Add benckmark tool. (fd2787f9) · Commits · Cianciosa, Mark / graph_framework

graph_benchmark/xrays_bench.cpp

+16 −22

Original line number	Diff line number	Diff line
		@@ -12,27 +12,25 @@
		template<typename T, size_t NUM_TIMES, size_t SUB_STEPS, size_t NUM_RAYS>
		void bench_runner() {
		if constexpr (std::is_same<T, float>::value) {
		std::cout << "Float ";
		std::cout << "Float --------------------------------------------------------------------------" << std::endl;
		} else if constexpr (std::is_same<T, double>::value) {
		std::cout << "Double ";
		std::cout << "Double -------------------------------------------------------------------------" << std::endl;
		} else if constexpr (std::is_same<T, std::complex<float>>::value) {
		std::cout << "Complex Float ";
		std::cout << "Complex Float ------------------------------------------------------------------" << std::endl;
		} else {
		std::cout << "Complex Double ";
		std::cout << "Complex Double -----------------------------------------------------------------" << std::endl;
		}

		std::cout << "--------------------------------------------------------------------------------" << std::endl;

		const size_t num_steps = NUM_TIMES/SUB_STEPS;

		std::vector<std::thread> threads(std::max(std::min(static_cast<unsigned int> (jit::context<T, false>::max_concurrency()),
		static_cast<unsigned int> (NUM_RAYS)),
		static_cast<unsigned int> (1)));

		std::vector<timeing::measure_diagnostic> timers;
		timeing::measure_diagnostic_threaded timing;

		for (size_t i = 0, ie = threads.size(); i < ie; i++) {
		threads[i] = std::thread([&timers] (const size_t thread_number,
		threads[i] = std::thread([&timing] (const size_t thread_number,
		const size_t num_threads) -> void {
		const size_t local_num_rays = NUM_RAYS/num_threads
		+ std::min(thread_number, NUM_RAYS%num_threads);
		@@ -73,28 +71,24 @@ void bench_runner() {
		solve.init(kx);
		solve.compile();

		std::ostringstream step_label;
		step_label << "Step Time thread " << thread_number;
		timers.emplace_back(step_label.str());
		timing.start_time(thread_number);
		for (size_t j = 0; j < num_steps; j++) {
		for (size_t k = 0; k < SUB_STEPS; k++) {
		solve.step();
		}
		}
		solve.sync_host();
		timers[thread_number].stop();
		timing.end_time(thread_number);
		}, i, threads.size());
		}

		for (std::thread &t : threads) {
		t.join();
		}
		timing.print();

		for (auto &time : timers) {
		time.print();
		}

		std::cout << "--------------------------------------------------------------------------------" << std::endl;
		std::cout << "--------------------------------------------------------------------------------"
		<< std::endl << std::endl;
		}

		//------------------------------------------------------------------------------
		@@ -106,10 +100,10 @@ void bench_runner() {
		int main(int argc, const char * argv[]) {
		START_GPU

		bench_runner<float, 100000, 10, 100000> ();
		bench_runner<double, 100000, 10, 100000> ();
		bench_runner<std::complex<float>, 100000, 10, 100000> ();
		bench_runner<std::complex<double>, 100000, 10, 100000> ();
		bench_runner<float, 1000, 10, 100000> ();
		bench_runner<double, 1000, 10, 100000> ();
		bench_runner<std::complex<float>, 1000, 10, 100000> ();
		bench_runner<std::complex<double>, 1000, 10, 100000> ();

		END_GPU
		}

graph_driver/xrays.cpp

+4 −2

Original line number	Diff line number	Diff line
		@@ -13,6 +13,7 @@
		const bool print = false;
		const bool write_step = true;
		const bool print_expressions = false;
		const bool verbose = true;

