Reorganice sources

.gitignore

@@ -1 +1,4 @@
 data
+build/
+install/
+tags

CMakeLists.txt

@@ -18,9 +18,6 @@ add_compile_options(
   -Werror
 )
 
-set(CMAKE_C_COMPILER "clang")
-set(CMAKE_CXX_COMPILER "clang++")
-
 set(CMAKE_C_VISIBILITY_PRESET hidden)
 
 set(CMAKE_C_STANDARD 11)
@@ -76,6 +73,23 @@ find_package(MPI)
 find_package(Nanos6)
 find_package(Nodes)
 
+set_property(GLOBAL PROPERTY bench6_list "")
+
+macro(mk_bench NAME)
+  if(NOT "${NAME}" MATCHES "b6_.*")
+    message(FATAL_ERROR "benchmark name must begin with b6_: ${NAME}")
+  endif()
+  add_executable(${NAME})
+
+  get_property(BENCH6_LIST GLOBAL PROPERTY bench6_list)
+  message(STATUS "Before BENCH6_LIST=${BENCH6_LIST}")
+  list(APPEND BENCH6_LIST ${NAME})
+  message(STATUS "After BENCH6_LIST=${BENCH6_LIST}")
+  set_property(GLOBAL PROPERTY bench6_list "${BENCH6_LIST}")
+
+  install(TARGETS ${NAME} RUNTIME DESTINATION bin)
+endmacro()
+
 add_subdirectory(src)
 
 include(FeatureSummary)

src/CMakeLists.txt

@@ -1,2 +1,4 @@
 add_subdirectory(bench6)
+add_subdirectory(ompss2)
 add_subdirectory(heat)
+add_subdirectory(tools)

src/bench6/CMakeLists.txt

@@ -1,13 +1,2 @@
-add_library(bench6_common STATIC common.c)
+add_library(bench6_lib STATIC bench6.c common.c)
-
+target_include_directories(bench6_lib PUBLIC ${CMAKE_CURRENT_SOURCE_DIR})
-macro(mk_bench6 NAME SOURCE)
-  add_executable(${NAME} ${SOURCE})
-  target_link_libraries(${NAME} PRIVATE bench6_common Nanos6::wrapper)
-  install(TARGETS ${NAME} RUNTIME DESTINATION bin)
-endmacro()
-
-# FIXME: broken in last clang
-#mk_bench6(bench6_register_deps register_deps.c)
-mk_bench6(bench6_sched_add sched_add.c)
-mk_bench6(bench6_sched_get sched_get.c)
-mk_bench6(bench6_readywave readywave.c)

src/bench6/bench6.c

@@ -0,0 +1,35 @@
+/* Copyright (c) 2022 Barcelona Supercomputing Center (BSC)
+ * SPDX-License-Identifier: GPL-3.0-or-later */
+
+#include "bench6.h"
+
+//#include <nanos6/debug.h>
+#include <time.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+/* Returns the current time in seconds since some point in the past */
+double bench6_time(void)
+{
+	struct timespec tv;
+	if(clock_gettime(CLOCK_MONOTONIC, &tv) != 0)
+	{
+		perror("clock_gettime failed");
+		exit(EXIT_FAILURE);
+	}
+
+	return (double)(tv.tv_sec) +
+		(double)tv.tv_nsec * 1.0e-9;
+}
+
+int get_ncpus(void)
+{
+	return -1;
+	//return (int) nanos6_get_num_cpus();
+}
+
+void
+bench6_report(double time)
+{
+	printf("time %e\n", time);
+}

src/bench6/bench6.h

@@ -4,14 +4,9 @@
 #ifndef BENCH6_H
 #define BENCH6_H
 
-#define UNUSED(x) (void)(x)
+double bench6_time(void);
+void bench6_report(double time);
 
-double get_time(void);
 int get_ncpus(void);
 
