Merge branch 'master' of git+ssh://scm.gforge.inria.fr/gitroot/starpu/starpu into fpga

doc/doxygen/chapters/320_scheduling.doxy

@@ -185,6 +185,11 @@ already gives the good results that a precise estimation would give.
 
 \section Energy-basedScheduling Energy-based Scheduling
 
+Note: by default StarPU does not let CPU workers sleep, to let them react to
+task release as quickly as possible. For idle time to really let CPU cores save
+energy, one needs to use the \ref enable-blocking-drivers
+"--enable-blocking-drivers" configuration option.
+
 If the application can provide some energy consumption performance model (through
 the field starpu_codelet::energy_model), StarPU will
 take it into account when distributing tasks. The target function that

doc/doxygen/chapters/380_offline_performance_tools.doxy

@@ -586,19 +586,31 @@ $ starpu_paje_sort paje.trace
 \section PapiCounters PAPI counters
 
 Performance counter values could be obtained from the PAPI framework if
-<c>./configure</c> detected the libpapi. One has to set the \ref STARPU_PROFILING
-environment variable to 1 and then specify which events to record with the
-\ref STARPU_PROF_PAPI_EVENTS environment variable. For instance:
+<c>./configure</c> detected the libpapi.
+
+In Debian, packages <c>libpapi-dev</c> and <c>libpapi5.7</c> provide required
+files.  Package <c>papi-tools</c> contains a set of useful tools, for example
+<c>papi_avail</c> to see which counters are available.
+
+To be able to use Papi counters, one may need to reduce the level of the kernel
+parameter <c>kernel.perf_event_paranoid</c> to at least 2. See
+https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html for the
+security impact of this parameter.
+
+Then one has to set the \ref STARPU_PROFILING environment variable to 1 and
+specify which events to record with the \ref STARPU_PROF_PAPI_EVENTS
+environment variable. For instance:
 
 \verbatim
 export STARPU_PROFILING=1 STARPU_PROF_PAPI_EVENTS="PAPI_TOT_INS PAPI_TOT_CYC"
 \endverbatim
 
+The comma can also be used to separate events to monitor.
+
 In the current simple implementation, only CPU tasks have their events measured
-and require CPUs that support the PAPI events. All events that PAPI support are
-available from their documentation (https://icl.cs.utk.edu/projects/papi/wiki/PAPIC:Preset_Event_Definitions).
-It is important to note that not all events are available on all systems, and
-general PAPI recommendations should be followed.
+and require CPUs that support the PAPI events. It is important to note that not
+all events are available on all systems, and general PAPI recommendations
+should be followed.
 
 The counter values can be accessed using the profiling interface:
 \code{.c}

doc/doxygen/chapters/510_configure_options.doxy

@@ -1,6 +1,7 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
  * Copyright (C) 2009-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2020       Federal University of Rio Grande do Sul (UFRGS)
  *
  * StarPU is free software; you can redistribute it and/or modify
  * it under the terms of the GNU Lesser General Public License as published by

doc/doxygen/chapters/code/vector_scal_opencl.c

@@ -57,6 +57,7 @@ void scal_opencl_func(void *buffers[], void *_args)
         err = clGetKernelWorkGroupInfo (kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(local), &local, &s);
         if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
         if (local > global) local=global;
+        else global = (global + local-1) / local * local;
 
         err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, &event);
         if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);

examples/axpy/axpy_opencl.c

@@ -60,6 +60,8 @@ void axpy_opencl(void *buffers[], void *_args)
 			STARPU_OPENCL_REPORT_ERROR(err);
                 if (local > global)
 			local=global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
 		if (err != CL_SUCCESS)

examples/basic_examples/multiformat_conversion_codelets_opencl.c

@@ -74,6 +74,8 @@ void cpu_to_opencl_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

examples/basic_examples/multiformat_opencl.c

@@ -68,6 +68,8 @@ void multiformat_scal_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

examples/basic_examples/vector_scal_opencl.c

@@ -57,6 +57,7 @@ void scal_opencl_func(void *buffers[], void *_args)
                 err = clGetKernelWorkGroupInfo (kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(local), &local, &s);
                 if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
                 if (local > global) local=global;
+                else global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
 		if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);

examples/filters/custom_mf/conversion_opencl.c

@@ -76,6 +76,8 @@ void cpu_to_opencl_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(
 				queue,

examples/filters/custom_mf/custom_opencl.c

@@ -75,6 +75,8 @@ void custom_scal_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(
 				queue,

examples/interface/complex_kernels_opencl.c

@@ -64,6 +64,8 @@ void copy_complex_codelet_opencl(void *buffers[], void *_args)
 			STARPU_OPENCL_REPORT_ERROR(err);
                 if (local > global)
 			local=global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
 		if (err != CL_SUCCESS)

examples/reductions/dot_product.c

@@ -185,18 +185,12 @@ void redux_opencl_func(void *buffers[], void *args)
 
 	{
 		size_t global=1;
-		size_t local;
+                size_t local=1;
                 size_t s;
                 cl_device_id device;
 
                 starpu_opencl_get_device(devid, &device);
 
-                err = clGetKernelWorkGroupInfo (kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(local), &local, &s);
-                if (err != CL_SUCCESS)
-			STARPU_OPENCL_REPORT_ERROR(err);
-                if (local > global)
-			local=global;
-
 		err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
 		if (err != CL_SUCCESS)
 			STARPU_OPENCL_REPORT_ERROR(err);
@@ -306,18 +300,12 @@ void dot_opencl_func(void *buffers[], void *cl_arg)
 
 	{
 		size_t global=1;
-		size_t local;
+                size_t local=1;
                 size_t s;
                 cl_device_id device;
 
                 starpu_opencl_get_device(devid, &device);
 
-                err = clGetKernelWorkGroupInfo (kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(local), &local, &s);
-                if (err != CL_SUCCESS)
-			STARPU_OPENCL_REPORT_ERROR(err);
-                if (local > global)
-			local=global;
-
 		err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
 		if (err != CL_SUCCESS)
 			STARPU_OPENCL_REPORT_ERROR(err);

examples/reductions/dot_product_opencl_kernels.cl

@@ -31,6 +31,7 @@ __kernel void _dot_opencl(__global float *x,
 			  __global DOT_TYPE *dot,
 			  unsigned n)
 {
+/* FIXME: real parallel implementation */
 	unsigned i;
 	__local double tmp;
 	tmp = 0.0;

julia/examples/cholesky/cholesky_common.jl

@@ -1,3 +1,18 @@
+# StarPU --- Runtime system for heterogeneous multicore architectures.
+#
+# Copyright (C) 2020       Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+#
+# StarPU is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation; either version 2.1 of the License, or (at
+# your option) any later version.
+#
+# StarPU is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+#
+# See the GNU Lesser General Public License in COPYING.LGPL for more details.
+#
 # Standard kernels for the Cholesky factorization
 # U22 is the gemm update
 # U21 is the trsm update

julia/examples/cholesky/cholesky_native.jl

@@ -1,3 +1,18 @@
+# StarPU --- Runtime system for heterogeneous multicore architectures.
+#
+# Copyright (C) 2020       Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+#
+# StarPU is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation; either version 2.1 of the License, or (at
+# your option) any later version.
+#
+# StarPU is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+#
+# See the GNU Lesser General Public License in COPYING.LGPL for more details.
+#
 using LinearAlgebra
 
 function check(mat::Matrix{Float32})

julia/src/openblas_ldflags.jl

@@ -1,3 +1,18 @@
+# StarPU --- Runtime system for heterogeneous multicore architectures.
+#
+# Copyright (C) 2020       Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+#
+# StarPU is free software; you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation; either version 2.1 of the License, or (at
+# your option) any later version.
+#
+# StarPU is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+#
+# See the GNU Lesser General Public License in COPYING.LGPL for more details.
+#
 import LinearAlgebra.BLAS
 import Libdl
 

mpi/examples/Makefile.am

@@ -83,6 +83,10 @@ EXTRA_DIST = 				\
 	matrix_decomposition/mpi_decomposition_params.h	\
 	matrix_decomposition/mpi_decomposition_matrix.h	\
 	user_datatype/my_interface.h			\
+	benchs/abstract_sendrecv_bench.h	\
+	benchs/bench_helper.h			\
+	benchs/gemm_helper.h			\
+	benchs/burst_helper.h			\
 	helper.h
 
 examplebindir = $(libdir)/starpu/mpi
@@ -399,3 +403,68 @@ native_fortran/nf_mm_task_build.o: nf_mm_cl.mod fstarpu_mpi_mod.mod fstarpu_mod.
 native_fortran/nf_basic_ring.o: fstarpu_mpi_mod.mod fstarpu_mod.mod
 endif
 endif
+
+
+##########
+# benchs #
+##########
+
+examplebin_PROGRAMS +=		\
+	benchs/sendrecv_bench	\
+	benchs/burst
+
+if !STARPU_USE_MPI_MPI
+examplebin_PROGRAMS +=		\
+	benchs/sendrecv_parallel_tasks_bench
+endif
+
+if !STARPU_NO_BLAS_LIB
+examplebin_PROGRAMS +=		\
+	benchs/sendrecv_gemm_bench			\
+	benchs/burst_gemm
+endif
+
+if !STARPU_SIMGRID
+starpu_mpi_EXAMPLES	+=	\
+	benchs/sendrecv_bench	\
+	benchs/burst
+
+if !STARPU_USE_MPI_MPI
+starpu_mpi_EXAMPLES	+=	\
+	benchs/sendrecv_parallel_tasks_bench
+endif
+
+if !STARPU_NO_BLAS_LIB
+starpu_mpi_EXAMPLES	+=	\
+	benchs/sendrecv_gemm_bench			\
+	benchs/burst_gemm
+endif
+endif
+
+benchs_sendrecv_bench_SOURCES = benchs/sendrecv_bench.c
+benchs_sendrecv_bench_SOURCES += benchs/bench_helper.c
+benchs_sendrecv_bench_SOURCES += benchs/abstract_sendrecv_bench.c
+
+benchs_sendrecv_parallel_tasks_bench_SOURCES = benchs/sendrecv_parallel_tasks_bench.c
+benchs_sendrecv_parallel_tasks_bench_SOURCES += benchs/bench_helper.c
+benchs_sendrecv_parallel_tasks_bench_SOURCES += benchs/abstract_sendrecv_bench.c
+
+benchs_burst_SOURCES = benchs/burst.c
+benchs_burst_SOURCES += benchs/burst_helper.c
+
+if !STARPU_NO_BLAS_LIB
+benchs_sendrecv_gemm_bench_SOURCES = benchs/sendrecv_gemm_bench.c
+benchs_sendrecv_gemm_bench_SOURCES += benchs/bench_helper.c
+benchs_sendrecv_gemm_bench_SOURCES += benchs/gemm_helper.c
+benchs_sendrecv_gemm_bench_SOURCES += benchs/abstract_sendrecv_bench.c
+benchs_sendrecv_gemm_bench_SOURCES += ../../examples/common/blas.c
+
+benchs_sendrecv_gemm_bench_LDADD = $(STARPU_BLAS_LDFLAGS)
+
+benchs_burst_gemm_SOURCES = benchs/burst_gemm.c
+benchs_burst_gemm_SOURCES += benchs/gemm_helper.c
+benchs_burst_gemm_SOURCES += benchs/burst_helper.c
+benchs_burst_gemm_SOURCES += ../../examples/common/blas.c
+
+benchs_burst_gemm_LDADD = $(STARPU_BLAS_LDFLAGS)
+endif

mpi/tests/burst.c

@@ -49,13 +49,11 @@ void parse_args(int argc, char **argv)
 
 int main(int argc, char **argv)
 {
-	int ret, rank, mpi_init, other_rank;
+	int ret, rank, other_rank;
 
 	parse_args(argc, argv);
 