		//------------------------------------------------------------------------------
		/// @brief Main program of the driver.
		@@ -23,6 +24,8 @@ const bool print_expressions = false;
		int main(int argc, const char * argv[]) {
		START_GPU

		jit::verbose = verbose;

		typedef float base;
		//typedef double base;
		//typedef std::complex<float> base;
		@@ -30,7 +33,7 @@ int main(int argc, const char * argv[]) {
		//constexpr bool use_safe_math = true;
		constexpr bool use_safe_math = false;

		timeing::measure_diagnostic total("Total Time");
		const timeing::measure_diagnostic total("Total Time");

		const size_t num_times = 100000;
		const size_t sub_steps = 10;
		@@ -182,7 +185,6 @@ int main(int argc, const char * argv[]) {
		}

		std::cout << std::endl << "Timing:" << std::endl;
		total.stop();
		total.print();

		END_GPU

graph_framework.xcodeproj/project.pbxproj

+283 −2

Original line number	Diff line number	Diff line
		@@ -32,6 +32,7 @@
		C73BBE7E29F816E60027BB7F /* piecewise_test.cpp in Sources / = {isa = PBXBuildFile; fileRef = C73BBE7D29F816E60027BB7F / piecewise_test.cpp */; };
		C73BBE8229F820810027BB7F /* Metal.framework in Frameworks / = {isa = PBXBuildFile; fileRef = C71342682947F36100672AD4 / Metal.framework */; };
		C73E2A7A2A4A216400BED03A /* Metal.framework in Frameworks / = {isa = PBXBuildFile; fileRef = C71342682947F36100672AD4 / Metal.framework */; };
		C74DF4602AA8BD1900319113 /* Metal.framework in Frameworks / = {isa = PBXBuildFile; fileRef = C71342682947F36100672AD4 / Metal.framework */; };
		C79141B622DAAD0C00E0BA0D /* xrays.cpp in Sources / = {isa = PBXBuildFile; fileRef = C79141B522DAAD0C00E0BA0D / xrays.cpp */; };
		C7D371132A0595A40074676E /* Metal.framework in Frameworks / = {isa = PBXBuildFile; fileRef = C71342682947F36100672AD4 / Metal.framework */; };
		C7E5644528A2A1AA000F31A2 /* backend_test.cpp in Sources / = {isa = PBXBuildFile; fileRef = C7931E7328074F540033B488 / backend_test.cpp */; };
		@@ -46,6 +47,107 @@
		C7FA0E0B29590F9F00A31E4D /* Metal.framework in Frameworks / = {isa = PBXBuildFile; fileRef = C71342682947F36100672AD4 / Metal.framework */; };
		/* End PBXBuildFile section */

		/* Begin PBXContainerItemProxy section */
		C74DF4612AA8BD2600319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4632AA8BD3400319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4652AA8BD3900319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4672AA8BD3E00319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4692AA8BD4200319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF46B2AA8BD4600319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF46D2AA8BD4B00319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF46F2AA8BD4F00319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4712AA8BD5300319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4732AA8BD5700319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4752AA8BD5B00319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4772AA8BD5F00319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF4792AA8BD6200319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		C74DF47B2AA8BD6600319113 /* PBXContainerItemProxy */ = {
		isa = PBXContainerItemProxy;
		containerPortal = C791419E22DA9BF200E0BA0D /* Project object */;
		proxyType = 1;
		remoteGlobalIDString = C79141A522DA9BF200E0BA0D;
		remoteInfo = graph_framework;
		};
		/* End PBXContainerItemProxy section */

