Revert "bench and commit ping pong"

This reverts commit f5238a23

mpi/examples/benchs/abstract_sendrecv_bench.c

@@ -22,7 +22,7 @@
 void sendrecv_bench(int mpi_rank, starpu_pthread_barrier_t* thread_barrier)
 {
 	uint64_t iterations = LOOPS_DEFAULT;
-	uint64_t s, j;
+
 	if (mpi_rank >= 2)
 	{
 		starpu_pause();
@@ -31,13 +31,13 @@ void sendrecv_bench(int mpi_rank, starpu_pthread_barrier_t* thread_barrier)
 			STARPU_PTHREAD_BARRIER_WAIT(thread_barrier);
 		}
 
-		for (s = NX_MIN; s <= NX_MAX; s = bench_next_size(s))
+		for (uint64_t s = NX_MIN; s <= NX_MAX; s = bench_next_size(s))
 		{
 			iterations = bench_nb_iterations(iterations, s);
 
 			starpu_mpi_barrier(MPI_COMM_WORLD);
 
-			for (j = 0; j < iterations; j++)
+			for (uint64_t j = 0; j < iterations; j++)
 			{
 				starpu_mpi_barrier(MPI_COMM_WORLD);
 			}
@@ -66,7 +66,7 @@ void sendrecv_bench(int mpi_rank, starpu_pthread_barrier_t* thread_barrier)
 	}
 
 	global_tstart = starpu_timing_now();
-	for (s = NX_MIN; s <= NX_MAX; s = bench_next_size(s))
+	for (uint64_t s = NX_MIN; s <= NX_MAX; s = bench_next_size(s))
 	{
 		vector_send = malloc(s);
 		vector_recv = malloc(s);
@@ -80,7 +80,7 @@ void sendrecv_bench(int mpi_rank, starpu_pthread_barrier_t* thread_barrier)
 
 		starpu_mpi_barrier(MPI_COMM_WORLD);
 
-		for (j = 0; j < iterations; j++)
+		for (uint64_t j = 0; j < iterations; j++)
 		{
 			if (mpi_rank == 0)
 			{
@@ -112,8 +112,8 @@ void sendrecv_bench(int mpi_rank, starpu_pthread_barrier_t* thread_barrier)
 			const double d1_lat = lats[(iterations - 1) / 10];
 			const double d9_lat = lats[9 * (iterations - 1) / 10];
 			double avg_lat = 0.0;
-			uint64_t k;
-			for(k = 0; k < iterations; k++)
+
+			for(uint64_t k = 0; k < iterations; k++)
 			{
 				avg_lat += lats[k];
 			}

mpi/examples/benchs/burst_helper.c

@@ -36,7 +36,6 @@ int burst_nb_requests = NB_REQUESTS;
 
 void burst_init_data(int rank)
 {
-	int i;
 	if (rank == 0 || rank == 1)
 	{
 		recv_handles = malloc(burst_nb_requests * sizeof(starpu_data_handle_t));
@@ -46,7 +45,7 @@ void burst_init_data(int rank)
 		recv_reqs = malloc(burst_nb_requests * sizeof(starpu_mpi_req));
 		send_reqs = malloc(burst_nb_requests * sizeof(starpu_mpi_req));
 
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			send_buffers[i] = malloc(NX_ARRAY * sizeof(float));
 			memset(send_buffers[i], 0, NX_ARRAY * sizeof(float));
@@ -61,10 +60,9 @@ void burst_init_data(int rank)
 
 void burst_free_data(int rank)
 {
-	int i;
 	if (rank == 0 || rank == 1)
 	{
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			starpu_data_unregister(send_handles[i]);
 			free(send_buffers[i]);
@@ -86,12 +84,12 @@ void burst_free_data(int rank)
 void burst_bidir(int rank)
 {
 	int other_rank = (rank == 0) ? 1 : 0;
-	int i;
+
 	FPRINTF(stderr, "Simultaneous....start (rank %d)\n", rank);
 
 	if (rank == 0 || rank == 1)
 	{
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			recv_reqs[i] = NULL;
 			starpu_mpi_irecv(recv_handles[i], &recv_reqs[i], other_rank, i, MPI_COMM_WORLD);
@@ -102,13 +100,13 @@ void burst_bidir(int rank)
 
 	if (rank == 0 || rank == 1)
 	{
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			send_reqs[i] = NULL;
 			starpu_mpi_isend_prio(send_handles[i], &send_reqs[i], other_rank, i, i, MPI_COMM_WORLD);
 		}
 