-	MPI_INIT_THREAD(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_init);
-
-	ret = starpu_mpi_init_conf(&argc, &argv, mpi_init, MPI_COMM_WORLD, NULL);
+	ret = starpu_mpi_init_conf(&argc, &argv, 1, MPI_COMM_WORLD, NULL);
 	STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_init_conf");
 
 	starpu_mpi_comm_rank(MPI_COMM_WORLD, &rank);
@@ -68,8 +66,6 @@ int main(int argc, char **argv)
 	burst_free_data(rank);
 
 	starpu_mpi_shutdown();
-	if (!mpi_init)
-		MPI_Finalize();
 
 	return 0;
 }

mpi/tests/burst_gemm.c

@@ -90,12 +90,11 @@ void parse_args(int argc, char **argv)
 
 int main(int argc, char **argv)
 {
-	int ret, mpi_init, worldsize, mpi_rank;
+	int ret, worldsize, mpi_rank;
 
 	parse_args(argc, argv);
 
-	MPI_INIT_THREAD(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_init);
-	ret = starpu_mpi_init_conf(&argc, &argv, mpi_init, MPI_COMM_WORLD, NULL);
+	ret = starpu_mpi_init_conf(&argc, &argv, 1, MPI_COMM_WORLD, NULL);
 	if (ret == -ENODEV)
 		return 77;
 	STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_init_conf");
@@ -109,8 +108,7 @@ int main(int argc, char **argv)
 			FPRINTF(stderr, "We need 2 processes.\n");
 
 		starpu_mpi_shutdown();
-		if (!mpi_init)
-			MPI_Finalize();
+
 		return STARPU_TEST_SKIPPED;
 	}
 
@@ -203,8 +201,6 @@ enodev:
 	burst_free_data(mpi_rank);
 
 	starpu_mpi_shutdown();
-	if (!mpi_init)
-		MPI_Finalize();
 
 	return ret;
 }

mpi/tests/sendrecv_bench.c

@@ -26,7 +26,6 @@
 int main(int argc, char **argv)
 {
 	int ret, rank, worldsize;
-	int mpi_init;
 	int pause_workers = 0;
 
 
@@ -52,8 +51,7 @@ int main(int argc, char **argv)
 	}
 
 
-	MPI_INIT_THREAD(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_init);
-	ret = starpu_mpi_init_conf(&argc, &argv, mpi_init, MPI_COMM_WORLD, NULL);
+	ret = starpu_mpi_init_conf(&argc, &argv, 1, MPI_COMM_WORLD, NULL);
 	STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_init_conf");
 
 	starpu_mpi_comm_rank(MPI_COMM_WORLD, &rank);
@@ -65,8 +63,7 @@ int main(int argc, char **argv)
 			FPRINTF(stderr, "We need 2 processes.\n");
 
 		starpu_mpi_shutdown();
-		if (!mpi_init)
-			MPI_Finalize();
+
 		return STARPU_TEST_SKIPPED;
 	}
 
@@ -85,8 +82,6 @@ int main(int argc, char **argv)
 	}
 
 	starpu_mpi_shutdown();
-	if (!mpi_init)
-		MPI_Finalize();
 
 	return 0;
 }

mpi/tests/sendrecv_gemm_bench.c

@@ -53,7 +53,7 @@ static void* comm_thread_func(void* arg)
 	{
 		char hostname[65];
 		gethostname(hostname, sizeof(hostname));
-		_STARPU_DISP("[%s] No core was available for the comm thread. You should increase STARPU_RESERVE_NCPU or decrease STARPU_NCPU\n", hostname);
+		fprintf(stderr, "[%s] No core was available for the comm thread. You should increase STARPU_RESERVE_NCPU or decrease STARPU_NCPU\n", hostname);
 	}
 
 	sendrecv_bench(mpi_rank, &thread_barrier);
@@ -118,7 +118,7 @@ void parse_args(int argc, char **argv)
 int main(int argc, char **argv)
 {
 	double start, end;
-	int ret, mpi_init, worldsize;
+	int ret, worldsize;
 	starpu_pthread_t comm_thread;
 
 	char hostname[255];
@@ -128,8 +128,7 @@ int main(int argc, char **argv)
 
 	starpu_fxt_autostart_profiling(0);
 
-	MPI_INIT_THREAD(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_init);
-	ret = starpu_mpi_init_conf(&argc, &argv, mpi_init, MPI_COMM_WORLD, NULL);
+	ret = starpu_mpi_init_conf(&argc, &argv, 1, MPI_COMM_WORLD, NULL);
 	if (ret == -ENODEV)
 		return 77;
 	STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_init_conf");
@@ -143,8 +142,7 @@ int main(int argc, char **argv)
 			FPRINTF(stderr, "We need 2 processes.\n");
 
 		starpu_mpi_shutdown();
-		if (!mpi_init)
-			MPI_Finalize();
+
 		return STARPU_TEST_SKIPPED;
 	}
 
@@ -162,7 +160,7 @@ int main(int argc, char **argv)
 
 	if (mpi_rank == 0)
 	{
-		PRINTF("# node\tx\ty\tz\tms\tGFlops\n");
+		printf("# node\tx\ty\tz\tms\tGFlops\n");
 	}
 
 	starpu_pause();
@@ -185,7 +183,7 @@ int main(int argc, char **argv)
 	double timing = end - start;
 	double flops = 2.0*((unsigned long long)matrix_dim) * ((unsigned long long)matrix_dim)*((unsigned long long)matrix_dim);
 
-	PRINTF("%s\t%u\t%u\t%u\t%.0f\t%.1f\n", hostname, matrix_dim, matrix_dim, matrix_dim, timing/1000.0, flops/timing/1000.0);
+	printf("%s\t%u\t%u\t%u\t%.0f\t%.1f\n", hostname, matrix_dim, matrix_dim, matrix_dim, timing/1000.0, flops/timing/1000.0);
 
 
 enodev:
@@ -200,8 +198,6 @@ enodev:
 
 	starpu_resume();
 	starpu_mpi_shutdown();
-	if (!mpi_init)
-		MPI_Finalize();
 
 	return ret;
 }

mpi/tests/sendrecv_parallel_tasks_bench.c

@@ -134,10 +134,8 @@ static struct starpu_codelet cl =
 int main(int argc, char **argv)
 {
 	int ret, rank, worldsize;
-	int mpi_init;
 
-	MPI_INIT_THREAD(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_init);
-	ret = starpu_mpi_init_conf(&argc, &argv, mpi_init, MPI_COMM_WORLD, NULL);
+	ret = starpu_mpi_init_conf(&argc, &argv, 1, MPI_COMM_WORLD, NULL);
 	STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_init_conf");
 
 	starpu_mpi_comm_rank(MPI_COMM_WORLD, &rank);
@@ -149,8 +147,7 @@ int main(int argc, char **argv)
 			FPRINTF(stderr, "We need 2 processes.\n");
 
 		starpu_mpi_shutdown();
-		if (!mpi_init)
-			MPI_Finalize();
+
 		return STARPU_TEST_SKIPPED;
 	}
 
@@ -162,8 +159,7 @@ int main(int argc, char **argv)
 	else if (rank >= 2)
 	{
 		starpu_mpi_shutdown();
-		if (!mpi_init)
-			MPI_Finalize();
+
 		return 0;
 	}
 
@@ -222,8 +218,6 @@ int main(int argc, char **argv)
 	free(mpi_tags);
 
 	starpu_mpi_shutdown();
-	if (!mpi_init)
-		MPI_Finalize();
 
 	return 0;
 }

mpi/tests/Makefile.am

@@ -62,11 +62,7 @@ BUILT_SOURCES =
 CLEANFILES = *.gcno *.gcda *.linkinfo starpu_idle_microsec.log
 
 EXTRA_DIST = 				\
-	abstract_sendrecv_bench.h	\
-	bench_helper.h			\
 	helper.h			\
-	gemm_helper.h			\
-	burst_helper.h			\
 	user_defined_datatype_value.h
 
 examplebindir = $(libdir)/starpu/examples/mpi
@@ -142,21 +138,8 @@ starpu_mpi_TESTS +=				\
 	temporary				\
 	user_defined_datatype			\
 	early_stuff				\
-	sendrecv_bench				\
-	burst						\
 	display_bindings
 
-if !STARPU_USE_MPI_MPI
-starpu_mpi_TESTS +=				\
-	sendrecv_parallel_tasks_bench
-endif
-
-if !STARPU_NO_BLAS_LIB
-starpu_mpi_TESTS +=				\
-	sendrecv_gemm_bench			\
-	burst_gemm
-endif
-
 if !STARPU_SIMGRID
 # missing support in simgrid
 starpu_mpi_TESTS +=				\
@@ -244,18 +227,9 @@ noinst_PROGRAMS +=				\
 	starpu_redefine				\
 	load_balancer				\
 	driver					\
-	sendrecv_bench				\
-	sendrecv_parallel_tasks_bench		\
-	burst					\
 	nothing							\
 	display_bindings
 
-if !STARPU_NO_BLAS_LIB
-noinst_PROGRAMS +=				\
-	sendrecv_gemm_bench			\
-	burst_gemm
-endif
-
 XFAIL_TESTS=					\
 	policy_register_toomany			\
 	policy_unregister			\
@@ -285,31 +259,3 @@ mpi_earlyrecv2_SOURCES = mpi_earlyrecv2.c
 mpi_earlyrecv2_SOURCES += ../../examples/interface/complex_interface.c
 mpi_earlyrecv2_sync_SOURCES = mpi_earlyrecv2_sync.c
 mpi_earlyrecv2_sync_SOURCES += ../../examples/interface/complex_interface.c
-
-sendrecv_bench_SOURCES = sendrecv_bench.c
-sendrecv_bench_SOURCES += bench_helper.c
-sendrecv_bench_SOURCES += abstract_sendrecv_bench.c
-
-sendrecv_parallel_tasks_bench_SOURCES = sendrecv_parallel_tasks_bench.c
-sendrecv_parallel_tasks_bench_SOURCES += bench_helper.c
-sendrecv_parallel_tasks_bench_SOURCES += abstract_sendrecv_bench.c
-
-burst_SOURCES = burst.c
-burst_SOURCES += burst_helper.c
-
-if !STARPU_NO_BLAS_LIB
-sendrecv_gemm_bench_SOURCES = sendrecv_gemm_bench.c
-sendrecv_gemm_bench_SOURCES += bench_helper.c
-sendrecv_gemm_bench_SOURCES += gemm_helper.c
-sendrecv_gemm_bench_SOURCES += abstract_sendrecv_bench.c
-sendrecv_gemm_bench_SOURCES += ../../examples/common/blas.c
-
-sendrecv_gemm_bench_LDADD = $(STARPU_BLAS_LDFLAGS)
-
-burst_gemm_SOURCES = burst_gemm.c
-burst_gemm_SOURCES += gemm_helper.c
-burst_gemm_SOURCES += burst_helper.c
-burst_gemm_SOURCES += ../../examples/common/blas.c
-
-burst_gemm_LDADD = $(STARPU_BLAS_LDFLAGS)
-endif

src/common/utils.c

@@ -743,7 +743,7 @@ int starpu_get_env_size_default(const char *str, int defval)
 
 void starpu_display_bindings(void)
 {
-#ifdef STARPU_HAVE_HWLOC
+#if defined(STARPU_HAVE_HWLOC) && !defined(STARPU_SIMGRID)
 	int hwloc_ret = system("hwloc-ps -a -t -c");
 	if (hwloc_ret)
 	{

src/profiling/profiling.c

@@ -46,6 +46,7 @@ static struct timespec executing_start_date[STARPU_NMAXWORKERS];
 #ifdef STARPU_PAPI
 static int papi_events[PAPI_MAX_HWCTRS];
 static int papi_nevents = 0;
+static int warned_component_unavailable = 0;
 #endif
 
