add -perTaskThreads and benchmark-based counts

EidosScribe/EidosAppDelegate.mm

@@ -72,7 +72,7 @@ - (void)applicationWillFinishLaunching:(NSNotification *)aNotification
 	// We always use 4 threads; we don't want to hog the whole machine, just run with a couple threads.
 	// We pass false for active_threads to let the worker threads sleep, otherwise the CPU is pegged
 	// the whole time EidosScribe is running, even when sitting idle.
-	Eidos_WarmUpOpenMP(&std::cout, true, 4, false);
+	Eidos_WarmUpOpenMP(&std::cout, true, 4, false, /* default per-task thread counts */ "");
 #endif
 
 	Eidos_WarmUp();

SLiMgui/AppDelegate.mm

@@ -224,7 +224,7 @@ - (void)applicationWillFinishLaunching:(NSNotification *)aNotification
 	// We always use 4 threads; we don't want to hog the whole machine, just run with a couple threads.
 	// We pass false for active_threads to let the worker threads sleep, otherwise the CPU is pegged
 	// the whole time SLiMgui is running, even when sitting idle.
-	Eidos_WarmUpOpenMP(&std::cout, true, 4, false);
+	Eidos_WarmUpOpenMP(&std::cout, true, 4, false, /* default per-task thread counts */ "");
 #endif
 
 	Eidos_WarmUp();

VERSIONS

@@ -30,7 +30,7 @@ PARALLEL changes (now in the master branch):
 	modify CMakeLists.txt to add /usr/local/include to all targets (through GSL_INCLUDES, which should maybe get cleaned up somehow)
 	modify CMakeLists.txt to build executables eidos_multi and slim_multi when PARALLEL is ON, and to link in libomp
 	disable parallel builds of EidosScribe, SLiMgui, and SLiMguiLegacy with #error directives; parallelization is for the command line only (active threads peg the CPU, passive threads are slower than single-threaded)
-	add a command-line option, -maxthreads <n>, to control the maximum number of threads used in OpenMP
+	add a command-line option, -maxThreads <n>, to control the maximum number of threads used in OpenMP
 	add a header eidos_openmp.h that should be used instead of #include "omp.h", providing various useful definitions etc.
 	add an initialization function for OpenMP, Eidos_WarmUpOpenMP(), that must be called when warming up; sets various options, logs diagnostic info
 	set default values for OMP_WAIT_POLICY (active) and OMP_PROC_BIND (false) and OMP_DYNAMIC (false) for slim and eidos
@@ -84,6 +84,7 @@ PARALLEL changes (now in the master branch):
 	add parallelSetTaskThreadCounts() function to set per-task thread counts, and parallelGetTaskThreadCounts() to get them
 	add recipe 22.7, Parallelizing nonWF reproduction and spatial interactions
 	add internal EidosBenchmark facility for timing of internal loops, for benchmarking of parallelization scaling
+	add a command-line option, -perTaskThreads, to let the user control how many threads to use for different tasks (if not overridden)
 
 development head (in the master branch):
 	fix a minor bug with autofix when opening multiple .slim documents at once in SLiMgui

core/main.cpp

@@ -104,8 +104,8 @@ static void PrintUsageAndDie(bool p_print_header, bool p_print_full_usage)
 	SLIM_OUTSTREAM << "   [-l[ong] [<l>]] [-s[eed] <seed>] [-t[ime]] [-m[em]] [-M[emhist]] [-x]" << std::endl;
 	SLIM_OUTSTREAM << "   [-d[efine] <def>] ";
 #ifdef _OPENMP
-	// The -maxthreads flag is visible only for a parallel build
-	SLIM_OUTSTREAM << "[-maxthreads <n>] ";
+	// Some flags are visible only for a parallel build
+	SLIM_OUTSTREAM << "[-maxThreads <n>] [-perTaskThreads \"x\"] ";
 #endif
 #if (SLIMPROFILING == 1)
 	// Some flags are visible only for a profile build
@@ -130,8 +130,9 @@ static void PrintUsageAndDie(bool p_print_header, bool p_print_full_usage)
 		SLIM_OUTSTREAM << "   -x                 : disable SLiM's runtime safety/consistency checks" << std::endl;
 		SLIM_OUTSTREAM << "   -d[efine] <def>    : define an Eidos constant, such as \"mu=1e-7\"" << std::endl;
 #ifdef _OPENMP
-		// The -maxthreads flag is visible only for a parallel build
-		SLIM_OUTSTREAM << "   -maxthreads <n>    : set the maximum number of threads used" << std::endl;
+		// Some flags are visible only for a parallel build
+		SLIM_OUTSTREAM << "   -maxThreads <n>    : set the maximum number of threads used" << std::endl;
+		SLIM_OUTSTREAM << "   -perTaskThreads \"x\": set per-task thread counts to named set \"x\"" << std::endl;
 #endif
 #if (SLIMPROFILING == 1)
 		SLIM_OUTSTREAM << "   " << std::endl;
@@ -187,6 +188,7 @@ int main(int argc, char *argv[])
 #ifdef _OPENMP
 	long max_thread_count = omp_get_max_threads();
 	bool changed_max_thread_count = false;
+	std::string per_task_thread_count_set_name = "";		// default per-task thread counts
 #endif
 