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			if (recv_reqs[i]) starpu_mpi_wait(&recv_reqs[i], MPI_STATUS_IGNORE);
 			if (send_reqs[i]) starpu_mpi_wait(&send_reqs[i], MPI_STATUS_IGNORE);
@@ -122,10 +120,10 @@ void burst_bidir(int rank)
 void burst_unidir(int sender, int receiver, int rank)
 {
 	FPRINTF(stderr, "%d -> %d... start (rank %d)\n", sender, receiver, rank);
-	int i;
+
 	if (rank == receiver)
 	{
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			recv_reqs[i] = NULL;
 			starpu_mpi_irecv(recv_handles[i], &recv_reqs[i], sender, i, MPI_COMM_WORLD);
@@ -136,7 +134,7 @@ void burst_unidir(int sender, int receiver, int rank)
 
 	if (rank == sender)
 	{
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			send_reqs[i] = NULL;
 			starpu_mpi_isend_prio(send_handles[i], &send_reqs[i], receiver, i, i, MPI_COMM_WORLD);
@@ -145,7 +143,7 @@ void burst_unidir(int sender, int receiver, int rank)
 
 	if (rank == sender || rank == receiver)
 	{
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			if (rank != sender && recv_reqs[i]) starpu_mpi_wait(&recv_reqs[i], MPI_STATUS_IGNORE);
 			if (rank == sender && send_reqs[i]) starpu_mpi_wait(&send_reqs[i], MPI_STATUS_IGNORE);
@@ -162,13 +160,12 @@ void burst_bidir_half_postponed(int rank)
 {
 	int other_rank = (rank == 0) ? 1 : 0;
 	int received = 0;
-	int i;
 
 	FPRINTF(stderr, "Half/half burst...start (rank %d)\n", rank);
 
 	if (rank == 0 || rank == 1)
 	{
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			recv_reqs[i] = NULL;
 			starpu_mpi_irecv(recv_handles[i], &recv_reqs[i], other_rank, i, MPI_COMM_WORLD);
@@ -179,7 +176,7 @@ void burst_bidir_half_postponed(int rank)
 
 	if (rank == 0 || rank == 1)
 	{
-		for (i = 0; i < (burst_nb_requests / 2); i++)
+		for (int i = 0; i < (burst_nb_requests / 2); i++)
 		{
 			send_reqs[i] = NULL;
 			starpu_mpi_isend_prio(send_handles[i], &send_reqs[i], other_rank, i, i, MPI_COMM_WORLD);
@@ -187,13 +184,13 @@ void burst_bidir_half_postponed(int rank)
 
 		if (recv_reqs[burst_nb_requests / 4]) starpu_mpi_wait(&recv_reqs[burst_nb_requests / 4], MPI_STATUS_IGNORE);
 
-		for (i = (burst_nb_requests / 2); i < burst_nb_requests; i++)
+		for (int i = (burst_nb_requests / 2); i < burst_nb_requests; i++)
 		{
 			send_reqs[i] = NULL;
 			starpu_mpi_isend_prio(send_handles[i], &send_reqs[i], other_rank, i, i, MPI_COMM_WORLD);
 		}
 