 /* Store the busid of the different (src, dst) pairs. busid_matrix[src][dst]
@@ -160,7 +161,7 @@ void _starpu_profiling_init(void)
 		conf_papi_events = starpu_getenv("STARPU_PROF_PAPI_EVENTS");
 		if (conf_papi_events != NULL)
 		{
-			while ((papi_event_name = strtok_r(conf_papi_events, " ", &conf_papi_events)))
+			while ((papi_event_name = strtok_r(conf_papi_events, " ,", &conf_papi_events)))
 			{
 				_STARPU_DEBUG("Loading PAPI Event:%s\n", papi_event_name);
 				retval = PAPI_event_name_to_code ((char*)papi_event_name, &papi_events[papi_nevents]);
@@ -188,7 +189,12 @@ void _starpu_profiling_papi_task_start_counters(struct starpu_task *task)
 		PAPI_create_eventset(&profiling_info->papi_event_set);
 		for(int i=0; i<papi_nevents; i++)
 		{
-			PAPI_add_event(profiling_info->papi_event_set, papi_events[i]);
+			int ret = PAPI_add_event(profiling_info->papi_event_set, papi_events[i]);
+			if (ret == PAPI_ECMP_DISABLED && !warned_component_unavailable)
+			{
+				_STARPU_MSG("Error while registering Papi event: Component containing event is disabled. Try running `papi_component_avail` to get more information.\n");
+				warned_component_unavailable = 1;
+			}
 			profiling_info->papi_values[i]=0;
 		}
 		PAPI_reset(profiling_info->papi_event_set);

src/sched_policies/component_heft.c

@@ -2,6 +2,7 @@
  *
  * Copyright (C) 2013-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  * Copyright (C) 2013       Simon Archipoff
+ * Copyright (C) 2020       Télécom-Sud Paris
  *
  * StarPU is free software; you can redistribute it and/or modify
  * it under the terms of the GNU Lesser General Public License as published by

src/sched_policies/component_heteroprio.c

@@ -2,6 +2,7 @@
  *
  * Copyright (C) 2013-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  * Copyright (C) 2013       Simon Archipoff
+ * Copyright (C) 2020       Télécom-Sud Paris
  *
  * StarPU is free software; you can redistribute it and/or modify
  * it under the terms of the GNU Lesser General Public License as published by

src/sched_policies/component_mct.c

@@ -2,6 +2,7 @@
  *
  * Copyright (C) 2013-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  * Copyright (C) 2013       Simon Archipoff
+ * Copyright (C) 2020       Télécom-Sud Paris
  *
  * StarPU is free software; you can redistribute it and/or modify
  * it under the terms of the GNU Lesser General Public License as published by

src/sched_policies/deque_modeling_policy_data_aware.c

@@ -6,6 +6,7 @@
  * Copyright (C) 2013       Simon Archipoff
  * Copyright (C) 2013       Thibaut Lambert
  * Copyright (C) 2016       Uppsala University
+ * Copyright (C) 2020       Télécom-Sud Paris
  *
  * StarPU is free software; you can redistribute it and/or modify
  * it under the terms of the GNU Lesser General Public License as published by
@@ -1014,6 +1015,10 @@ static void initialize_dmda_policy(unsigned sched_ctx_id)
 	dt->alpha = starpu_get_env_float_default("STARPU_SCHED_ALPHA", _STARPU_SCHED_ALPHA_DEFAULT);
 	dt->beta = starpu_get_env_float_default("STARPU_SCHED_BETA", _STARPU_SCHED_BETA_DEFAULT);
 	/* data->_gamma: cost of one Joule in us. If gamma is set to 10^6, then one Joule cost 1s */
+#ifdef STARPU_NON_BLOCKING_DRIVERS
+	if (starpu_getenv("STARPU_SCHED_GAMMA"))
+		_STARPU_DISP("Warning: STARPU_SCHED_GAMMA was used, but --enable-blocking-drivers configuration was not set, CPU cores will not actually be sleeping\n");
+#endif
 	dt->_gamma = starpu_get_env_float_default("STARPU_SCHED_GAMMA", _STARPU_SCHED_GAMMA_DEFAULT);
 	/* data->idle_power: Idle power of the whole machine in Watt */
 	dt->idle_power = starpu_get_env_float_default("STARPU_IDLE_POWER", 0.0);

src/sched_policies/helper_mct.c

@@ -2,6 +2,7 @@
  *
  * Copyright (C) 2013-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  * Copyright (C) 2013       Simon Archipoff
+ * Copyright (C) 2020       Télécom-Sud Paris
  *
  * StarPU is free software; you can redistribute it and/or modify
  * it under the terms of the GNU Lesser General Public License as published by
@@ -45,6 +46,10 @@ struct _starpu_mct_data *starpu_mct_init_parameters(struct starpu_sched_componen
 	{
 		data->alpha = starpu_get_env_float_default("STARPU_SCHED_ALPHA", _STARPU_SCHED_ALPHA_DEFAULT);
 		data->beta = starpu_get_env_float_default("STARPU_SCHED_BETA", _STARPU_SCHED_BETA_DEFAULT);
+#ifdef STARPU_NON_BLOCKING_DRIVERS
+		if (starpu_getenv("STARPU_SCHED_GAMMA"))
+			_STARPU_DISP("Warning: STARPU_SCHED_GAMMA was used, but --enable-blocking-drivers configuration was not set, CPU cores will not actually be sleeping\n");
+#endif
 		data->_gamma = starpu_get_env_float_default("STARPU_SCHED_GAMMA", _STARPU_SCHED_GAMMA_DEFAULT);
 		data->idle_power = starpu_get_env_float_default("STARPU_IDLE_POWER", 0.0);
 	}

src/sched_policies/helper_mct.h

@@ -1,6 +1,7 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
  * Copyright (C) 2013-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2020       Télécom-Sud Paris
  *
  * StarPU is free software; you can redistribute it and/or modify
  * it under the terms of the GNU Lesser General Public License as published by

src/sched_policies/parallel_heft.c

@@ -567,6 +567,10 @@ static void initialize_parallel_heft_policy(unsigned sched_ctx_id)
 
 	hd->alpha = starpu_get_env_float_default("STARPU_SCHED_ALPHA", _STARPU_SCHED_ALPHA_DEFAULT);
 	hd->beta = starpu_get_env_float_default("STARPU_SCHED_BETA", _STARPU_SCHED_BETA_DEFAULT);
+#ifdef STARPU_NON_BLOCKING_DRIVERS
+	if (starpu_getenv("STARPU_SCHED_GAMMA"))
+		_STARPU_DISP("Warning: STARPU_SCHED_GAMMA was used, but --enable-blocking-drivers configuration was not set, CPU cores will not actually be sleeping\n");
+#endif
 	hd->_gamma = starpu_get_env_float_default("STARPU_SCHED_GAMMA", _STARPU_SCHED_GAMMA_DEFAULT);
 	hd->idle_power = starpu_get_env_float_default("STARPU_IDLE_POWER", 0.0);
 

starpufft/src/starpufftx3d.c

@@ -164,7 +164,6 @@ static struct starpu_task *
 STARPUFFT(start3dC2C)(STARPUFFT(plan) plan, starpu_data_handle_t in, starpu_data_handle_t out)
 {
 	STARPU_ASSERT(plan->type == C2C);
-	int z;
 	int ret;
 
 if (PARALLEL) {

tests/Makefile.am

@@ -61,7 +61,6 @@ EXTRA_DIST =					\
 	datawizard/interfaces/block/block_opencl_kernel.cl \
 	datawizard/interfaces/tensor/tensor_opencl_kernel.cl \
 	perfmodels/opencl_memset_kernel.cl \
-	perfmodels/opencl_memset_kernel_01.cl \
 	$(MICROBENCHS:=.sh) \
 	microbenchs/microbench.sh \
 	model-checking/platform.xml \
@@ -359,7 +358,8 @@ myPROGRAMS +=				\
 	perfmodels/regression_based		\
 	perfmodels/regression_based_01		\
 	perfmodels/regression_based_02		\
-	perfmodels/regression_based_03		\	
+	perfmodels/regression_based_03		\
+	perfmodels/regression_based_04		\
 	perfmodels/non_linear_regression_based	\
 	perfmodels/feed				\
 	perfmodels/user_base			\
@@ -1011,6 +1011,9 @@ perfmodels_regression_based_02_SOURCES=\
 perfmodels_regression_based_03_SOURCES=\
 	perfmodels/regression_based_03.c
 
+perfmodels_regression_based_04_SOURCES=\
+	perfmodels/regression_based_04.c
+
 perfmodels_max_fpga_SOURCES=\
 	perfmodels/max_fpga.c
 perfmodels_max_fpga_LDADD = $(LDADD) \
@@ -1020,6 +1023,9 @@ if STARPU_USE_OPENCL
 perfmodels_regression_based_SOURCES+=\
 	perfmodels/opencl_memset.c
 
+perfmodels_regression_based_04_SOURCES+=\
+	perfmodels/opencl_memset.c
+
 nobase_STARPU_OPENCL_DATA_DATA += \
 	perfmodels/opencl_memset_kernel.cl
 endif

tests/datawizard/interfaces/bcsr/bcsr_opencl.c

@@ -95,6 +95,8 @@ test_bcsr_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

tests/datawizard/interfaces/coo/coo_opencl.c

@@ -93,6 +93,8 @@ test_coo_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

tests/datawizard/interfaces/csr/csr_opencl.c

@@ -93,6 +93,8 @@ test_csr_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

tests/datawizard/interfaces/matrix/matrix_opencl.c

@@ -92,6 +92,8 @@ void test_matrix_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

tests/datawizard/interfaces/multiformat/multiformat_conversion_codelets_opencl.c

@@ -84,6 +84,8 @@ void cpu_to_opencl_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

tests/datawizard/interfaces/multiformat/multiformat_opencl.c

@@ -98,6 +98,8 @@ void test_multiformat_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

tests/datawizard/interfaces/tensor/tensor_opencl.c

@@ -87,7 +87,7 @@ test_tensor_opencl_func(void *buffers[], void *args)
 	}
 			
 	{
-		size_t global = nx * ny * nz * nt;
+                size_t global = 1;
 		err = clEnqueueNDRangeKernel(queue,
 					     kernel,
 					     1,

tests/datawizard/interfaces/variable/variable_opencl.c

@@ -73,24 +73,12 @@ void test_variable_opencl_func(void *buffers[], void *args)
 
 	{
 		size_t global = 1;
-		size_t local;
+                size_t local = 1;
                 size_t s;
                 cl_device_id device;
 
                 starpu_opencl_get_device(devid, &device);
 