 #if (SLIMPROFILING == 1)
@@ -335,7 +337,7 @@ int main(int argc, char *argv[])
 		if (strcmp(arg, "--testEidos") == 0 || strcmp(arg, "-testEidos") == 0 || strcmp(arg, "-te") == 0)
 		{
 #ifdef _OPENMP
-			Eidos_WarmUpOpenMP(&SLIM_ERRSTREAM, changed_max_thread_count, (int)max_thread_count, true);
+			Eidos_WarmUpOpenMP(&SLIM_ERRSTREAM, changed_max_thread_count, (int)max_thread_count, true, /* max per-task thread counts */ "maxThreads");
 #endif
 			Eidos_WarmUp();
 
@@ -351,7 +353,7 @@ int main(int argc, char *argv[])
 		if (strcmp(arg, "--testSLiM") == 0 || strcmp(arg, "-testSLiM") == 0 || strcmp(arg, "-ts") == 0)
 		{
 #ifdef _OPENMP
-			Eidos_WarmUpOpenMP(&SLIM_ERRSTREAM, changed_max_thread_count, (int)max_thread_count, true);
+			Eidos_WarmUpOpenMP(&SLIM_ERRSTREAM, changed_max_thread_count, (int)max_thread_count, true, /* max per-task thread counts */ "maxThreads");
 #endif
 			Eidos_WarmUp();
 			SLiM_WarmUp();
@@ -383,8 +385,8 @@ int main(int argc, char *argv[])
 			continue;
 		}
 
-		// -maxthreads <x>: set the maximum number of OpenMP threads that will be used
-		if (strcmp(arg, "-maxthreads") == 0)
+		// -maxThreads <x>: set the maximum number of OpenMP threads that will be used
+		if (strcmp(arg, "-maxThreads") == 0)
 		{
 			if (++arg_index == argc)
 				PrintUsageAndDie(false, true);
@@ -397,20 +399,33 @@ int main(int argc, char *argv[])
 
 			if ((max_thread_count < 1) || (max_thread_count > EIDOS_OMP_MAX_THREADS))
 			{
-				SLIM_OUTSTREAM << "The -maxthreads command-line option enforces a range of [1, " << EIDOS_OMP_MAX_THREADS << "]." << std::endl;
+				SLIM_OUTSTREAM << "The -maxThreads command-line option enforces a range of [1, " << EIDOS_OMP_MAX_THREADS << "]." << std::endl;
 				exit(EXIT_FAILURE);
 			}
 
 			continue;
 #else
 			if (count != 1)
 			{
-				SLIM_OUTSTREAM << "The -maxthreads command-line option only allows a value of 1 when not running a PARALLEL build." << std::endl;
+				SLIM_OUTSTREAM << "The -maxThreads command-line option only allows a value of 1 when not running a PARALLEL build." << std::endl;
 				exit(EXIT_FAILURE);
 			}
 #endif
 		}
 
+		// -perTaskThreads "x": set the per-task thread counts to be used in OpenMP to a named set "x"
+		if (strcmp(arg, "-perTaskThreads") == 0)
+		{
+			if (++arg_index == argc)
+				PrintUsageAndDie(false, true);
+
+#ifdef _OPENMP
+			// We just take the name as given; testing against known values will be done later
+			// This command-line argument is ignored completely when not parallel
+			per_task_thread_count_set_name = std::string(argv[arg_index]);
+#endif
+		}
+
 #if (SLIMPROFILING == 1)
 		if (strcmp(arg, "-profileStart") == 0)
 		{
@@ -500,7 +515,7 @@ int main(int argc, char *argv[])
 #endif
 
 #ifdef _OPENMP
-	Eidos_WarmUpOpenMP((SLiM_verbosity_level >= 1) ? &SLIM_ERRSTREAM : nullptr, changed_max_thread_count, (int)max_thread_count, true);
+	Eidos_WarmUpOpenMP((SLiM_verbosity_level >= 1) ? &SLIM_ERRSTREAM : nullptr, changed_max_thread_count, (int)max_thread_count, true, per_task_thread_count_set_name);
 #endif
 
 	if (SLiM_verbosity_level >= 2)

eidos/eidos_functions_other.cpp

@@ -891,9 +891,9 @@ EidosValue_SP Eidos_ExecuteFunction_parallelSetTaskThreadCounts(__attribute__((u
 
 	if (source_value->Type() == EidosValueType::kValueNULL)
 	{
-		// A dict value of NULL means "reset to default settings", which we have a function for
+		// A dict value of NULL means "reset to the command-line default settings"
 #ifdef _OPENMP
-		_Eidos_SetDefaultOpenMPThreadCounts();
+		_Eidos_SetOpenMPThreadCounts(gEidosDefaultPerTaskThreadCounts);
 #endif
 	}
 	else
@@ -1048,6 +1048,10 @@ EidosValue_SP Eidos_ExecuteFunction_parallelSetTaskThreadCounts(__attribute__((u
 						else if (key == "SURVIVAL")						gEidos_OMP_threads_SURVIVAL = (int)value_int64;
 						else
 							EIDOS_TERMINATION << "ERROR (Eidos_ExecuteFunction_parallelSetTaskThreadCounts): parallelSetTaskThreadCounts() does not recognize the task name " << key << "." << EidosTerminate(nullptr);
+
+						// This assumes that any thread count set might push the maximum per-task thread count higher, but not lower
+						gEidosPerTaskThreadCountsSetName = "UserDefined";
+						gEidosPerTaskOriginalMaxThreadCount = std::max(gEidosPerTaskOriginalMaxThreadCount, (int)value_int64);
 #endif
 					}
 					else