-int bench6_creator(int argc, char *argv[]);
-int bench6_sched_get(int argc, char *argv[]);
-int bench6_sched_add(int argc, char *argv[]);
-int bench6_register_deps(int argc, char *argv[]);
-
 #endif /* BENCH6_H */

src/bench6/common.c

@@ -1,28 +1,69 @@
-/* Copyright (c) 2022 Barcelona Supercomputing Center (BSC)
+/* Copyright (c) 2021-2023 Barcelona Supercomputing Center (BSC)
  * SPDX-License-Identifier: GPL-3.0-or-later */
 
-#include "bench6.h"
+#include "common.h"
 
-#include <nanos6/debug.h>
-#include <time.h>
 #include <stdio.h>
+#include <stdarg.h>
+#include <string.h>
+#include <errno.h>
 #include <stdlib.h>
 
-/* Returns the current time in seconds since some point in the past */
+char *progname = NULL;
-double get_time(void)
+int is_debug_enabled = 0;
+
+void
+progname_set(char *name)
 {
-	struct timespec tv;
+	progname = name;
-	if(clock_gettime(CLOCK_MONOTONIC, &tv) != 0)
-	{
-		perror("clock_gettime failed");
-		exit(EXIT_FAILURE);
 }
 
-	return (double)(tv.tv_sec) +
+void
-		(double)tv.tv_nsec * 1.0e-9;
+enable_debug(void)
+{
+	is_debug_enabled = 1;
 }
 
-int get_ncpus(void)
+static void
+vaerr(const char *prefix, const char *func, const char *errstr, va_list ap)
 {
-	return (int) nanos6_get_num_cpus();
+	if (progname != NULL)
+		fprintf(stderr, "%s: ", progname);
+
+	if (prefix != NULL)
+		fprintf(stderr, "%s: ", prefix);
+
+	if (func != NULL)
+		fprintf(stderr, "%s: ", func);
+
+	vfprintf(stderr, errstr, ap);
+
+	int len = strlen(errstr);
+
+	if (len > 0) {
+		char last = errstr[len - 1];
+		if (last == ':')
+			fprintf(stderr, " %s\n", strerror(errno));
+		else if (last != '\n' && last != '\r')
+			fprintf(stderr, "\n");
+	}
+}
+
+void
+verr(const char *prefix, const char *func, const char *errstr, ...)
+{
+	va_list ap;
+	va_start(ap, errstr);
+	vaerr(prefix, func, errstr, ap);
+	va_end(ap);
+}
+
+void
+vdie(const char *prefix, const char *func, const char *errstr, ...)
+{
+	va_list ap;
+	va_start(ap, errstr);
+	vaerr(prefix, func, errstr, ap);
+	va_end(ap);
+	abort();
 }

src/bench6/common.h

@@ -0,0 +1,46 @@
+/* Copyright (c) 2021-2023 Barcelona Supercomputing Center (BSC)
+ * SPDX-License-Identifier: GPL-3.0-or-later */
+
+#ifndef COMMON_H
+#define COMMON_H
+
+#include <stdio.h>
+
+extern int is_debug_enabled;
+
+/* Debug macros */
+
+void progname_set(char *name);
+void enable_debug(void);
+void verr(const char *prefix, const char *func, const char *errstr, ...);
+void vdie(const char *prefix, const char *func, const char *errstr, ...);
+
+/* clang-format off */
+
+#define rerr(...) fprintf(stderr, __VA_ARGS__)
+#define err(...)  verr("ERROR", __func__, __VA_ARGS__)
+#define die(...)  vdie("FATAL", __func__, __VA_ARGS__)
+#define info(...) verr("INFO", NULL, __VA_ARGS__)
+#define finfo(...) verr("INFO", __func__, __VA_ARGS__)
+#define warn(...) verr("WARN", NULL, __VA_ARGS__)
+
+#define dbg(...) do { \
+	if (unlikely(is_debug_enabled)) verr("DEBUG", __func__, __VA_ARGS__); \
+} while (0);
+
+#define likely(x) __builtin_expect(!!(x), 1)
+#define unlikely(x) __builtin_expect(!!(x), 0)
+#define UNUSED(x) (void)(x)
+
+/* Poison assert */
+#pragma GCC poison assert
+
+#define USE_RET __attribute__((warn_unused_result))
+
+#define ARRAYLEN(x) (sizeof(x)/sizeof((x)[0]))
+
+/* clang-format on */
+
+
+
+#endif /* COMMON_H */