-		for (i = 0; i < burst_nb_requests; i++)
+		for (int i = 0; i < burst_nb_requests; i++)
 		{
 			if (recv_reqs[i]) starpu_mpi_wait(&recv_reqs[i], MPI_STATUS_IGNORE);
 			if (send_reqs[i]) starpu_mpi_wait(&send_reqs[i], MPI_STATUS_IGNORE);

mpi/examples/benchs/gemm_helper.c

@@ -98,9 +98,8 @@ static void cpu_init_matrix_random(void *descr[], void *arg)
 	TYPE *subB = (TYPE *)STARPU_MATRIX_GET_PTR(descr[1]);
 	unsigned nx = STARPU_MATRIX_GET_NX(descr[0]);
 	unsigned ny = STARPU_MATRIX_GET_NY(descr[0]);
-	unsigned i;
 
-	for (i = 0; i < nx *ny; i++)
+	for (unsigned i = 0; i < nx *ny; i++)
 	{
 		subA[i] = (TYPE) (starpu_drand48());
 		subB[i] = (TYPE) (starpu_drand48());
@@ -114,9 +113,8 @@ static void cpu_init_matrix_zero(void *descr[], void *arg)
 	TYPE *subA = (TYPE *)STARPU_MATRIX_GET_PTR(descr[0]);
 	unsigned nx = STARPU_MATRIX_GET_NX(descr[0]);
 	unsigned ny = STARPU_MATRIX_GET_NY(descr[0]);
-	unsigned i;
 
-	for (i = 0; i < nx *ny; i++)
+	for (unsigned i = 0; i < nx *ny; i++)
 	{
 		subA[i] = (TYPE) (0);
 	}
@@ -292,18 +290,18 @@ void gemm_add_polling_dependencies()
 {
 	starpu_tag_t nb_tasks = (starpu_tag_t) nslices * (starpu_tag_t) nslices;
 	unsigned nb_workers = starpu_worker_get_count();
-	starpu_tag_t synchro_tag, previous_tag, next_tag;
-	for (synchro_tag = nb_workers+1; synchro_tag <= nb_tasks; synchro_tag += (nb_workers+1))
+
+	for (starpu_tag_t synchro_tag = nb_workers+1; synchro_tag <= nb_tasks; synchro_tag += (nb_workers+1))
 	{
 		// this synchro tag depends on tasks of previous column of tasks:
-		for (previous_tag = synchro_tag - nb_workers; previous_tag < synchro_tag; previous_tag++)
+		for (starpu_tag_t previous_tag = synchro_tag - nb_workers; previous_tag < synchro_tag; previous_tag++)
 		{
 			starpu_tag_declare_deps(synchro_tag, 1, previous_tag);
 		}
 
 		// tasks of the next column of tasks depend on this synchro tag:
 		// this actually allows workers to poll for new tasks, while no task is available
-		for (next_tag = synchro_tag+1; next_tag < (synchro_tag + nb_workers + 1) && next_tag <= nb_tasks; next_tag++)
+		for (starpu_tag_t next_tag = synchro_tag+1; next_tag < (synchro_tag + nb_workers + 1) && next_tag <= nb_tasks; next_tag++)
 		{
 			starpu_tag_declare_deps(next_tag, 1, synchro_tag);
 		}

mpi/examples/benchs/sendrecv_bench.c

@@ -27,9 +27,9 @@ int main(int argc, char **argv)
 {
 	int ret, rank, worldsize;
 	int pause_workers = 0;
-	int i;
 
-	for (i = 1; i < argc; i++)
+
+	for (int i = 1; i < argc; i++)
 	{
 		if (strcmp(argv[i], "-p") == 0)
 		{

mpi/examples/benchs/sendrecv_parallel_tasks_bench.c

@@ -56,7 +56,7 @@ void cpu_task(void* descr[], void* args)
 	double t1, t2;
 	int asked_worker;
 	int current_worker = starpu_worker_get_id();
-	uint64_t j, k;
+
 	starpu_codelet_unpack_args(args, &mpi_rank, &asked_worker, &s, &handle_send, &handle_recv);
 