-                err = clGetKernelWorkGroupInfo (kernel,
-						device,
-						CL_KERNEL_WORK_GROUP_SIZE,
-						sizeof(local),
-						&local,
-						&s);
-                if (err != CL_SUCCESS)
-			STARPU_OPENCL_REPORT_ERROR(err);
-
-                if (local > global)
-			local = global;
-
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,
 					1,

tests/datawizard/interfaces/vector/vector_opencl.c

@@ -91,6 +91,8 @@ test_vector_opencl_func(void *buffers[], void *args)
 
                 if (local > global)
 			local = global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue,
 					kernel,

tests/datawizard/scal.c

@@ -73,6 +73,7 @@ void scal_func_opencl(void *buffers[], void *cl_arg)
                 err = clGetKernelWorkGroupInfo (kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(local), &local, &s);
                 if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
                 if (local > global) local=global;
+                else global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
 		if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);

tests/datawizard/scratch_opencl.c

@@ -73,6 +73,8 @@ void opencl_f(void *buffers[], void *args)
 			STARPU_OPENCL_REPORT_ERROR(err);
                 if (local > global)
 			local=global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
 		if (err != CL_SUCCESS)

tests/main/display_binding.c

@@ -38,6 +38,8 @@ int main(void)
 	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
 	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
 
+	starpu_shutdown();
+
 	return EXIT_SUCCESS;
 }
 #endif

tests/perfmodels/opencl_memset.c

@@ -22,7 +22,7 @@
 
 extern struct starpu_opencl_program opencl_program;
 
-void memset_opencl(void *buffers[], void *args)
+void _memset_opencl(void *buffers[], void *args, const char *name)
 {
 	(void) args;
 	int id, devid;
@@ -36,7 +36,7 @@ void memset_opencl(void *buffers[], void *args)
 	id = starpu_worker_get_id_check();
 	devid = starpu_worker_get_devid(id);
 
-	err = starpu_opencl_load_kernel(&kernel, &queue, &opencl_program, "_memset_opencl", devid);
+	err = starpu_opencl_load_kernel(&kernel, &queue, &opencl_program, name, devid);
 	if (err != CL_SUCCESS)
 		STARPU_OPENCL_REPORT_ERROR(err);
 
@@ -58,6 +58,8 @@ void memset_opencl(void *buffers[], void *args)
 			STARPU_OPENCL_REPORT_ERROR(err);
                 if (local > global)
 			local=global;
+                else
+                        global = (global + local-1) / local * local;
 
 		err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
 		if (err != CL_SUCCESS)
@@ -65,3 +67,13 @@ void memset_opencl(void *buffers[], void *args)
 	}
 	starpu_opencl_release_kernel(kernel);
 }
+
+void memset_opencl(void *buffers[], void *args, const char *kernel)
+{
+	_memset_opencl(buffers, args, "_memset_opencl");
+}
+
+void memset0_opencl(void *buffers[], void *args, const char *kernel)
+{
+	_memset_opencl(buffers, args, "_memset0_opencl");
+}

tests/perfmodels/opencl_memset_kernel.cl

@@ -14,6 +14,13 @@
  * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  */
 
+ __kernel void _memset0_opencl(__global int *val, int nx)
+{
+        const int i = get_global_id(0);
+        if (i < nx)
+                val[0] += i;
+}
+
 __kernel void _memset_opencl(__global int *val, int nx)
 {
         const int i = get_global_id(0);

tests/perfmodels/opencl_memset_kernel_01.cl

@@ -1,31 +0,0 @@
-/* StarPU --- Runtime system for heterogeneous multicore architectures.
- *
- * Copyright (C) 2012-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
- *
- * StarPU is free software; you can redistribute it and/or modify
- * it under the terms of the GNU Lesser General Public License as published by
- * the Free Software Foundation; either version 2.1 of the License, or (at
- * your option) any later version.
- *
- * StarPU is distributed in the hope that it will be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
- *
- * See the GNU Lesser General Public License in COPYING.LGPL for more details.
- */
-
- __kernel void _memset0_opencl(__global int *val, int nx)
-{
-        const int i = get_global_id(0);
-        if (i < nx)
-                val[0] += i;
-}
-
-__kernel void _memset_opencl(__global int *val, int nx)
-{
-        const int i = get_global_id(0);
-        if (i < nx)
-                val[i] = 42;
-}
-
-

tests/perfmodels/regression_based.c

@@ -23,6 +23,7 @@
  * Benchmark memset with a linear regression
  */
 
+#define STARTlin 1024
 #define START 1024
 #ifdef STARPU_QUICK_CHECK
 #define END 1048576
@@ -184,11 +185,14 @@ int main(int argc, char **argv)
 #endif
 
 	int size;
-	for (size = START; size < END; size *= 2)
+	for (size = STARTlin; size < END; size *= 2)
 	{
 		/* Use a linear regression */
 		test_memset(size, &memset_cl);
+	}
 
+	for (size = START; size < END; size *= 2)
+	{
 		/* Use a non-linear regression */
 		test_memset(size, &nl_memset_cl);
 	}

tests/perfmodels/regression_based_01.c

@@ -1,6 +1,7 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2010-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2011-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2011       Télécom-SudParis
  *
  * StarPU is free software; you can redistribute it and/or modify
  * it under the terms of the GNU Lesser General Public License as published by
@@ -14,23 +15,20 @@
  * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  */
 
-/*
- * Dans ce benchmark:
- - calibrer le modèle linéaire seulement pour des grandes tailles STARTlin 1048576
- - séparer la boucle test_memset en deux boucles:
-        *linéaire: démarrer à partir de 1 048 576
-        *non linéaire: conserver le démarrage à 1024
- */
-
 #include <starpu.h>
 #include <assert.h>
 #include <starpu_scheduler.h>
 #include <unistd.h>
 #include "../helper.h"
 
-#define FPRINTF(ofile, fmt, ...) do { if (!getenv("STARPU_SSILENT")) {fprintf(ofile, fmt, ## __VA_ARGS__); }} while(0)
+/*
+ * - Calibrate the linear model only for large sizes: STARTline 1048576
+ * - Separate the test_memset loop in two loops:
+ *   - linear: start from 1048576
+ *   - non-linear: keep start at 1024
+ */
 
-#define STARTlin 1048576
+#define STARTlin 131072
 #define START 1024
 #ifdef STARPU_QUICK_CHECK
 #define END 1048576
@@ -38,168 +36,159 @@
 #define END 16777216
 #endif
 
-int ret;
-
 
 void memset_cpu(void *descr[], void *arg)
 {
-    (void)arg;
-    STARPU_SKIP_IF_VALGRIND;
+	(void)arg;
+	STARPU_SKIP_IF_VALGRIND;
 
-    int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
-    unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+	int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
+	unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+	unsigned i;
 
-    usleep(1000);
-    int i;
+	usleep(1000);
 
-    for (i=0; i<n ; i++)
-    {
+	for (i=0; i<n ; i++)
+	{
 
-        ptr[0] += i;
-    }
+		ptr[0] += i;
+	}
 }
 
-
 static struct starpu_perfmodel model =
 {
-    .type = STARPU_REGRESSION_BASED,
-    .symbol = "memset_regression_based"
+	.type = STARPU_REGRESSION_BASED,
+	.symbol = "memset_regression_based"
 };
 
 static struct starpu_perfmodel nl_model =
 {
-    .type = STARPU_NL_REGRESSION_BASED,
-    .symbol = "non_linear_memset_regression_based"
+	.type = STARPU_NL_REGRESSION_BASED,
+	.symbol = "non_linear_memset_regression_based"
 };
 
 static struct starpu_codelet memset_cl =
 {
-    .cpu_funcs = {memset_cpu},
-    .cpu_funcs_name = {"memset_cpu"},
-    .model = &model,
-    .nbuffers = 1,
-    .modes = {STARPU_W}
+	.cpu_funcs = {memset_cpu},
+	.cpu_funcs_name = {"memset_cpu"},
+	.model = &model,
+	.nbuffers = 1,
+	.modes = {STARPU_W}
 };
 
 static struct starpu_codelet nl_memset_cl =
 {
-    .cpu_funcs = {memset_cpu},
-    .cpu_funcs_name = {"memset_cpu"},
-    .model = &nl_model,
-    .nbuffers = 1,
-    .modes = {STARPU_W}
+	.cpu_funcs = {memset_cpu},
+	.cpu_funcs_name = {"memset_cpu"},
+	.model = &nl_model,
+	.nbuffers = 1,
+	.modes = {STARPU_W}
 };
 
-
 static void test_memset(int nelems, struct starpu_codelet *codelet)
 {
-    int nloops = 100;
-    int loop;
-    starpu_data_handle_t handle;
+	int nloops = 100;
+	int loop;
+	starpu_data_handle_t handle;
 
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, nelems, sizeof(int));
-    for (loop = 0; loop < nloops; loop++)
-    {
-        struct starpu_task *task = starpu_task_create();
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, nelems, sizeof(int));
+	for (loop = 0; loop < nloops; loop++)
+	{
+		struct starpu_task *task = starpu_task_create();
 
-        task->cl = codelet;
-        task->handles[0] = handle;
+		task->cl = codelet;
+		task->handles[0] = handle;
 
-        int ret = starpu_task_submit(task);
-        if (ret == -ENODEV)
-            exit(STARPU_TEST_SKIPPED);
-        STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
-    }
+		int ret = starpu_task_submit(task);
+		if (ret == -ENODEV)
+			exit(STARPU_TEST_SKIPPED);
+		STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
+	}
 
-    starpu_data_unregister(handle);
+	starpu_data_unregister(handle);
 }
 
-
-static void compare_performance(int size, struct starpu_codelet *codelet, struct starpu_task *task)
+static void compare_performance(int size, struct starpu_codelet *codelet, struct starpu_task *compar_task)
 {
-    unsigned i;
-    int niter = 100;
-    starpu_data_handle_t handle;
-
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
+	unsigned i;
+	unsigned niter = 100;
+	starpu_data_handle_t handle;
 
-    struct starpu_task **tasks = (struct starpu_task **) malloc(niter*sizeof(struct starpu_task *));
-    assert(tasks);
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
 
-    for (i = 0; i < niter; i++)
-    {
-        struct starpu_task *task = starpu_task_create();
+	struct starpu_task **tasks = (struct starpu_task **) malloc(niter*sizeof(struct starpu_task *));
+	assert(tasks);
 
-        task->cl = codelet;
-        task->handles[0] = handle;
+	for (i = 0; i < niter; i++)
+	{
+		struct starpu_task *task = starpu_task_create();
 
-        /* create a synchronous task: any call to starpu_task_submit will block
-         * until it is terminated */
+		task->cl = codelet;
+		task->handles[0] = handle;
 
-        task->synchronous = 1;
+		task->synchronous = 1;
 
-        /* We will destroy the task structure by hand so that we can
-         * query the profiling info before the task is destroyed. */
-        task->destroy = 0;
+		/* We will destroy the task structure by hand so that we can
+		 * query the profiling info before the task is destroyed. */
+		task->destroy = 0;
 
-        tasks[i] = task;
+		tasks[i] = task;
 
-        ret = starpu_task_submit(task);
+		int ret = starpu_task_submit(task);
 
-        if (STARPU_UNLIKELY(ret == -ENODEV))
-        {
-            FPRINTF(stderr, "No worker may execute this task\n");
-            exit(0);
-        }
-    }
+		if (STARPU_UNLIKELY(ret == -ENODEV))
+		{
+			FPRINTF(stderr, "No worker may execute this task\n");
+			exit(0);
+		}
+	}
 