src/heat/CMakeLists.txt

@@ -1,5 +1,6 @@
 add_library(heat_common STATIC common/misc.c common/kernel.c)
 target_include_directories(heat_common PUBLIC ${CMAKE_CURRENT_SOURCE_DIR})
+target_link_libraries(heat_common PUBLIC m)
 
 add_subdirectory(smp)
 add_subdirectory(mpi)

src/heat/common/heat.h

@@ -3,11 +3,11 @@
 
 #include <stdbool.h>
 #include <stdint.h>
+#include <limits.h>
 
 #define IGNORE_RESIDUAL ((double) -1.0)
 #define DEFAULT_DELTA ((double) 0.00005)
 #define DEFAULT_BS 1024
-#define MAX_STRING_SIZE 100
 
 #define ROUND(a, b) ((((a) + (b) - 1) / (b)) * (b))
 
@@ -30,8 +30,8 @@ typedef struct {
 	double *matrix;
 	int numHeatSources;
 	HeatSource *heatSources;
-	char confFileName[MAX_STRING_SIZE];
+	char confFileName[PATH_MAX];
-	char imageFileName[MAX_STRING_SIZE];
+	char imageFileName[PATH_MAX];
 	bool generateImage;
 	bool warmup;
 	bool verbose;
@@ -47,6 +47,7 @@ void printConfiguration(const HeatConfiguration *conf);
 void initializeMatrix(const HeatConfiguration *conf, double *matrix, int64_t rows, int64_t cols, int64_t rowOffset);
 double getTime(void);
 
+const char *summary(void);
 double solve(HeatConfiguration *conf, int64_t rows, int64_t cols, int timesteps, void *extraData);
 void computeBlock(const int64_t rows, const int64_t cols, const int rstart, const int rend, const int cstart, const int cend, double M[rows][cols]);
 double computeBlockResidual(const int64_t rows, const int64_t cols, const int rstart, const int rend, const int cstart, const int cend, double M[rows][cols]);

src/heat/common/main.c

@@ -0,0 +1,52 @@
+#include <assert.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#include "common/heat.h"
+
+
+int main(int argc, char **argv)
+{
+	HeatConfiguration conf;
+	readConfiguration(argc, argv, &conf);
+	refineConfiguration(&conf, conf.rbs, conf.cbs);
+	if (conf.verbose)
+		printConfiguration(&conf);
+
+	int64_t rows = conf.rows+2;
+	int64_t cols = conf.cols+2;
+
+	initialize(&conf, rows, cols, 0);
+
+	if (conf.warmup)
+		solve(&conf, rows, cols, 1, NULL);
+
+	// Solve the problem
+	double start = getTime();
+	double residual = solve(&conf, rows, cols, conf.timesteps, NULL);
+	double end = getTime();
+
+	int64_t totalElements = conf.rows*conf.cols;
+	double throughput = (totalElements*conf.timesteps)/(end-start);
+
+#ifdef _OMPSS_2
+	int threads = sysconf(_SC_NPROCESSORS_ONLN);
+#else
+	int threads = 1;
+#endif
+
+	fprintf(stderr,"%8s %8s %8s %8s %8s %8s %14s %14s %14s",
+			"rows", "cols", "rbs", "cbs", "threads",
+			"steps", "error", "time", "updates/s\n");
+	fprintf(stdout, "%8ld %8ld %8d %8d %8d %8d %14e %14e %14e\n", 
+			conf.rows, conf.cols, 
+			conf.rbs, conf.cbs, threads,
+			conf.convergenceTimesteps, residual, end-start, throughput);
+
+	if (conf.generateImage)
+		writeImage(conf.imageFileName, conf.matrix, rows, cols);
+
+	finalize(&conf);
+
+	return 0;
+}

