added flag to turn on and off mpi thread funneled, implemented H2, no...

added flag to turn on and off mpi thread funneled, implemented H2, no guarantee for binary equality of results for now

CMakeLists.txt

+cmake_minimum_required(VERSION 2.4)
+project(dadp)
+
+set(CMAKE_C_COMPILER "mpicc")
+add_compile_options("-O3")
+add_compile_options("-g")
+add_compile_options("-fopenmp")
+link_libraries("-lm")
+link_libraries("-fopenmp")
+
+include_directories(${PROJECT_SOURCE_DIR}/include)
+
+add_executable(main src/main/main.c src/tree/tree.c src/common/common.c src/tree/kdtreeV2.c src/tree/heap.c)
+
+

Makefile

 CC=mpicc
-CFLAGS=-O3 -g
-LDFLAGS=-lm -fopenmp
+CFLAGS=-O3 -g -fopenmp
+LDFLAGS=-lm 
 
 all: main
 

run_pleiadi

 #!/bin/bash
 
-#SBATCH --nodes=2
+#SBATCH --nodes=1
 #SBATCH --ntasks-per-node=2
 #SBATCH --cpus-per-task=18
 #SBATCH --time=01:00:00
@@ -26,6 +26,7 @@ rm bb/*
 
 #time mpirun -n ${SLURM_NTASKS} --map-by ppr:1:node:PE=${SLURM_CPUS_PER_TASK}  main
 time mpirun -n ${SLURM_NTASKS} --mca orte_base_help_aggregate 0 --map-by ppr:1:socket:PE=${SLURM_CPUS_PER_TASK}  main
+#time mpirun -n ${SLURM_NTASKS} --mca orte_base_help_aggregate 0 --map-by core  main
 #time mpirun -n ${SLURM_NTASKS} main
 
 #time python3 check.py

run_pleiadi_no_openmp

+#!/bin/bash
+
+#SBATCH --nodes=8
+#SBATCH --ntasks-per-node=36
+#SBATCH --cpus-per-task=1
+#SBATCH --time=01:00:00
+#SBATCH --job-name=dADP-test
+#SBATCH --account=ulearn
+#SBATCH --partition=pleiadi
+#SBATCH --output=out_pleiadi
+#SBATCH --error=err_pleiadi
+#SBATCH --mem=230G
+
+cd $SLURM_SUBMIT_DIR
+module restore dev_pleiadi
+source /u/ftomba/my_envs/dadac-dev/bin/activate
+make clean
+make
+
+#export PSM2_MQ_SENDREQS_MAX=268435456
+#export PSM2_MQ_RECVREQS_MAX=268435456
+
+rm bb/*
+
+time mpirun -n ${SLURM_NTASKS} --mca orte_base_help_aggregate 0 --map-by core main
+
+#time python3 check.py
+

src/common/common.h

@@ -35,6 +35,9 @@ typedef struct datapoint_info_t {
 
 #define CHECK_ALLOCATION(x) if(!x){printf("[!!!] %d rank encountered failed allocation at line %s ", ctx -> mpi_rank, __LINE__ ); exit(1);};
 
+#define CHECK_ALLOCATION_NO_CTX(x) if(!x){printf("[!!!] Failed allocation at line %d ", __LINE__ ); exit(1);}
+#define MY_MALLOC(n) ({void* p = calloc(n,1); CHECK_ALLOCATION_NO_CTX(p); p; })
+
 #define DB_PRINT(...) printf(__VA_ARGS__)
 #ifdef NDEBUG
 	#undef DB_PRINT(...)
@@ -64,6 +67,7 @@ typedef struct datapoint_info_t {
         (clock_gettime(CLOCK_MONOTONIC,&__end), \
         (double)(__end.tv_sec - __start.tv_sec) + (__end.tv_nsec - __start.tv_nsec)/1e9)
     #define LOG_WRITE(sec_name,time) { \
+        MPI_Barrier(ctx -> mpi_communicator); \
         if(time > 0) \
         { \
             double max, min, avg; \

src/main/main.c

@@ -4,7 +4,13 @@
 #include "../common/common.h"
 #include "../tree/tree.h"
 
+//
+
+#ifdef THREAD_FUNNELED
     #define THREAD_LEVEL MPI_THREAD_FUNNELED
+#else
+    #define THREAD_LEVEL MPI_THREAD_MULTIPLE
+#endif
 
 int main(int argc, char** argv) {
     #if defined (_OPENMP)
@@ -50,6 +56,12 @@ int main(int argc, char** argv) {
         mpi_printf(&ctx,"Running pure MPI code\n");
     #endif
 
+    #if defined (THREAD_FUNNELED)
+        mpi_printf(&ctx,"/!\\ Code build with MPI_THREAD_FUNNELED level\n");
+    #else
+        mpi_printf(&ctx,"/!\\ Code build with MPI_THREAD_MULTIPLE level\n");
+    #endif
+
 	/*
 	 * Mock reading some files, one for each processor
 	 */