jacobi_solution/gpu/openacc added

jacobi/gpu/openacc/make.def

-CC     = gcc
-CFLAGS = -Wall -Wextra -march=native
-LIBS   = -lm
-
-SYSTYPE = $(strip $(shell uname -n))
-
-PROG           = jacobi_serial_opt_$(SYSTYPE)
-PROG_DEBUG      = $(PROG)_DEBUG
-PROG_MEMCHECK   = $(PROG)_MEMCHECK
-PROG_CALLGRIND  = $(PROG)_CALLGRIND
-PROG_CACHEGRIND = $(PROG)_CACHEGRIND
-
-HEADERS       = $(wildcard ./include/*.h)
-SOURCES       = $(wildcard ./src/*.c)
-DEPENDENCIES  = $(SOURCES) $(HEADERS) Makefile
-
-$(PROG): $(DEPENDENCIES)
-	$(CC) $(CFLAGS) -O3 -I./include $(SOURCES) -o $@ $(LIBS)
-	@echo ' '
-	@echo 'Program' $(PROG) 'compiled for' $(SYSTYPE) 'machine'
-	@echo ' '
-
-$(PROG_DEBUG): $(DEPENDENCIES)
-	$(CC) $(CFLAGS) -Og -ggdb3 -fno-omit-frame-pointer -I./include $(SOURCES) -o $@ $(LIBS)
-	@echo ' '
-	@echo 'Program' $(PROG_DEBUG) 'compiled for' $(SYSTYPE) 'machine'
-	@echo ' '
-
-$(PROG_MEMCHECK): $(DEPENDENCIES)
-	$(CC) $(CFLAGS) -Og -I./include $(SOURCES) -o $@ $(LIBS)
-	@echo ' '
-	@echo 'Program' $(PROG_MEMCHECK) 'compiled for' $(SYSTYPE) 'machine'
-	@echo ' '
-
-$(PROG_CALLGRIND): $(DEPENDENCIES)
-	$(CC) $(CFLAGS) -g -O3 -I./include $(SOURCES) -o $@ $(LIBS)
-	@echo ' '
-	@echo 'Program' $(PROG_CALLGRIND) 'compiled for' $(SYSTYPE) 'machine'
-	@echo ' '
-
-$(PROG_CACHEGRIND): $(DEPENDENCIES)
-	$(CC) $(CFLAGS) -g -O3 -I./include $(SOURCES) -o $@ $(LIBS)
-	@echo ' '
-	@echo 'Program' $(PROG_CACHEGRIND) 'compiled for' $(SYSTYPE) 'machine'
-	@echo ' '

jacobi/gpu/openacc/Makefile → jacobi_solutions/gpu/openacc/Makefile

@@ -11,53 +11,24 @@
 include make.def
 #######################################################################
 
-.PHONY: info serial valgrind_memcheck valgrind_callgrind valgrind_cachegrind debug clean 
+.PHONY: info openacc serial clean
 
 info:
 	@echo ' '
 	@echo '-----------------------------------------------------------------------------------------------'
+	@echo '$$ make OpenACC                ---> compile the OpenACC application                            '
 	@echo '$$ make serial                 ---> compile the serial application                             '
-	@echo '$$ make debug                  ---> debug the serial application                               '
-	@echo '$$ make valgrind_memcheck      ---> run the serial application using Valgrind under Memcheck   '
-	@echo '$$ make valgrind_callgrind     ---> run the serial application using Valgrind under Callgrind  '
-	@echo '$$ make valgrind_cachegrind    ---> run the serial application using Valgrind under Cachegrind '
-	@echo '$$ make clean                  ---> clean up all                                               '
 	@echo '$$ make info                   ---> get make info                                              '
 	@echo '-----------------------------------------------------------------------------------------------'
 	@echo ' '
 
-serial: $(PROG)
+openacc: $(PROG)
 
-debug: $(PROG_DEBUG)
-	@echo 'OOOooo... debugging ...oooOOO'
-	gdb --args ./$<
-	@echo 'OOOooo... debugging ... oooOOO'
-
-valgrind_memcheck: $(PROG_MEMCHECK)
-	@echo 'oooOOO... valgrind_memcheck ...OOOooo'
-	valgrind --tool=memcheck -s --leak-check=full --show-leak-kinds=all --track-origins=yes --read-var-info=yes --log-file=valgrind_memcheck_log_%p.txt ./$< 10 10
-	@echo 'oooOOO... valgrind_memcheck ...OOOooo'
-
-valgrind_callgrind: $(PROG_CALLGRIND)
-	@echo 'oooOOO... valgrind_callgrind ...OOOooo'
-	valgrind --tool=callgrind --dump-instr=yes --collect-jumps=yes --log-file=valgrind_callgrind_log_.%p.txt ./$< 128 128
-	@echo ' '
-	@echo 'To generate a function-by-function summary from the profile data file:'
-	@echo '$$ callgrind_annotate --auto=yes callgrind.out.<pid> | less'
-	@echo '(kcachegrind is required in order to visualize the output using the GUI)'
-
-valgrind_cachegrind: $(PROG_CACHEGRIND)
-	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'
-	valgrind --tool=cachegrind --cache-sim=yes --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128
-	@echo '$$ cg_annotate --auto=yes cachegrind.out.<pid> | less'
-	@echo '(kcachegrind is required in order to visualize the output using the GUI)'
-	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'
+serial: $(PROG_SERIAL)
 
 clean:
 	rm -f *~ .*~ ./src/*~ ./src/*# ./include/*~ ./include/*# *~
-	rm -f $(PROG) $(PROG_DEBUG) $(PROG_MEMCHECK) $(PROG_CALLGRIND) $(PROG_CACHEGRIND)
-	rm -f valgrind_*.txt
-	rm -f cachegrind.out.*
-	rm -f callgrind.*
+	rm -f $(PROG) $(PROG_SERIAL)
 	rm -f *bin
-	rm -f jacobi_serial_opt_*
+	rm -f jacobi_OpeACC*
+	rm -rf jacobi_serial*