src/heat/common/misc.c

@@ -1,11 +1,11 @@
-#define _POSIX_C_SOURCE 200809L
-
 #include <assert.h>
 #include <float.h>
 #include <getopt.h>
+#include <libgen.h>
+#include <limits.h>
 #include <math.h>
-#include <stdlib.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
 #include <time.h>
 #include <unistd.h>
@@ -105,7 +105,11 @@ static void printUsage(int argc, char **argv)
 {
 	(void) argc;
 
-	fprintf(stdout, "Usage: %s [OPTION]...\n", argv[0]);
+	const char *prog = basename(argv[0]);
+	fprintf(stdout, "%s - %s\n", prog, summary());
+	fprintf(stdout, "\n");
+	fprintf(stdout, "Usage: %s [OPTION]...\n", prog);
+	fprintf(stdout, "\n");
 	fprintf(stdout, "Parameters:\n");
 	fprintf(stdout, "  -s, --size=SIZE          use SIZExSIZE matrix as the surface\n");
 	fprintf(stdout, "  -r, --rows=ROWS          use ROWS as the number of rows of the surface\n");
@@ -176,7 +180,7 @@ static void readParameters(int argc, char **argv, HeatConfiguration *conf)
 				conf->verbose = true;
 				break;
 			case 'f':
-				if (strlen(optarg) >= MAX_STRING_SIZE) {
+				if (strlen(optarg) >= PATH_MAX) {
 					fprintf(stderr, "Error: Configuration name is too long!\n");
 					exit(1);
 				}
@@ -185,7 +189,7 @@ static void readParameters(int argc, char **argv, HeatConfiguration *conf)
 			case 'o':
 				conf->generateImage = true;
 				conf->warmup = false;
-				if (strlen(optarg) >= MAX_STRING_SIZE) {
+				if (strlen(optarg) >= PATH_MAX) {
 					fprintf(stderr, "Error: Image name is too long!\n");
 					exit(1);
 				}
@@ -263,8 +267,8 @@ static void readParameters(int argc, char **argv, HeatConfiguration *conf)
 
 static void readSourcesFile(HeatConfiguration *conf, FILE *file)
 {
-	char line[MAX_STRING_SIZE];
+	char line[4096];
-	if (!fgets(line, MAX_STRING_SIZE, file)) {
+	if (!fgets(line, 4096, file)) {
 		fprintf(stderr, "Error: Configuration file is not correct!\n");
 		exit(1);
 	}
@@ -283,7 +287,7 @@ static void readSourcesFile(HeatConfiguration *conf, FILE *file)
 	assert(conf->heatSources != NULL);
 
 	for (int i = 0; i < conf->numHeatSources; i++) {
-		if (!fgets(line, MAX_STRING_SIZE, file)) {
+		if (!fgets(line, 4096, file)) {
 			fprintf(stderr, "Error: Configuration file is not correct!\n");
 			exit(1);
 		}

src/heat/mpi/CMakeLists.txt

@@ -2,34 +2,36 @@ if(NOT MPI_FOUND)
   return()
 endif()
 
-add_library(heat_mpi_common STATIC main.c utils.c)
-target_link_libraries(heat_mpi_common PUBLIC heat_common MPI::MPI_C)
-
 macro(mk_heat_mpi NAME SOURCE)
-  add_executable(${NAME} ${SOURCE})
+  mk_bench(${NAME})
+  target_sources(${NAME} PRIVATE ${SOURCE})
   target_link_libraries(${NAME} PRIVATE heat_mpi_common)
-  install(TARGETS ${NAME} RUNTIME DESTINATION bin)
 endmacro()
 
