR=/apps/ATLAS/3.10.3/INTEL_BK/lib

common-files/configure_nord3

@@ -14,6 +14,7 @@ export MPICC=mpiicc
 export MPICXX=mpiicpc
 export MPI_NATIVE_CC=I_MPI_CC
 export MPI_NATIVE_CXX=I_MPI_CXX
+export MPIRUN_COMMAND="mpirun"
 
 # Note (vlopez); wxparaver needs glibcxx from gcc >= 5
 module load gcc/5.1.0

common-files/configure_nvidia

@@ -2,18 +2,19 @@
 
 DIRNAME=$(readlink -f $(dirname ${BASH_SOURCE[0]}))
 
-export OMPSS_HOME=/apps/PM/ompss/17.06/
+export OMPSS_HOME=/apps/PM/ompss/19.06/
 export DLB_HOME=/apps/PM/dlb/latest/bullxmpi
-export EXTRAE_HOME=/apps/BSCTOOLS/extrae/latest/bullxmpi_1_2_9_1
+export EXTRAE_HOME=/apps/BSCTOOLS/extrae/latest/impi_2017_1_132
 export PARAVER_HOME=/apps/BSCTOOLS/wxparaver/latest
 export TEMANEJO_HOME=
-export MPI_HOME=/opt/mpi/bullxmpi/1.1.11.1
-export MPI_LIB_DIR=$MPI_HOME/lib
-export MPI_INC_DIR=$MPI_HOME/include
-export MPICC=mpicc
-export MPICXX=mpicxx
-export MPI_NATIVE_CC=OMPI_CC
-export MPI_NATIVE_CXX=OMPI_CXX
+export MPI_HOME=/apps/INTEL/2017.1-043/impi/2017.1.132
+export MPI_LIB_DIR=$MPI_HOME/lib64
+export MPI_INC_DIR=$MPI_HOME/include64
+export MPICC=mpiicc
+export MPICXX=mpiicpc
+export MPI_NATIVE_CC=I_MPI_CC
+export MPI_NATIVE_CXX=I_MPI_CXX
+export MPIRUN_COMMAND="srun --cpu_bind=cores"
 export MKL_LIB_DIR=/opt/compilers/intel/2016.3.067/mkl/lib/intel64/
 export MKL_INC_DIR=/opt/compilers/intel/2016.3.067/mkl/include
 export ATLAS_LIB_DIR=/gpfs/apps/NVIDIA/ATLAS/3.9.51/lib
@@ -23,8 +24,12 @@ ln -sf $DIRNAME/sched-job-smp_nvidia $DIRNAME/sched-job-smp
 ln -sf $DIRNAME/sched-job-mpi_nvidia $DIRNAME/sched-job-mpi
 
 module unload gcc
-module load gcc/4.9.4
-
+module load gcc/4.6.1
+module load GRAPHVIZ
+module unload bullxmpi
+module load impi/2017.1
+module unload cuda
+module load cuda/8.0
 
 alias submit=mnsubmit
 alias queue=mnq

common-files/graph.sh

+#!/bin/bash
+
+# Instrumentation to generate a task dependency graph
+export NX_INSTRUMENTATION=tdg
+
+$*

common-files/sched-job-mpi_nvidia

@@ -8,4 +8,5 @@
 # @ gpus_per_node = 2
 # @ cpus_per_task = 6
 # @ node_usage = not_shared
+# @ features = k80
 # @ wall_clock_limit = 00:15:00