jacobi_solutions/gpu/openacc/make.def

+CC             = gcc
+CC_OPENACC     = pgcc
+CFLAGS_OPENACC = -acc -fast -Minfo=accel -ta=tesla:cc80 -ta=tesla:maxregcount:32 -ta=tesla:ptxinfo -g
+CFLAGS         = -Wall -Wextra -march=native -mtune=native
+LIBS           = -lm
+
+SYSTYPE = $(strip $(shell uname -n))
+
+PROG          = jacobi_OpenACC_$(SYSTYPE)
+PROG_SERIAL   = jacobi_serial_$(SYSTYPE)
+
+HEADERS       = $(wildcard ./include/*.h)
+SOURCES       = $(wildcard ./src/*.c)
+DEPENDENCIES  = $(SOURCES) $(HEADERS) Makefile
+
+$(PROG): $(DEPENDENCIES)
+	$(CC_OPENACC) $(CFLAGS_OPENACC) -O3 -I./include $(SOURCES) -o $@ $(LIBS)
+	@echo ' '
+	@echo 'Program' $(PROG) 'compiled for' $(SYSTYPE) 'machine'
+	@echo ' '
+
+$(PROG_SERIAL): $(DEPENDENCIES)
+	$(CC) $(CFLAGS) -O3 -I./include $(SOURCES) -o $@ $(LIBS)
+	@echo ' '
+	@echo 'Program' $(PROG_SERIAL) 'compiled for' $(SYSTYPE) 'machine'
+	@echo ' '

jacobi_solutions/gpu/openacc/script/leonardo.sbatch

+#!/bin/bash
+
+#SBATCH --job-name=Jacobi-OpenACC
+#SBATCH --account=IscrC_SCGPCT
+#SBATCH --partition=boost_usr_prod
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --cpus-per-task=1
+#SBATCH --gres=gpu:1
+#SBATCH --output=Jacobi-OpenACC-%j.out
+#SBATCH --error=Jacobi-OpenACC-%j.err
+#SBATCH --time=0:10:00
+
+# module purge
+# module load scorep/8.1--openmpi--4.1.4--nvhpc--23.1-cuda-11.8
+# module load openmpi/4.1.6--nvhpc--23.11 nvhpc/23.11
+module load nvhpc/23.11
+
+SCOREP= #"scorep --user"
+CC = "nvcc"
+OPT = 
+#OPT="-g -mp=gpu -gpu=ccnative,debug,lineinfo -target=gpu -Minfo=all -v"
+#LIB="-L/leonardo/prod/opt/libraries/openmpi/4.1.6/nvhpc--24.3/lib/ -lmpi -lm"
+LIB = "-lm"
+
+COMPUTE_SANITIZER="compute-sanitizer --generate-coredump "
+
+### script path
+SCRIPT_PATH=/leonardo/home/userexternal/dgoz0000/test/openmp-gpu/script
+
+################## source list (C CPP)
+# on Leonardo compile the CPP code
+SOURCE_CODE=( CPP )
+
+#################################################################################
+#################################################################################
+if [ ! -d ${SCRIPT_PATH} ]
+then
+    printf "\n\t SCRIPT_PATH invalid... aborting..."
+    exit 0
+fi
+
+printf "\n\t SLURM:"
+printf "\n\t\t NODES       : ${SLURM_NNODES} (${SLURM_NODELIST})"
+printf "\n\t\t NTASKS      : ${SLURM_NTASKS}"
+printf "\n\t\t CPU-PER-TASK: ${SLURM_CPUS_PER_TASK}"
+printf "\n\t\t GPUs        : ${SLURM_JOB_GPUS}"
+printf "\n\n"
+
+# working directory
+WORKDIR=${PWD}
+
+# compile the sources
+source ${SCRIPT_PATH}/compile_OpenMP-GPU.sh
+
+# run the executable getting the profile using score-p
+source ${SCRIPT_PATH}/run_OpenMP-GPU.sh
+
+printf "\n\t End of game!!! \n"

jacobi/gpu/openacc/src/jacobi_2D_OpenACC.c → jacobi_solutions/gpu/openacc/src/jacobi_2D_OpenACC.c

@@ -214,7 +214,7 @@ void JacobiAlgorithm(MyData      **const restrict Phi,
 {
   double gpu_err=0.0;
 
-#pragma acc parallel loop reduction(+: err)
+#pragma acc parallel loop reduction(+: gpu_err)
   for (int j=jbeg ; j<=jend ; j++)
     {
       for (int i=ibeg ; i<=iend ; i++)
@@ -226,7 +226,7 @@ void JacobiAlgorithm(MyData      **const restrict Phi,
 
 	  /* avoid fabs from math library */
 	  const MyData diff = (Phi[j][i] - Phi0[j][i]);
-	  *error += ((diff > 0) ? diff : -diff);
+	  gpu_err += ((diff > 0) ? diff : -diff);
 	} /* loop over columns */
     } /* loop over rows */
   

jacobi/gpu/openacc/src/tools.c → jacobi_solutions/gpu/openacc/src/tools.c

@@ -16,7 +16,7 @@ MyData **Allocate_2DdblArray(const int nx, const int ny)
  * nx rows and ny columns
  *********************************************************************** */
 {
-  MyData **buf = malloc(nx * sizeof(MyData *));
+  MyData **buf = (MyData **)malloc(nx * sizeof(MyData *));
   assert(buf != NULL);
 
   buf[0] = (MyData *) malloc(nx * ny * sizeof(MyData));