-mk_heat_mpi(heat_mpi solver_mpi.c)
-mk_heat_mpi(heat_mpi_nbuffer solver_mpi_nbuffer.c)
-
-if(NANOS6_FOUND)
 macro(mk_heat_mpi_nanos6 NAME SOURCE)
   mk_heat_mpi(${NAME} ${SOURCE})
   target_link_libraries(${NAME} PRIVATE Nanos6::wrapper)
 endmacro()
 
-  mk_heat_mpi_nanos6(heat_mpi_nanos6_forkjoin solver_mpi_ompss2_forkjoin.c)
-  mk_heat_mpi_nanos6(heat_mpi_nanos6_tasks solver_mpi_ompss2_tasks.c)
-endif()
-
-if(NODES_FOUND)
 macro(mk_heat_mpi_nodes NAME SOURCE)
   mk_heat_mpi(${NAME} ${SOURCE})
   target_link_libraries(${NAME} PRIVATE Nodes::wrapper)
 endmacro()
 
-  mk_heat_mpi_nodes(heat_mpi_nodes_forkjoin solver_mpi_ompss2_forkjoin.c)
+# -------------------------------------------------------------------
-  mk_heat_mpi_nodes(heat_mpi_nodes_tasks solver_mpi_ompss2_tasks.c)
+
+add_library(heat_mpi_common STATIC main.c utils.c)
+target_link_libraries(heat_mpi_common PUBLIC heat_common MPI::MPI_C)
+
+mk_heat_mpi(b6_heat_mpi solver_mpi.c)
+mk_heat_mpi(b6_heat_mpi_nbuffer solver_mpi_nbuffer.c)
+
+if(NANOS6_FOUND)
+  mk_heat_mpi_nanos6(b6_heat_mpi_nanos6_forkjoin solver_mpi_ompss2_forkjoin.c)
+  mk_heat_mpi_nanos6(b6_heat_mpi_nanos6_tasks solver_mpi_ompss2_tasks.c)
+endif()
+
+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()

src/heat/mpi/solver_mpi.c

@@ -3,6 +3,11 @@
 #include "utils.h"
 #include "common/heat.h"
 
+const char *
+summary(void)
+{
+	return "Parallel version using MPI and blocking primitives";
+}
 
 static inline void send(const double *data, int nelems, int dst, int tag)
 {

src/heat/mpi/solver_mpi_nbuffer.c

@@ -8,6 +8,13 @@ typedef struct {
 	MPI_Request recv;
 } HaloRequests;
 
+const char *
+summary(void)
+{
+	return "Parallel version using MPI and non-blocking primitives with\n"
+		"overlap of computation and communication phases";
+}
+
 static inline void isend(const double *data, int nelems, int dst, int tag, HaloRequests *reqs)
 {
 	MPI_Isend(data, nelems, MPI_DOUBLE, dst, tag, MPI_COMM_WORLD, &reqs->send);

src/heat/mpi/solver_mpi_ompss2_forkjoin.c

@@ -3,6 +3,12 @@
 #include "utils.h"
 #include "common/heat.h"
 
+const char *
+summary(void)
+{
+	return "Parallel version using MPI + OmpSs-2 following a fork-join\n"
+		"parallelization.";
+}
 
 static inline void send(const double *data, int nelems, int dst, int tag)
 {

src/heat/mpi/solver_mpi_ompss2_tasks.c

@@ -3,9 +3,15 @@
 #include "utils.h"
 #include "common/heat.h"
 
-
 static int serial;
 
+const char *
+summary(void)
+{
+	return "Parallel version using MPI + OmpSs-2 tasks where communication\n"
+		"tasks are serialized";
+}
+
 static inline void send(const double *data, int nelems, int dst, int tag)
 {
 	MPI_Send(data, nelems, MPI_DOUBLE, dst, tag, MPI_COMM_WORLD);

src/heat/smp/CMakeLists.txt

@@ -1,25 +1,32 @@
-add_library(heat_seq_common STATIC main.c)
+macro(mk_heat_smp NAME SOURCE)
-target_link_libraries(heat_seq_common PUBLIC heat_common)
+  mk_bench(${NAME})
+  target_sources(${NAME} PRIVATE ${SOURCE})
+  target_link_libraries(${NAME} PRIVATE heat_smp_common)
+endmacro()
 