-    starpu_data_unregister(handle);
+	starpu_data_unregister(handle);
 
-    starpu_task_wait_for_all();
+	starpu_task_wait_for_all();
 
-    double length_sum = 0.0;
+	double length_sum = 0.0;
 
-    for (i = 0; i < niter; i++)
-    {
-        struct starpu_task *task = tasks[i];
-        struct starpu_profiling_task_info *info = task->profiling_info;
+	for (i = 0; i < niter; i++)
+	{
+		struct starpu_task *task = tasks[i];
+		struct starpu_profiling_task_info *info = task->profiling_info;
 
 
-        /* How long was the task execution ? */
-        length_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
+		/* How long was the task execution ? */
+		length_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
 
-        /* We don't need the task structure anymore */
-        starpu_task_destroy(task);
-    }
+		/* We don't need the task structure anymore */
+		starpu_task_destroy(task);
+	}
 
 
-    /* Display the occupancy of all workers during the test */
-    unsigned worker;
-    for (worker = 0; worker < starpu_worker_get_count(); worker++)
-    {
-        struct starpu_profiling_worker_info worker_info;
-        ret = starpu_profiling_worker_get_info(worker, &worker_info);
-        STARPU_ASSERT(!ret);
+	/* Display the occupancy of all workers during the test */
+	unsigned worker;
+	for (worker = 0; worker < starpu_worker_get_count(); worker++)
+	{
+		struct starpu_profiling_worker_info worker_info;
+		int ret = starpu_profiling_worker_get_info(worker, &worker_info);
+		STARPU_ASSERT(!ret);
 
-        char workername[128];
-        starpu_worker_get_name(worker, workername, sizeof(workername));
-        unsigned nimpl;
+		char workername[128];
+		starpu_worker_get_name(worker, workername, sizeof(workername));
+		unsigned nimpl;
 
 
-        if (starpu_worker_get_type(worker)==STARPU_CPU_WORKER)
-        {
-            FPRINTF(stdout, "\n Worker :%s ::::::::::\n\n", workername);
+		if (starpu_worker_get_type(worker)==STARPU_CPU_WORKER)
+		{
+			FPRINTF(stdout, "\n Worker :%s ::::::::::\n\n", workername);
 
-            for (nimpl = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
-            {
+			for (nimpl = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
+			{
 
-                FPRINTF(stdout, "Expected time for %d on %s (impl %u): %f, Measured time: %f\n",
-                        size, workername, nimpl,starpu_task_expected_length(task, starpu_worker_get_perf_archtype(worker, task->sched_ctx), nimpl), ((length_sum)/niter));
+				FPRINTF(stdout, "Expected time for %d on %s (impl %u): %f, Measured time: %f\n",
+						size, workername, nimpl,starpu_task_expected_length(compar_task, starpu_worker_get_perf_archtype(worker, compar_task->sched_ctx), nimpl), ((length_sum)/niter));
 
-            }
-        }
-
-    }
+			}
+		}
+	}
 
 
 }
@@ -207,78 +196,76 @@ static void compare_performance(int size, struct starpu_codelet *codelet, struct
 
 int main(int argc, char **argv)
 {
-    /* Enable profiling */
-    starpu_profiling_status_set(STARPU_PROFILING_ENABLE);
-
-    struct starpu_conf conf;
-    starpu_data_handle_t handle;
-    int ret;
-
-    starpu_conf_init(&conf);
+	/* Enable profiling */
+	starpu_profiling_status_set(STARPU_PROFILING_ENABLE);
 
-    conf.sched_policy_name = "eager";
-    conf.calibrate = 2;
+	struct starpu_conf conf;
+	starpu_data_handle_t handle;
+	int ret;
 
-    ret = starpu_initialize(&conf, &argc, &argv);
-    if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
+	starpu_conf_init(&conf);
 
-    int size;
-    for (size = STARTlin; size < END; size *= 2)
-    {
-        /* Use a linear regression */
-        test_memset(size, &memset_cl);
-    }
+	conf.sched_policy_name = "eager";
+	conf.calibrate = 2;
 
-    for (size = START; size < END; size *= 2)
-    {
-        /* Use a non-linear regression */
-        test_memset(size, &nl_memset_cl);
-    }
+	ret = starpu_initialize(&conf, &argc, &argv);
+	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
 
-    ret = starpu_task_wait_for_all();
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
+	int size;
+	for (size = STARTlin; size < END; size *= 2)
+	{
+		/* Use a linear regression */
+		test_memset(size, &memset_cl);
+	}
 
-    starpu_shutdown();
+	for (size = START; size < END; size *= 2)
+	{
+		/* Use a non-linear regression */
+		test_memset(size, &nl_memset_cl);
+	}
 
+	ret = starpu_task_wait_for_all();
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
 
-    /* Test Phase */
-    starpu_conf_init(&conf);
+	starpu_shutdown();
 
-    conf.sched_policy_name = "eager";
-    conf.calibrate = 0;
 
-    ret = starpu_initialize(&conf, &argc, &argv);
-    if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
+	/* Test Phase */
+	starpu_conf_init(&conf);
 
-    /* Now create a dummy task just to estimate its duration according to the regression */
+	conf.sched_policy_name = "eager";
+	conf.calibrate = 0;
 
-    size = 1234567;
+	ret = starpu_initialize(&conf, &argc, &argv);
+	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
 
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
+	/* Now create a dummy task just to estimate its duration according to the regression */
 
-    struct starpu_task *task = starpu_task_create();
-    task->cl = &memset_cl;
-    task->handles[0] = handle;
-    task->destroy = 0;
+	size = 1234567;
 
-    FPRINTF(stdout, "\n ////linear regression results////\n");
-    compare_performance(size, &memset_cl,task);
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
 
+	struct starpu_task *task = starpu_task_create();
+	task->cl = &memset_cl;
+	task->handles[0] = handle;
+	task->destroy = 0;
 
-    task->cl = &nl_memset_cl;
+	FPRINTF(stdout, "\n ////linear regression results////\n");
+	compare_performance(size, &memset_cl, task);
 
-    FPRINTF(stdout, "\n ////non linear regression results////\n");
+	task->cl = &nl_memset_cl;
 
-    compare_performance(size, &nl_memset_cl,task);
+	FPRINTF(stdout, "\n ////non linear regression results////\n");
 
+	compare_performance(size, &nl_memset_cl, task);
 
-    starpu_task_destroy(task);
+	starpu_task_destroy(task);
 
-    starpu_data_unregister(handle);
+	starpu_data_unregister(handle);
 
-    starpu_shutdown();
+	starpu_shutdown();
 
-    return 0;
+	return EXIT_SUCCESS;
 }

tests/perfmodels/regression_based_02.c

@@ -33,202 +33,197 @@
 #define END 16777216
 #endif
 
-int ret;
-
-//1er implémentation avec un delai initial (100 us)
+// first implementation with an initial delay (100 us)
 void memset0_cpu(void *descr[], void *arg)
 {
-    (void)arg;
-    STARPU_SKIP_IF_VALGRIND;
+	(void)arg;
+	STARPU_SKIP_IF_VALGRIND;
 
-    int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
-    unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
-    int i;
+	int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
+	unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+	unsigned i;
 
-    //usleep(100);
+	usleep(100);
 
-    for (i=0; i<n ; i++)
-    {
-        ptr[0] += i;
-    }
+	for (i=0; i<n ; i++)
+	{
+		ptr[0] += i;
+	}
 }
 
-//deuxième implémentation sans delai initial usleep() et fait 2.5 plus de tours de boucles
+// second implementation without initial delay but 2.5 more loops
 void memset_cpu(void *descr[], void *arg)
 {
-    (void)arg;
-    STARPU_SKIP_IF_VALGRIND;
+	(void)arg;
+	STARPU_SKIP_IF_VALGRIND;
 
-    int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
-    unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
-    int i;
+	int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
+	unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+	int i;
 
-    for (i=0; i<6.5*n ; i++)
-    {
-        ptr[0] += i;
-    }
+	for (i=0; i<6.5*n ; i++)
+	{
+		ptr[0] += i;
+	}
 }
 
 //fonction pour mesurer l'energie
 double energy_function(struct starpu_task *task, struct starpu_perfmodel_arch *arch, unsigned nimpl)
 
 {
-    double energy;
-    int factor;
-    if (nimpl == 0)
-        factor = 10;
-    else
-        factor = 1;
+	double energy;
+	int factor;
+	if (nimpl == 0)
+		factor = 10;
+	else
+		factor = 1;
 
-    energy=starpu_task_expected_length(task, arch, nimpl)*factor;
+	energy=starpu_task_expected_length(task, arch, nimpl)*factor;
 
-    return energy;
+	return energy;
 }
 
 static struct starpu_perfmodel model =
 {
-    .type = STARPU_REGRESSION_BASED,
-    .symbol = "memset_regression_based"
+	.type = STARPU_REGRESSION_BASED,
+	.symbol = "memset_regression_based"
 };
 
 static struct starpu_perfmodel nl_model =
 {
-    .type = STARPU_NL_REGRESSION_BASED,
-    .symbol = "non_linear_memset_regression_based"
+	.type = STARPU_NL_REGRESSION_BASED,
+	.symbol = "non_linear_memset_regression_based"
 };
 
 static struct starpu_perfmodel nl_energy_model=
 {
-    .type = STARPU_PER_ARCH,
-    .symbol = "non_linear_energy_model",
-    .arch_cost_function={energy_function},
+	.type = STARPU_PER_ARCH,
+	.symbol = "non_linear_energy_model",
+	.arch_cost_function=energy_function,
 };
 
 static struct starpu_codelet memset_cl =
 {
-    .cpu_funcs = {memset0_cpu, memset_cpu},
-    .cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
-    .model = &model,
-    .nbuffers = 1,
-    .modes = {STARPU_W}
+	.cpu_funcs = {memset0_cpu, memset_cpu},
+	.cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
+	.model = &model,
+	.nbuffers = 1,
+	.modes = {STARPU_W}
 };
 
 static struct starpu_codelet nl_memset_cl =
 {
-    .cpu_funcs = {memset0_cpu, memset_cpu},
-    .cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
-    .model = &nl_model,
-    .energy_model = &nl_energy_model,
-    .nbuffers = 1,
-    .modes = {STARPU_W}
+	.cpu_funcs = {memset0_cpu, memset_cpu},
+	.cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
+	.model = &nl_model,
+	.energy_model = &nl_energy_model,
+	.nbuffers = 1,
+	.modes = {STARPU_W}
 };
 
 static void test_memset(int nelems, struct starpu_codelet *codelet)
 {
-    int nloops = 100;
-    int loop;
-    starpu_data_handle_t handle;
+	int nloops = 100;
+	int loop;
+	starpu_data_handle_t handle;
 
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, nelems, sizeof(int));
-    for (loop = 0; loop < nloops; loop++)
-    {
-        struct starpu_task *task = starpu_task_create();
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, nelems, sizeof(int));
+	for (loop = 0; loop < nloops; loop++)
+	{
+		struct starpu_task *task = starpu_task_create();
 
-        task->cl = codelet;
-        task->handles[0] = handle;
+		task->cl = codelet;
+		task->handles[0] = handle;
 
-        int ret = starpu_task_submit(task);
-        if (ret == -ENODEV)
-            exit(STARPU_TEST_SKIPPED);
-        STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
-    }
+		int ret = starpu_task_submit(task);
+		if (ret == -ENODEV)
+			exit(STARPU_TEST_SKIPPED);
+		STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
+	}
 