 	STARPU_ASSERT(asked_worker == current_worker);
@@ -64,7 +64,7 @@ void cpu_task(void* descr[], void* args)
 	iterations = bench_nb_iterations(iterations, s);
 	double* lats = malloc(sizeof(double) * iterations);
 
-	for (j = 0; j < NB_WARMUP_PINGPONGS; j++)
+	for (uint64_t j = 0; j < NB_WARMUP_PINGPONGS; j++)
 	{
 		if (mpi_rank == 0)
 		{
@@ -78,7 +78,7 @@ void cpu_task(void* descr[], void* args)
 		}
 	}
 
-	for (j = 0; j < iterations; j++)
+	for (uint64_t j = 0; j < iterations; j++)
 	{
 		if (mpi_rank == 0)
 		{
@@ -107,7 +107,7 @@ void cpu_task(void* descr[], void* args)
 		const double d9_lat = lats[9 * (iterations - 1) / 10];
 		double avg_lat = 0.0;
 
-		for(k = 0; k < iterations; k++)
+		for(uint64_t k = 0; k < iterations; k++)
 		{
 			avg_lat += lats[k];
 		}
@@ -167,19 +167,18 @@ int main(int argc, char **argv)
 	unsigned cpu_count = starpu_cpu_worker_get_count();
 	unsigned* mpi_tags = malloc(cpu_count * sizeof(unsigned));
 	unsigned tag = 0;
-	uint64_t s;
-	unsigned i;
+
 	int* workers = malloc(cpu_count * sizeof(int));
 	float** vectors_send = malloc(cpu_count * sizeof(float*));
 	float** vectors_recv = malloc(cpu_count * sizeof(float*));
 	starpu_data_handle_t* handles_send = malloc(cpu_count * sizeof(starpu_data_handle_t));
 	starpu_data_handle_t* handles_recv = malloc(cpu_count * sizeof(starpu_data_handle_t));
 
-	for ( s = NX_MIN; s <= NX_MAX; s = bench_next_size(s))
+	for (uint64_t s = NX_MIN; s <= NX_MAX; s = bench_next_size(s))
 	{
 		starpu_pause();
 
-		for (i = 0; i < cpu_count; i++)
+		for (unsigned i = 0; i < cpu_count; i++)
 		{
 			workers[i] = i;
 			vectors_send[i] = malloc(s);
@@ -202,7 +201,7 @@ int main(int argc, char **argv)
 		starpu_resume();
 		starpu_task_wait_for_all();
 
-		for (i = 0; i < cpu_count; i++)
+		for (unsigned i = 0; i < cpu_count; i++)
 		{
 			starpu_data_unregister(handles_send[i]);
 			starpu_data_unregister(handles_recv[i]);

mpi/tests/pingpong.c

@@ -43,13 +43,13 @@ int main(int argc, char **argv)
 {
 	int ret, rank, size;
 	int mpi_init;
-	int i, r;
+
 	int niter = DEFAULT_NITER;
 	int data_size = DEFAULT_DATA_SIZE;
 	int sleep_time = DEFAULT_SLEEP_TIME;
 	int method = DEFAULT_METHOD;
 
-	for (i = 1; i < argc; i++)
+	for (int i = 1; i < argc; i++)
 	{
 		if (strcmp(argv[i], "-n") == 0)
 		{
@@ -161,7 +161,7 @@ int main(int argc, char **argv)
 			sender = loop % size;
 			if (sender == rank)
 			{
-				for (r = 0; r < size; r++)
+				for (int r = 0; r < size; r++)
 				{
 					if (r != rank)
 					{
@@ -175,7 +175,7 @@ int main(int argc, char **argv)
 				MPI_Status status;
 				starpu_mpi_recv(tab_handle, sender, (rank * niter) + loop, MPI_COMM_WORLD, &status);
 
-				for (r = 0; r < (size-1); r++)
+				for (int r = 0; r < (size-1); r++)
 					starpu_sleep(sleep_time / 1000);
 			}
 		}