-add_executable(heat_seq solver_seq.c)
+macro(mk_heat_nanos6 NAME SOURCE)
-target_link_libraries(heat_seq PUBLIC heat_seq_common)
+  mk_heat_smp(${NAME} ${SOURCE})
+  target_link_libraries(${NAME} PRIVATE Nanos6::wrapper)
+endmacro()
+
+macro(mk_heat_nodes NAME SOURCE)
+  mk_heat_smp(${NAME} ${SOURCE})
+  target_link_libraries(${NAME} PRIVATE Nodes::wrapper)
+endmacro()
+
+# -------------------------------------------------------------------
+
+add_library(heat_smp_common STATIC main.c)
+target_link_libraries(heat_smp_common PUBLIC heat_common)
+
+mk_heat_smp(b6_heat_seq solver_seq.c)
 
 if(NANOS6_FOUND)
-  macro(mk_heat_nanos6 NAME SOURCE)
+  mk_heat_nanos6(b6_heat_nanos6 solver_ompss2.c)
-    add_executable(${NAME} ${SOURCE})
+  mk_heat_nanos6(b6_heat_nanos6_residual solver_ompss2_residual.c)
-    target_link_libraries(${NAME} PRIVATE heat_seq_common Nanos6::wrapper)
-    install(TARGETS ${NAME} RUNTIME DESTINATION bin)
-  endmacro()
-  mk_heat_nanos6(heat_nanos6 solver_ompss2.c)
-  mk_heat_nanos6(heat_nanos6_residual solver_ompss2_residual.c)
 endif()
 
 if(NODES_FOUND)
-  macro(mk_heat_nodes NAME SOURCE)
+  mk_heat_nodes(b6_heat_nodes solver_ompss2.c)
-    add_executable(${NAME} ${SOURCE})
+  mk_heat_nodes(b6_heat_nodes_residual solver_ompss2_residual.c)
-    target_link_libraries(${NAME} PRIVATE heat_seq_common Nodes::wrapper)
-    install(TARGETS ${NAME} RUNTIME DESTINATION bin)
-  endmacro()
-  mk_heat_nodes(heat_nodes solver_ompss2.c)
-  mk_heat_nodes(heat_nodes_residual solver_ompss2_residual.c)
 endif()

src/heat/smp/main.c

@@ -35,13 +35,16 @@ int main(int argc, char **argv)
 	int threads = 1;
 #endif
 
-	fprintf(stderr,"%8s %8s %8s %8s %8s %8s %14s %14s %14s",
+	fprintf(stderr, "%14s %14s %14s %8s %8s %8s %8s %8s %8s\n",
-			"rows", "cols", "rbs", "cbs", "threads",
+			"time", "updates/s", "error", 
-			"steps", "error", "time", "updates/s\n");
+			"rows", "cols",
-	fprintf(stdout, "%8ld %8ld %8d %8d %8d %8d %14e %14e %14e\n", 
+			"rbs", "cbs", "threads",
+			"steps");
+	fprintf(stdout, "%14e %14e %14e %8ld %8ld %8d %8d %8d %8d\n", 
+			end-start, throughput, residual,
 			conf.rows, conf.cols, 
 			conf.rbs, conf.cbs, threads,
-			conf.convergenceTimesteps, residual, end-start, throughput);
+			conf.convergenceTimesteps);
 
 	if (conf.generateImage)
 		writeImage(conf.imageFileName, conf.matrix, rows, cols);

src/heat/smp/solver_ompss2.c

@@ -1,5 +1,10 @@
 #include "common/heat.h"
 
