Fix heat MPI benchmarks

src/heat/mpi/CMakeLists.txt

@@ -10,7 +10,8 @@ endmacro()
 
 macro(mk_heat_mpi_nanos6 NAME SOURCE)
   mk_heat_mpi(${NAME} ${SOURCE})
-  target_link_libraries(${NAME} PRIVATE Nanos6::wrapper)
+  target_compile_options(${NAME} PRIVATE "-fompss-2=libnanos6")
+  target_link_options(${NAME} PRIVATE "-fompss-2=libnanos6")
 endmacro()
 
 macro(mk_heat_mpi_nodes NAME SOURCE)
@@ -18,6 +19,12 @@ macro(mk_heat_mpi_nodes NAME SOURCE)
   target_link_libraries(${NAME} PRIVATE Nodes::wrapper)
 endmacro()
 
+macro(mk_heat_tampi_nanos6 NAME SOURCE)
+  mk_heat_mpi_nanos6(${NAME} ${SOURCE})
+  target_link_libraries(${NAME} PRIVATE Tampi::tampi-c)
+endmacro()
+
+
 # -------------------------------------------------------------------
 
 add_library(heat_mpi_common STATIC main.c utils.c)
@@ -35,3 +42,7 @@ if(NODES_FOUND)
   mk_heat_mpi_nodes(b6_heat_mpi_nodes_forkjoin solver_mpi_ompss2_forkjoin.c)
   mk_heat_mpi_nodes(b6_heat_mpi_nodes_tasks solver_mpi_ompss2_tasks.c)
 endif()
+
+if(NANOS6_FOUND AND TAMPI_FOUND)
+  mk_heat_tampi_nanos6(b6_heat_itampi_nanos6_tasks solver_itampi_ompss2_tasks.c)
+endif()

src/heat/mpi/main.c

@@ -1,4 +1,5 @@
 #include <mpi.h>
+#include <math.h>
 #include <assert.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -19,16 +20,7 @@ void generateImage(const HeatConfiguration *conf, int64_t rows, int64_t cols, in
 
 int main(int argc, char **argv)
 {
-#if defined(TAMPI)
+	const int required = mpi_level();
-	// TAMPI+OmpSs-2 variants
-	const int required = MPI_TASK_MULTIPLE;
-#elif defined(_OMPSS_2)
-	// MPI+OmpSs-2 variants
-	const int required = MPI_THREAD_SERIALIZED;
-#else
-	// MPI-only variants
-	const int required = MPI_THREAD_SINGLE;
-#endif
 
 	int provided;
 	MPI_Init_thread(&argc, &argv, required, &provided);
@@ -76,8 +68,10 @@ int main(int argc, char **argv)
 
 	if (!rank) {
 		int64_t totalElements = conf.rows*conf.cols;
+		//double time_element = (end-start)/(totalElements*conf.timesteps);
 		double throughput = (totalElements*conf.timesteps)/(end-start);
-		throughput = throughput/1000000.0;
+		//throughput = throughput/1000000.0;
+		double residual = NAN;
 
 #ifdef _OMPSS_2
 		int threads = nanos6_get_num_cpus();
@@ -85,10 +79,17 @@ int main(int argc, char **argv)
 		int threads = 1;
 #endif
 
-		fprintf(stdout, "rows, %ld, cols, %ld, rows/rank, %ld, total, %ld, total/rank, %ld, rbs, %d, "
+		fprintf(stderr, "%14s %14s %14s %8s %8s %8s %8s %8s %8s\n",
-				"cbs, %d, ranks, %d, threads, %d, timesteps, %d, time, %f, Mupdates/s, %f\n",
+				"throughput", "time", "error", 
-				conf.rows, conf.cols, conf.rows/nranks, totalElements, totalElements/nranks,
+				"rows", "cols",
-				conf.rbs, conf.cbs, nranks, threads, conf.timesteps, end-start, throughput);
+				"rbs", "cbs", "threads",
+				"steps");
+		fprintf(stdout, "%14e %14e %14e %8ld %8ld %8d %8d %8d %8d\n", 
+				throughput, end-start, residual,
+				conf.rows, conf.cols, 
+				conf.rbs, conf.cbs, threads,
+				conf.convergenceTimesteps);
+
 	}
 
 	if (conf.generateImage) {

src/heat/mpi/solver_itampi_ompss2_tasks.c

@@ -4,6 +4,17 @@
 #include "utils.h"
 #include "common/heat.h"
 
+const char *
+summary(void)
+{
+	return "Parallel version using MPI + OmpSs-2 tasks + Non-blocking TAMPI";
+}
+
+int
+mpi_level(void)
+{
+	return MPI_TASK_MULTIPLE;
+}
 
 static inline void send(const double *data, int nelems, int dst, int tag)
 {
@@ -55,6 +66,7 @@ static inline void gaussSeidelSolver(int64_t rows, int64_t cols, int rbs, int cb
 
 double solve(HeatConfiguration *conf, int64_t rows, int64_t cols, int timesteps, void *extraData)
 {
+	(void) extraData;
 	double (*matrix)[cols] = (double (*)[cols]) conf->matrix;
 	const int rbs = conf->rbs;
 	const int cbs = conf->cbs;

src/heat/mpi/solver_mpi.c

@@ -3,6 +3,12 @@
 #include "utils.h"
 #include "common/heat.h"
 
+int
+mpi_level(void)
+{
+	return MPI_THREAD_SINGLE;
+}
+
 const char *
 summary(void)
 {

src/heat/mpi/solver_mpi_nbuffer.c

@@ -8,6 +8,12 @@ typedef struct {
 	MPI_Request recv;
 } HaloRequests;
 
+int
+mpi_level(void)
+{
+	return MPI_THREAD_SINGLE;
+}
+
 const char *
 summary(void)
 {

src/heat/mpi/solver_mpi_ompss2_forkjoin.c

@@ -3,6 +3,12 @@
 #include "utils.h"
 #include "common/heat.h"
 
+int
+mpi_level(void)
+{
+	return MPI_THREAD_SERIALIZED;
+}
+
 const char *
 summary(void)
 {

src/heat/mpi/solver_mpi_ompss2_tasks.c

@@ -5,6 +5,12 @@
 
 static int serial;
 
+int
+mpi_level(void)
+{
+	return MPI_THREAD_SERIALIZED;
+}
+
 const char *
 summary(void)
 {

src/heat/mpi/utils.h

@@ -17,5 +17,6 @@ typedef struct {
 void broadcastConfiguration(HeatConfiguration *configuration);
 void initializeWindows(HeatConfiguration *configuration, int64_t rows, int64_t cols, MPIRMAInfo *info);
 void finalizeWindows(MPIRMAInfo *info);
+int mpi_level(void);
 
 #endif // MPI_UTILS_H