-    starpu_data_unregister(handle);
+	starpu_data_unregister(handle);
 }
 
-static void compare_performance(int size, struct starpu_codelet *codelet, struct starpu_task *task)
+static void compare_performance(int size, struct starpu_codelet *codelet, struct starpu_task *compar_task)
 {
-    unsigned i;
-    int niter = 100;
-    starpu_data_handle_t handle;
-
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
+	unsigned i;
+	unsigned niter = 100;
+	starpu_data_handle_t handle;
 
-    struct starpu_task **tasks = (struct starpu_task **) malloc(niter*sizeof(struct starpu_task *));
-    assert(tasks);
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
 
-    for (i = 0; i < niter; i++)
-    {
-        //fabriquer la tache
-        struct starpu_task *task = starpu_task_create();
+	struct starpu_task **tasks = (struct starpu_task **) malloc(niter*sizeof(struct starpu_task *));
+	assert(tasks);
 
-        task->cl = codelet;
-        task->handles[0] = handle;
+	for (i = 0; i < niter; i++)
+	{
+		struct starpu_task *task = starpu_task_create();
 
-        task->synchronous = 1;
+		task->cl = codelet;
+		task->handles[0] = handle;
 
-        /* We will destroy the task structure by hand so that we can
-         * query the profiling info before the task is destroyed. */
-        task->destroy = 0;
+		task->synchronous = 1;
 
-        tasks[i] = task;
+		/* We will destroy the task structure by hand so that we can
+		 * query the profiling info before the task is destroyed. */
+		task->destroy = 0;
 
-        //soumettre la tache
-        ret = starpu_task_submit(task);
+		tasks[i] = task;
 
-        if (STARPU_UNLIKELY(ret == -ENODEV))
-        {
-            FPRINTF(stderr, "No worker may execute this task\n");
-            exit(0);
-        }
-    }
+		int ret = starpu_task_submit(task);
 
-    starpu_data_unregister(handle);
+		if (STARPU_UNLIKELY(ret == -ENODEV))
+		{
+			FPRINTF(stderr, "No worker may execute this task\n");
+			exit(0);
+		}
+	}
 
-    starpu_task_wait_for_all();
+	starpu_data_unregister(handle);
 
-    double length_sum = 0.0;
+	starpu_task_wait_for_all();
 
-    for (i = 0; i < niter; i++)
-    {
-        struct starpu_task *task = tasks[i];
+	double length_sum = 0.0;
 
-        struct starpu_profiling_task_info *info = task->profiling_info;
+	for (i = 0; i < niter; i++)
+	{
+		struct starpu_task *task = tasks[i];
+		struct starpu_profiling_task_info *info = task->profiling_info;
 
 
-        /* How long was the task execution ? */
-        length_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
+		/* How long was the task execution ? */
+		length_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
 
-        /* We don't need the task structure anymore */
-        starpu_task_destroy(task);
-    }
+		/* We don't need the task structure anymore */
+		starpu_task_destroy(task);
+	}
 
 
-    /* Display the occupancy of all workers during the test */
-    unsigned worker;
-    for (worker = 0; worker < starpu_worker_get_count(); worker++)
-    {
-        struct starpu_profiling_worker_info worker_info;
-        ret = starpu_profiling_worker_get_info(worker, &worker_info);
-        STARPU_ASSERT(!ret);
+	/* Display the occupancy of all workers during the test */
+	unsigned worker;
+	for (worker = 0; worker < starpu_worker_get_count(); worker++)
+	{
+		struct starpu_profiling_worker_info worker_info;
+		int ret = starpu_profiling_worker_get_info(worker, &worker_info);
+		STARPU_ASSERT(!ret);
 
-        char workername[128];
-        starpu_worker_get_name(worker, workername, sizeof(workername));
-        unsigned nimpl;
+		char workername[128];
+		starpu_worker_get_name(worker, workername, sizeof(workername));
+		unsigned nimpl;
 
-        if (starpu_worker_get_type(worker)==STARPU_CPU_WORKER)
-        {
-            FPRINTF(stdout, "\n Worker :%s ::::::::::\n\n", workername);
+		if (starpu_worker_get_type(worker)==STARPU_CPU_WORKER)
+		{
+			FPRINTF(stdout, "\n Worker :%s ::::::::::\n\n", workername);
 
-            for (nimpl = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
-            {
+			for (nimpl = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
+			{
 
-                FPRINTF(stdout, "Expected time for %d on %s (impl %u): %f, Measured time: %f, Expected energy: %f\n",
-                        size, workername, nimpl,starpu_task_expected_length(task, starpu_worker_get_perf_archtype(worker, task->sched_ctx), nimpl), ((length_sum)/niter),
-                        starpu_task_expected_energy(task, starpu_worker_get_perf_archtype(worker, task->sched_ctx), nimpl));
-            }
-        }
-    }
+				FPRINTF(stdout, "Expected time for %d on %s (impl %u): %f, Measured time: %f, Expected energy: %f\n",
+						size, workername, nimpl,starpu_task_expected_length(compar_task, starpu_worker_get_perf_archtype(worker, compar_task->sched_ctx), nimpl), ((length_sum)/niter),
+						starpu_task_expected_energy(compar_task, starpu_worker_get_perf_archtype(worker, compar_task->sched_ctx), nimpl));
+			}
+		}
+	}
 
 
 }
@@ -236,74 +231,75 @@ static void compare_performance(int size, struct starpu_codelet *codelet, struct
 
 int main(int argc, char **argv)
 {
+	/* Enable profiling */
+	starpu_profiling_status_set(STARPU_PROFILING_ENABLE);
 
-    /* Enable profiling */
-    starpu_profiling_status_set(1);
+	struct starpu_conf conf;
+	starpu_data_handle_t handle;
+	int ret;
 
-    struct starpu_conf conf;
-    starpu_data_handle_t handle;
-    int ret;
+	starpu_conf_init(&conf);
 
-    starpu_conf_init(&conf);
+	conf.sched_policy_name = "dmda";
+	conf.calibrate = 2;
 
-    conf.sched_policy_name = "dmda";
-    conf.calibrate = 2;
+	ret = starpu_initialize(&conf, &argc, &argv);
+	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
 
-    ret = starpu_initialize(&conf, &argc, &argv);
-    if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
-
-    int size;
-    /*for (size = STARTlin; size < END; size *= 2)
-    {
-        /* Use a linear regression */
-    //test_memset(size, &memset_cl);
-    //}
+	int size;
+#if 0
+	for (size = STARTlin; size < END; size *= 2)
+	{
+		/* Use a linear regression */
+		test_memset(size, &memset_cl);
+	}
+#endif
 
-    for (size = START; size < END; size *= 2)
-    {
-        /* Use a non-linear regression */
-        test_memset(size, &nl_memset_cl);
-    }
+	for (size = START; size < END; size *= 2)
+	{
+		/* Use a non-linear regression */
+		test_memset(size, &nl_memset_cl);
+	}
 
-    ret = starpu_task_wait_for_all();
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
+	ret = starpu_task_wait_for_all();
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
 
-    starpu_shutdown();
+	starpu_shutdown();
 
 
-    /* Test Phase */
-    starpu_conf_init(&conf);
+	/* Test Phase */
+	starpu_conf_init(&conf);
 
-    conf.sched_policy_name = "dmda";
-    conf.calibrate = 0;
+	conf.sched_policy_name = "dmda";
+	conf.calibrate = 0;
 
-    ret = starpu_initialize(&conf, &argc, &argv);
-    if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
+	ret = starpu_initialize(&conf, &argc, &argv);
+	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
 
-    /* Now create a dummy task just to estimate its duration according to the regression */
+	/* Now create a dummy task just to estimate its duration according to the regression */
 
-    size = 1234567;
+	size = 1234567;
 
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
 
-    struct starpu_task *task = starpu_task_create();
-    task->cl = &memset_cl;
-    task->handles[0] = handle;
-    task->destroy = 0;
+	struct starpu_task *task = starpu_task_create();
+	task->cl = &memset_cl;
+	task->handles[0] = handle;
+	task->destroy = 0;
 
-    task->cl = &nl_memset_cl;
+	task->cl = &nl_memset_cl;
 
-    FPRINTF(stdout, "\n ////non linear regression results////\n");
+	FPRINTF(stdout, "\n ////non linear regression results////\n");
 
-    compare_performance(size, &nl_memset_cl,task);
+	compare_performance(size, &nl_memset_cl, task);
 
-    starpu_task_destroy(task);
+	starpu_task_destroy(task);
 
-    starpu_data_unregister(handle);
+	starpu_data_unregister(handle);
 
-    starpu_shutdown();
+	starpu_shutdown();
 
-    return EXIT_SUCCESS;
+	return EXIT_SUCCESS;
 }

tests/perfmodels/regression_based_03.c

@@ -34,204 +34,197 @@
 #define END 16777216
 #endif
 
-
-int ret;
-
-//1er implémentation avec un delai initial (100 us)
+// first implementation with an initial delay (100 us)
 void memset0_cpu(void *descr[], void *arg)
 {
-    (void)arg;
-    STARPU_SKIP_IF_VALGRIND;
+	(void)arg;
+	STARPU_SKIP_IF_VALGRIND;
 
-    int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
-    unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
-    int i;
+	int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
+	unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+	unsigned i;
 
-    //usleep () function
-    //usleep(100);
+	usleep(100);
 
-    for (i=0; i<n ; i++)
-    {
-        ptr[0] += i;
-    }
+	for (i=0; i<n ; i++)
+	{
+		ptr[0] += i;
+	}
 }
 
-//deuxième implémentation sans delai initial usleep() et fait 2.5 plus de tours de boucles
+// second implementation without initial delay but 2.5 more loops
 void memset_cpu(void *descr[], void *arg)
 {
-    (void)arg;
-    STARPU_SKIP_IF_VALGRIND;
+	(void)arg;
+	STARPU_SKIP_IF_VALGRIND;
 
-    int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
-    unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
-    int i;
+	int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
+	unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+	int i;
 
-    for (i=0; i<6.5*n ; i++)
-    {
-        ptr[0] += i;
-    }
+	for (i=0; i<6.5*n ; i++)
+	{
+		ptr[0] += i;
+	}
 }
 
 //fonction pour mesurer l'energie
 double energy_function(struct starpu_task *task, struct starpu_perfmodel_arch *arch, unsigned nimpl)
 
 {
-    double energy;
-    int factor;
-    if (nimpl == 0)
-        factor = 10;
-    else
-        factor = 1;
+	double energy;
+	int factor;
+	if (nimpl == 0)
+		factor = 10;
+	else
+		factor = 1;
 
-    energy=starpu_task_expected_length(task, arch, nimpl)*factor;
+	energy=starpu_task_expected_length(task, arch, nimpl)*factor;
 