+const char *
+summary(void)
+{
+	return "Parallel version using OmpSs-2 tasks";
+}
 
 static inline void gaussSeidelSolver(int64_t rows, int64_t cols, int rbs, int cbs, int nrb, int ncb, double M[rows][cols], char reps[nrb][ncb])
 {

src/heat/smp/solver_ompss2_residual.c

@@ -2,6 +2,12 @@
 
 #include "common/heat.h"
 
+const char *
+summary(void)
+{
+	return "Parallel version using OmpSs-2 tasks and taking into account the\n"
+		"residual";
+}
 
 static inline void gaussSeidelSolver(int64_t rows, int64_t cols, int rbs, int cbs, int nrb, int ncb, double M[rows][cols], char reps[nrb][ncb], double *residual)
 {

src/heat/smp/solver_ompss2_taskloop.c

@@ -1,5 +1,10 @@
 #include "common/heat.h"
 
+const char *
+summary(void)
+{
+	return "Parallel version using OmpSs-2 tasklook";
+}
 
 static inline void gaussSeidelSolver(int64_t rows, int64_t cols, int rbs, int cbs, int nrb, int ncb, double M[rows][cols], char reps[nrb][ncb])
 {

src/heat/smp/solver_seq.c

@@ -1,5 +1,10 @@
 #include "common/heat.h"
 
+const char *
+summary(void)
+{
+	return "Sequential solver with one CPU";
+}
 
 static inline void gaussSeidelSolver(int64_t rows, int64_t cols, int rbs, int cbs, double M[rows][cols])
 {

src/ompss2/CMakeLists.txt

@@ -0,0 +1,12 @@
+macro(mk_nanos6 NAME SOURCE)
+  mk_bench(${NAME})
+  target_sources(${NAME} PRIVATE ${SOURCE})
+  target_link_libraries(${NAME} PRIVATE Nanos6::wrapper)
+endmacro()
+
+if(NANOS6_FOUND)
+  mk_bench6(b6_nanos6_register_deps register_deps.c)
+  mk_bench6(b6_nanos6_sched_add sched_add.c)
+  mk_bench6(b6_nanos6_sched_get sched_get.c)
+  mk_bench6(b6_nanos6_readywave readywave.c)
+endif()

src/ompss2/README.md

@@ -1,6 +1,4 @@
-## Bench6: A set of micro-benchmarks for Nanos6
+This directory contains a set of microbenchmarks for Nanos6,
-
-This repository contains a set of microbenchmarks for Nanos6,
 specifically target to expose the limitations of the runtime following
 the breakdown analysis.
 

src/tools/CMakeLists.txt

@@ -0,0 +1,7 @@
+get_property(BENCH6_LIST GLOBAL PROPERTY bench6_list)
+configure_file(config.in.h config.h)
+include_directories(${CMAKE_CURRENT_BINARY_DIR})
+
+add_executable(bench6_runner runner.c)
+target_link_libraries(bench6_runner PRIVATE m bench6_lib)
+install(TARGETS bench6_runner RUNTIME DESTINATION bin)

src/tools/config.in.h

@@ -0,0 +1,11 @@
+/* Copyright (c) 2023 Barcelona Supercomputing Center (BSC)
+ * SPDX-License-Identifier: GPL-3.0-or-later */
+
+#ifndef CONFIG_H
+#define CONFIG_H
+
+#define BENCH6_PREFIX "@CMAKE_INSTALL_PREFIX@"
+#define BENCH6_BIN (BENCH6_PREFIX "/bin")
+#define BENCH6_LIST "@BENCH6_LIST@"
+
+#endif /* CONFIG_H */

src/tools/runner.c

@@ -0,0 +1,199 @@
+#include "common.h"
+#include "config.h"
+#include <limits.h>
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+
+//static void
+//usage(void)
+//{
+//	exit(1);
+//}
+
+struct sampling {
+	int nmax;
+	int nmin;
+	int n;
+	double *samples;
+	double rse;
+	double last;
+};
+
+static int
+do_run(char *argv[], double *ptime)
+{
+	/* Gather binary path */
+	char path[PATH_MAX];
+	sprintf(path, "%s/%s", BENCH6_BIN, argv[0]);
+
+	if (access(path, R_OK | X_OK) != 0) {
+		err("cannot find benchmark %s:", path);
+		return -1;
+	}
+
+	int pipefd[2];
+	if (pipe(pipefd) != 0) {
+		err("pipe failed:");
+		return -1;
+	}
+
+	/* Fork */
+	pid_t p = fork();
+
+	if (p < 0) {
+		err("fork failed:");
+		return -1;
+	}
+
+	/* In children execute benchmark */
+	if (p == 0) {
+		close(pipefd[0]);
+		dup2(pipefd[1], 1);
+		close(2);
+		if (execve(path, argv, NULL) != 0) {
+			err("execve failed:");
+			return -1;
+		}
+		/* Not reached */
+	} else {
+		close(pipefd[1]);
+		char line[4096];
+		FILE *f = fdopen(pipefd[0], "r");
+		if (f == NULL) {
+			err("fdopen failed:");
+			return -1;
+		}
+
+		if (fgets(line, 4096, f) == NULL) {
+			err("missing stdout line");
+			return -1;
+		}
+
+		char *nl = strchr(line, '\n');
+		if (nl != NULL)
+			*nl = '\0';
+
+		double time;
+		sscanf(line, "%le", &time);
+		//printf("got %e\n", time);
+		*ptime = time;
+
+		/* Drain the rest of the stdout */
+		while (fgets(line, 4096, f) != NULL) { }
+		fclose(f);
+		close(pipefd[0]);
+	}
+
+	return 0;
+}
+
+static void
+stats(struct sampling *s)
+{
+	if (s->n < 2)
+		return;
+
+	double n = s->n;
+	double sum = 0.0;
+	for (int i = 0; i < s->n; i++)
+		sum += s->samples[i];
+
+	double mean = sum / n;
+	double sumsqr = 0.0;
+	for (int i = 0; i < s->n; i++) {
+		double dev = s->samples[i] - mean;
+		sumsqr += dev * dev;
+	}
+
+	double var = sumsqr / n;
+	double stdev = sqrt(var);
+	double se = stdev / sqrt(n);
+	double rse = se * 1.96 / mean;
+
+	fprintf(stderr, "\rn=%d last=%e mean=%e stdev=%e se=%e rse=%e",
+			s->n, s->last, mean, stdev, se, rse);
+
+	s->rse = rse;
+}
+
+static int
+should_continue(struct sampling *s)
+{
+	stats(s);
+
+	if (s->n < s->nmin)
+		return 1;
+
+	if (s->rse * 100.0 > 1.0 /* % */)
+		return 1;
+
+	return 0;
+}
+
+static void
+add_sample(struct sampling *s, double time)
+{
+	if (s->n >= s->nmax) {
+		die("overflowing samples");
+	} else {
+		s->samples[s->n] = time;
+		s->n++;
+		s->last = time;
+	}
+}
+
+//static int
+//compare_double(const void *a, const void *b)
+//{
+//	double aa = *(const double *) a;
+//	double bb = *(const double *) b;
+//
+//	if (aa < bb)
+//		return -1;
+//	else if (aa > bb)
+//		return +1;
+//	else
+//		return 0;
+//}
+
+static int
+sample(char *argv[])
+{
+	struct sampling s = { 0 };
+	s.nmax = 4000;
+	s.nmin = 30;
+	s.samples = calloc(s.nmax, sizeof(double));
+	s.n = 0;
+
+	while (should_continue(&s)) {
+		double time;
+		if (do_run(argv, &time) != 0) {
+			err("failed to run benchmark");
+			return 1;
+		}
+
+		add_sample(&s, time);
+	}
+
+	free(s.samples);
+
+	return 0;
+}
+
+int
+main(int argc, char *argv[])
+{
+	(void) argc;
+
+	if (sample(argv+1) != 0) {
+		err("failed to sample the benchmark");
+		return 1;
+	}
+
+	return 0;
+}