-    return energy;
+	return energy;
 }
 
 static struct starpu_perfmodel model =
 {
-    .type = STARPU_REGRESSION_BASED,
-    .symbol = "memset_regression_based"
+	.type = STARPU_REGRESSION_BASED,
+	.symbol = "memset_regression_based"
 };
 
 static struct starpu_perfmodel nl_model =
 {
-    .type = STARPU_NL_REGRESSION_BASED,
-    .symbol = "non_linear_memset_regression_based"
+	.type = STARPU_NL_REGRESSION_BASED,
+	.symbol = "non_linear_memset_regression_based"
 };
 
 static struct starpu_perfmodel nl_energy_model=
 {
-    .type = STARPU_PER_ARCH,
-    .symbol = "non_linear_energy_model",
-    .arch_cost_function={energy_function},
+	.type = STARPU_PER_ARCH,
+	.symbol = "non_linear_energy_model",
+	.arch_cost_function=energy_function,
 };
 
 static struct starpu_codelet memset_cl =
 {
-    .cpu_funcs = {memset0_cpu, memset_cpu},
-    .cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
-    .model = &model,
-    .nbuffers = 1,
-    .modes = {STARPU_W}
+	.cpu_funcs = {memset0_cpu, memset_cpu},
+	.cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
+	.model = &model,
+	.nbuffers = 1,
+	.modes = {STARPU_W}
 };
 
 static struct starpu_codelet nl_memset_cl =
 {
-    .cpu_funcs = {memset0_cpu, memset_cpu},
-    .cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
-    .model = &nl_model,
-    .energy_model = &nl_energy_model,
-    .nbuffers = 1,
-    .modes = {STARPU_W}
+	.cpu_funcs = {memset0_cpu, memset_cpu},
+	.cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
+	.model = &nl_model,
+	.energy_model = &nl_energy_model,
+	.nbuffers = 1,
+	.modes = {STARPU_W}
 };
 
 static void test_memset(int nelems, struct starpu_codelet *codelet)
 {
-    int nloops = 100;
-    int loop;
-    starpu_data_handle_t handle;
+	int nloops = 100;
+	int loop;
+	starpu_data_handle_t handle;
 
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, nelems, sizeof(int));
-    for (loop = 0; loop < nloops; loop++)
-    {
-        struct starpu_task *task = starpu_task_create();
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, nelems, sizeof(int));
+	for (loop = 0; loop < nloops; loop++)
+	{
+		struct starpu_task *task = starpu_task_create();
 
-        task->cl = codelet;
-        task->handles[0] = handle;
+		task->cl = codelet;
+		task->handles[0] = handle;
 
-        int ret = starpu_task_submit(task);
-        if (ret == -ENODEV)
-            exit(STARPU_TEST_SKIPPED);
-        STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
-    }
+		int ret = starpu_task_submit(task);
+		if (ret == -ENODEV)
+			exit(STARPU_TEST_SKIPPED);
+		STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
+	}
 
-    starpu_data_unregister(handle);
+	starpu_data_unregister(handle);
 }
 
-static void compare_performance(int size, struct starpu_codelet *codelet, struct starpu_task *task)
+static void compare_performance(int size, struct starpu_codelet *codelet, struct starpu_task *compar_task)
 {
-    unsigned i;
-    int niter = 100;
-    starpu_data_handle_t handle;
+	unsigned i;
+	unsigned niter = 100;
+	starpu_data_handle_t handle;
 
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
 
-    struct starpu_task **tasks = (struct starpu_task **) malloc(niter*sizeof(struct starpu_task *));
-    assert(tasks);
+	struct starpu_task **tasks = (struct starpu_task **) malloc(niter*sizeof(struct starpu_task *));
+	assert(tasks);
 
-    for (i = 0; i < niter; i++)
-    {
-        //fabriquer la tache
-        struct starpu_task *task = starpu_task_create();
+	for (i = 0; i < niter; i++)
+	{
+		struct starpu_task *task = starpu_task_create();
 
-        task->cl = codelet;
-        task->handles[0] = handle;
+		task->cl = codelet;
+		task->handles[0] = handle;
 
-        task->synchronous = 1;
+		task->synchronous = 1;
 
-        /* We will destroy the task structure by hand so that we can
-         * query the profiling info before the task is destroyed. */
-        task->destroy = 0;
+		/* We will destroy the task structure by hand so that we can
+		 * query the profiling info before the task is destroyed. */
+		task->destroy = 0;
 
-        tasks[i] = task;
+		tasks[i] = task;
 
-        //soumettre la tache
-        ret = starpu_task_submit(task);
+		int ret = starpu_task_submit(task);
 
-        if (STARPU_UNLIKELY(ret == -ENODEV))
-        {
-            FPRINTF(stderr, "No worker may execute this task\n");
-            exit(0);
-        }
-    }
+		if (STARPU_UNLIKELY(ret == -ENODEV))
+		{
+			FPRINTF(stderr, "No worker may execute this task\n");
+			exit(0);
+		}
+	}
 
-    starpu_data_unregister(handle);
+	starpu_data_unregister(handle);
 
-    starpu_task_wait_for_all();
+	starpu_task_wait_for_all();
 
-    double length_sum = 0.0;
+	double length_sum = 0.0;
 
-    for (i = 0; i < niter; i++)
-    {
-        struct starpu_task *task = tasks[i];
+	for (i = 0; i < niter; i++)
+	{
+		struct starpu_task *task = tasks[i];
+		struct starpu_profiling_task_info *info = task->profiling_info;
 
-        struct starpu_profiling_task_info *info = task->profiling_info;
 
+		/* How long was the task execution ? */
+		length_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
 
-        /* How long was the task execution ? */
-        length_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
+		/* We don't need the task structure anymore */
+		starpu_task_destroy(task);
+	}
 
-        /* We don't need the task structure anymore */
-        starpu_task_destroy(task);
-    }
 
+	/* Display the occupancy of all workers during the test */
+	unsigned worker;
+	for (worker = 0; worker < starpu_worker_get_count(); worker++)
+	{
+		struct starpu_profiling_worker_info worker_info;
+		int ret = starpu_profiling_worker_get_info(worker, &worker_info);
+		STARPU_ASSERT(!ret);
 
-    /* Display the occupancy of all workers during the test */
-    unsigned worker;
-    for (worker = 0; worker < starpu_worker_get_count(); worker++)
-    {
-        struct starpu_profiling_worker_info worker_info;
-        ret = starpu_profiling_worker_get_info(worker, &worker_info);
-        STARPU_ASSERT(!ret);
+		char workername[128];
+		starpu_worker_get_name(worker, workername, sizeof(workername));
+		unsigned nimpl;
 
-        char workername[128];
-        starpu_worker_get_name(worker, workername, sizeof(workername));
-        unsigned nimpl;
+		if (starpu_worker_get_type(worker)==STARPU_CPU_WORKER)
+		{
+			FPRINTF(stdout, "\n Worker :%s ::::::::::\n\n", workername);
 
-        if (starpu_worker_get_type(worker)==STARPU_CPU_WORKER)
-        {
-            FPRINTF(stdout, "\n Worker :%s ::::::::::\n\n", workername);
+			for (nimpl = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
+			{
 
-            for (nimpl = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
-            {
-
-                FPRINTF(stdout, "Expected time for %d on %s (impl %u): %f, Measured time: %f, Expected energy: %f\n",
-                        size, workername, nimpl,starpu_task_expected_length(task, starpu_worker_get_perf_archtype(worker, task->sched_ctx), nimpl), ((length_sum)/niter),
-                        starpu_task_expected_energy(task, starpu_worker_get_perf_archtype(worker, task->sched_ctx), nimpl));
-            }
-        }
-    }
+				FPRINTF(stdout, "Expected time for %d on %s (impl %u): %f, Measured time: %f, Expected energy: %f\n",
+						size, workername, nimpl,starpu_task_expected_length(compar_task, starpu_worker_get_perf_archtype(worker, compar_task->sched_ctx), nimpl), ((length_sum)/niter),
+						starpu_task_expected_energy(compar_task, starpu_worker_get_perf_archtype(worker, compar_task->sched_ctx), nimpl));
+			}
+		}
+	}
 
 
 }
@@ -239,74 +232,73 @@ static void compare_performance(int size, struct starpu_codelet *codelet, struct
 
 int main(int argc, char **argv)
 {
+	/* Enable profiling */
+	starpu_profiling_status_set(STARPU_PROFILING_ENABLE);
 
-    /* Enable profiling */
-    starpu_profiling_status_set(1);
-
-    struct starpu_conf conf;
-    starpu_data_handle_t handle;
-    int ret;
+	struct starpu_conf conf;
+	starpu_data_handle_t handle;
+	int ret;
 
-    starpu_conf_init(&conf);
+	starpu_conf_init(&conf);
 
-    conf.sched_policy_name = "dmda";
-    conf.calibrate = 2;
+	conf.sched_policy_name = "dmda";
+	conf.calibrate = 2;
 
-    ret = starpu_initialize(&conf, &argc, &argv);
-    if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
+	ret = starpu_initialize(&conf, &argc, &argv);
+	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
 
-    int size;
-    for (size = STARTlin; size < END; size *= 2)
-    {
-        /* Use a linear regression */
-        test_memset(size, &memset_cl);
-    }
+	int size;
+	for (size = STARTlin; size < END; size *= 2)
+	{
+		/* Use a linear regression */
+		test_memset(size, &memset_cl);
+	}
 
-    for (size = START; size < END; size *= 2)
-    {
-        /* Use a non-linear regression */
-        test_memset(size, &nl_memset_cl);
-    }
+	for (size = START; size < END; size *= 2)
+	{
+		/* Use a non-linear regression */
+		test_memset(size, &nl_memset_cl);
+	}
 
-    ret = starpu_task_wait_for_all();
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
+	ret = starpu_task_wait_for_all();
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
 
-    starpu_shutdown();
+	starpu_shutdown();
 
 
-    /* Test Phase */
-    starpu_conf_init(&conf);
+	/* Test Phase */
+	starpu_conf_init(&conf);
 
-    conf.sched_policy_name = "dmda";
-    conf.calibrate = 0;
+	conf.sched_policy_name = "dmda";
+	conf.calibrate = 0;
 
-    ret = starpu_initialize(&conf, &argc, &argv);
-    if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
-    STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
+	ret = starpu_initialize(&conf, &argc, &argv);
+	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
 
-    /* Now create a dummy task just to estimate its duration according to the regression */
+	/* Now create a dummy task just to estimate its duration according to the regression */
 
-    size = 1234567;
+	size = 1234567;
 
-    starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
 
-    struct starpu_task *task = starpu_task_create();
-    task->cl = &memset_cl;
-    task->handles[0] = handle;
-    task->destroy = 0;
+	struct starpu_task *task = starpu_task_create();
+	task->cl = &memset_cl;
+	task->handles[0] = handle;
+	task->destroy = 0;
 
-    task->cl = &nl_memset_cl;
+	task->cl = &nl_memset_cl;
 
-    FPRINTF(stdout, "\n ////non linear regression results////\n");
+	FPRINTF(stdout, "\n ////non linear regression results////\n");
 
-    compare_performance(size, &nl_memset_cl,task);
+	compare_performance(size, &nl_memset_cl, task);
 
-    starpu_task_destroy(task);
+	starpu_task_destroy(task);
 
-    starpu_data_unregister(handle);
+	starpu_data_unregister(handle);
 
-    starpu_shutdown();
+	starpu_shutdown();
 
-    return EXIT_SUCCESS;
+	return EXIT_SUCCESS;
 }

tests/perfmodels/regression_based_04.c

@@ -0,0 +1,387 @@
+/* StarPU --- Runtime system for heterogeneous multicore architectures.
+ *
+ * Copyright (C) 2011-2020  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2011       Télécom-SudParis
+ *
+ * StarPU is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU Lesser General Public License as published by
+ * the Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ *
+ * StarPU is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * See the GNU Lesser General Public License in COPYING.LGPL for more details.
+ */
+
+#include <starpu.h>
+#include <starpu_scheduler.h>
+#include "../helper.h"
+
+/*
+ * A multi-implementation benchmark with dmda scheduler
+ * we aim to test OPENCL workers and calculate the estimated time for each type of worker (CPU or OPENCL or CUDA)
+ * dmda choose OPENCL workers for lage size (variable size of compare_performance) size=1234567
+ * dmda choose CPU workers for small size (size=1234)
+ */
+
+#define STARTlin (512*1024)
+#define START 1024
+#ifdef STARPU_QUICK_CHECK
+#define END 1048576
+#else
+#define END 16777216
+#endif
+
+#ifdef STARPU_USE_CUDA
+static void memset_cuda(void *descr[], void *arg)
+{
+	(void)arg;
+	STARPU_SKIP_IF_VALGRIND;
+
+	int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
+	unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+
+	cudaMemsetAsync(ptr, 42, n * sizeof(*ptr), starpu_cuda_get_local_stream());
+}
+#endif
+
+#ifdef STARPU_USE_OPENCL
+extern void memset0_opencl(void *buffers[], void *args);
+extern void memset_opencl(void *buffers[], void *args);
+#endif
+
+void memset0_cpu(void *descr[], void *arg)
+{
+	(void)arg;
+	STARPU_SKIP_IF_VALGRIND;
+
+	int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
+	unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+	unsigned i;
+
+	//starpu_usleep(100);
+
+	for (i = 0; i < n; i++)
+
+		ptr[0] += i;
+}
+
+void memset_cpu(void *descr[], void *arg)
+{
+	(void)arg;
+	STARPU_SKIP_IF_VALGRIND;
+
+	int *ptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
+	unsigned n = STARPU_VECTOR_GET_NX(descr[0]);
+
+	//starpu_usleep(10);
+	memset(ptr, 42, n * sizeof(*ptr));
+}
+
+static struct starpu_perfmodel model =
+{
+	.type = STARPU_REGRESSION_BASED,
+	.symbol = "memset_regression_based"
+};
+
+static struct starpu_perfmodel nl_model =
+{
+	.type = STARPU_NL_REGRESSION_BASED,
+	.symbol = "non_linear_memset_regression_based"
+};
+
+static struct starpu_codelet memset_cl =
+{
+#ifdef STARPU_USE_CUDA
+	.cuda_funcs = {memset_cuda},
+	.cuda_flags = {STARPU_CUDA_ASYNC},
+#endif
+#ifdef STARPU_USE_OPENCL
+	.opencl_funcs = {memset0_opencl, memset_opencl},
+	.opencl_flags = {STARPU_OPENCL_ASYNC},
+#endif
+	.cpu_funcs = {memset0_cpu, memset_cpu},
+	.cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
+	.model = &model,
+	.nbuffers = 1,
+	.modes = {STARPU_W}
+};
+
+static struct starpu_codelet nl_memset_cl =
+{
+#ifdef STARPU_USE_CUDA
+	.cuda_funcs = {memset_cuda},
+	.cuda_flags = {STARPU_CUDA_ASYNC},
+#endif
+#ifdef STARPU_USE_OPENCL
+	.opencl_funcs = {memset0_opencl, memset_opencl},
+	.opencl_flags = {STARPU_OPENCL_ASYNC},
+#endif
+	.cpu_funcs = {memset0_cpu, memset_cpu},
+	.cpu_funcs_name = {"memset0_cpu", "memset_cpu"},
+	.model = &nl_model,
+	.nbuffers = 1,
+	.modes = {STARPU_W}
+};
+
+static void test_memset(int nelems, struct starpu_codelet *codelet)
+{
+	int nloops = 100;
+	int loop;
+	starpu_data_handle_t handle;
+
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, nelems, sizeof(int));
+	for (loop = 0; loop < nloops; loop++)
+	{
+		struct starpu_task *task = starpu_task_create();
+
+		task->cl = codelet;
+		task->handles[0] = handle;
+
+		int ret = starpu_task_submit(task);
+		if (ret == -ENODEV)
+			exit(STARPU_TEST_SKIPPED);
+		STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
+	}
+
+	starpu_data_unregister(handle);
+}
+
+static void compare_performance(int size, struct starpu_codelet *codelet, struct starpu_task *compar_task)
+{
+	unsigned i;
+	unsigned niter = 100;
+	starpu_data_handle_t handle;
+
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
+
+	struct starpu_task **tasks = (struct starpu_task **) malloc(niter*sizeof(struct starpu_task *));
+	assert(tasks);
+
+	for (i = 0; i < niter; i++)
+	{
+		struct starpu_task *task = starpu_task_create();
+
+		task->cl = codelet;
+		task->handles[0] = handle;
+
+		task->synchronous = 1;
+
+		/* We will destroy the task structure by hand so that we can
+		 * query the profiling info before the task is destroyed. */
+		task->destroy = 0;
+
+		tasks[i] = task;
+
+		int ret = starpu_task_submit(task);
+
+		if (STARPU_UNLIKELY(ret == -ENODEV))
+		{
+			FPRINTF(stderr, "No worker may execute this task\n");
+			exit(0);
+		}
+	}
+
+	starpu_data_unregister(handle);
+
+	starpu_task_wait_for_all();
+
+	double length_cpu_sum = 0.0;
+	double length_gpu_sum = 0.0;
+
+	enum starpu_worker_archtype archi;
+
+	for (i = 0; i < niter; i++)
+	{
+		struct starpu_task *task = tasks[i];
+		struct starpu_profiling_task_info *info = task->profiling_info;
+
+		//archi=starpu_worker_get_type(0);
+		archi=starpu_worker_get_type(info->workerid);
+
+		switch (archi)
+		{
+		case STARPU_CPU_WORKER:
+			FPRINTF(stdout, "cpuuu\n");
+			/* How long was the task execution ? */
+			length_cpu_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
+			break;
+
+		case STARPU_OPENCL_WORKER:
+
+			FPRINTF(stdout, "openclllllll\n");
+			/* How long was the task execution ? */
+			length_gpu_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
+			break;
+
+		case STARPU_CUDA_WORKER:
+
+			FPRINTF(stdout, "cudaaaaaa\n");
+			/* How long was the task execution ? */
+			length_gpu_sum += starpu_timing_timespec_delay_us(&info->start_time, &info->end_time);
+			break;
+
+
+	default:
+			FPRINTF(stdout, "unsupported!\n");
+		break;
+		}
+
+		/* We don't need the task structure anymore */
+		starpu_task_destroy(task);
+
+	}
+
+	unsigned worker;
+
+	/* Display the occupancy of all workers during the test */
+	unsigned ncpus =  starpu_cpu_worker_get_count();
+	unsigned ngpus =  starpu_opencl_worker_get_count()+starpu_cuda_worker_get_count();
+	//unsigned ncpu= starpu_worker_get_count_by_type(STARPU_CPU_WORKER);
+
+	FPRINTF(stderr, "ncpus %u \n", ncpus);
+	FPRINTF(stderr, "ngpus %u \n", ngpus);
+	for (worker= 0; worker< starpu_worker_get_count(); worker++)
+	{
+
+		struct starpu_profiling_worker_info worker_info;
+		int ret = starpu_profiling_worker_get_info(worker, &worker_info);
+		STARPU_ASSERT(!ret);
+
+		char workername[128];
+		starpu_worker_get_name(worker, workername, sizeof(workername));
+		unsigned nimpl;
+
+		FPRINTF(stdout, "\n Worker :%s ::::::::::\n\n", workername);
+
+		for (nimpl = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
+		{
+			switch (starpu_worker_get_type(worker))
+
+			{
+			case STARPU_CPU_WORKER:
+
+				FPRINTF(stdout, "Expected time for %d on %s (impl %u): %f, Measured time: %f \n",
+						size, workername, nimpl,starpu_task_expected_length(compar_task, starpu_worker_get_perf_archtype(worker, compar_task->sched_ctx), nimpl), ((length_cpu_sum)/niter));
+
+				break;
+
+			case STARPU_OPENCL_WORKER:
+
+				FPRINTF(stdout, "Expectedd time for %d on %s (impl %u): %f, Measuredd time: %f \n",
+						size, workername, nimpl,starpu_task_expected_length(compar_task, starpu_worker_get_perf_archtype(worker, compar_task->sched_ctx), nimpl), ((length_gpu_sum)/niter));
+
+				break;
+
+			case STARPU_CUDA_WORKER:
+
+				FPRINTF(stdout, "Expectedd time for %d on %s (impl %u): %f, Measuredd time: %f \n",
+						size, workername, nimpl,starpu_task_expected_length(compar_task, starpu_worker_get_perf_archtype(worker, compar_task->sched_ctx), nimpl), ((length_gpu_sum)/niter));
+
+				break;
+
+			default:
+				FPRINTF(stdout, "unsupported!\n");
+				break;
+			}
+		}
+	}
+
+
+}
+
+#ifdef STARPU_USE_OPENCL
+struct starpu_opencl_program opencl_program;
+#endif
+
+int main(int argc, char **argv)
+{
+	/* Enable profiling */
+	starpu_profiling_status_set(STARPU_PROFILING_ENABLE);
+
+	struct starpu_conf conf;
+	starpu_data_handle_t handle;
+	int ret;
+
+	starpu_conf_init(&conf);
+
+	conf.sched_policy_name = "dmda";
+	conf.calibrate = 2;
+
+	ret = starpu_initialize(&conf, &argc, &argv);
+	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
+
+#ifdef STARPU_USE_OPENCL
+	ret = starpu_opencl_load_opencl_from_file("tests/perfmodels/opencl_memset_kernel.cl",
+			&opencl_program, NULL);
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
+#endif
+
+	int size;
+	for (size = STARTlin; size < END; size *= 2)
+	{
+		/* Use a linear regression */
+		test_memset(size, &memset_cl);
+	}
+
+	for (size = START*1.5; size < END; size *= 2)
+	{
+		/* Use a non-linear regression */
+		test_memset(size, &nl_memset_cl);
+	}
+
+	ret = starpu_task_wait_for_all();
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
+
+	starpu_shutdown();
+
+
+	/* Test Phase */
+	starpu_conf_init(&conf);
+
+	conf.sched_policy_name = "dmda";
+	conf.calibrate = 0;
+
+	ret = starpu_initialize(&conf, &argc, &argv);
+	if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
+
+	ret = starpu_opencl_load_opencl_from_file("tests/perfmodels/opencl_memset_kernel.cl",
+			&opencl_program, NULL);
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
+
+	/* Now create a dummy task just to estimate its duration according to the regression */
+
+	size = 1234567;
+
+	starpu_vector_data_register(&handle, -1, (uintptr_t)NULL, size, sizeof(int));
+
+	struct starpu_task *task = starpu_task_create();
+	task->cl = &memset_cl;
+	task->handles[0] = handle;
+	task->destroy = 0;
+
+	//FPRINTF(stdout, "\n ////linear regression results////\n");
+	//compare_performance(size, &memset_cl, task);
+
+	task->cl = &nl_memset_cl;
+
+	FPRINTF(stdout, "\n ////non linear regression results////\n");
+
+	compare_performance(size, &nl_memset_cl, task);
+
+	starpu_task_destroy(task);
+
+	starpu_data_unregister(handle);
+
+#ifdef STARPU_USE_OPENCL
+	ret = starpu_opencl_unload_opencl(&opencl_program);
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
+#endif
+	starpu_shutdown();
+
+	return EXIT_SUCCESS;
+}