		/* Begin PBXCopyFilesBuildPhase section */
		C736902F2A38C498001733B0 /* CopyFiles */ = {
		isa = PBXCopyFilesBuildPhase;
		@@ -74,6 +176,15 @@
		);
		runOnlyForDeploymentPostprocessing = 1;
		};
		C74DF4552AA8BC7300319113 /* CopyFiles */ = {
		isa = PBXCopyFilesBuildPhase;
		buildActionMask = 2147483647;
		dstPath = /usr/share/man/man1/;
		dstSubfolderSpec = 0;
		files = (
		);
		runOnlyForDeploymentPostprocessing = 1;
		};
		C79141B122DAAD0C00E0BA0D /* CopyFiles */ = {
		isa = PBXCopyFilesBuildPhase;
		buildActionMask = 2147483647;
		@@ -188,7 +299,9 @@
		C73BBE7B29F816E50027BB7F /* piecewise_test */ = {isa = PBXFileReference; explicitFileType = "compiled.mach-o.executable"; includeInIndex = 0; path = piecewise_test; sourceTree = BUILT_PRODUCTS_DIR; };
		C73BBE7D29F816E60027BB7F /* piecewise_test.cpp */ = {isa = PBXFileReference; explicitFileType = sourcecode.cpp.objcpp; path = piecewise_test.cpp; sourceTree = "<group>"; };
		C73BBE9629F8669F0027BB7F /* newton.hpp */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.h; path = newton.hpp; sourceTree = "<group>"; };
		C7616FBC2AA7944400A831DE /* graph_benchmark */ = {isa = PBXFileReference; lastKnownFileType = folder; path = graph_benchmark; sourceTree = "<group>"; };
		C74DF4572AA8BC7300319113 /* graph_benchmark */ = {isa = PBXFileReference; explicitFileType = "compiled.mach-o.executable"; includeInIndex = 0; path = graph_benchmark; sourceTree = BUILT_PRODUCTS_DIR; };
		C74DF4592AA8BC7300319113 /* main.cpp */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.cpp; path = main.cpp; sourceTree = "<group>"; };
		C74DF45F2AA8BCE600319113 /* graph_benchmark */ = {isa = PBXFileReference; lastKnownFileType = folder; path = graph_benchmark; sourceTree = "<group>"; };
		C77E6DF522DD64E700469621 /* trigonometry.hpp */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.h; path = trigonometry.hpp; sourceTree = "<group>"; };
		C79141A622DA9BF200E0BA0D /* libgraph_framework.a */ = {isa = PBXFileReference; explicitFileType = archive.ar; includeInIndex = 0; path = libgraph_framework.a; sourceTree = BUILT_PRODUCTS_DIR; };
		C79141AE22DA9C3000E0BA0D /* node.hpp */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.h; path = node.hpp; sourceTree = "<group>"; };
		@@ -246,6 +359,14 @@
		);
		runOnlyForDeploymentPostprocessing = 0;
		};
		C74DF4542AA8BC7300319113 /* Frameworks */ = {
		isa = PBXFrameworksBuildPhase;
		buildActionMask = 2147483647;
		files = (
		C74DF4602AA8BD1900319113 /* Metal.framework in Frameworks */,
		);
		runOnlyForDeploymentPostprocessing = 0;
		};
		C79141A422DA9BF200E0BA0D /* Frameworks */ = {
		isa = PBXFrameworksBuildPhase;
		buildActionMask = 2147483647;
		@@ -346,6 +467,14 @@
		path = cmake;
		sourceTree = "<group>";
		};
		C74DF4582AA8BC7300319113 /* graph_benchmark */ = {
		isa = PBXGroup;
		children = (
		C74DF4592AA8BC7300319113 /* main.cpp */,
		);
		path = graph_benchmark;
		sourceTree = "<group>";
		};
		C791419D22DA9BF200E0BA0D = {
		isa = PBXGroup;
		children = (
		@@ -353,8 +482,9 @@
		C79141AD22DA9C0600E0BA0D /* graph_framework */,
		C7931E6928073BCA0033B488 /* graph_tests */,
		C79141B422DAAD0C00E0BA0D /* graph_driver */,
		C7616FBC2AA7944400A831DE /* graph_benchmark */,
		C74DF45F2AA8BCE600319113 /* graph_benchmark */,
		C717CB8C2A02E361008FBDD8 /* cmake */,
		C74DF4582AA8BC7300319113 /* graph_benchmark */,
		C79141A722DA9BF200E0BA0D /* Products */,
		C71342672947F36100672AD4 /* Frameworks */,
		);
		@@ -377,6 +507,7 @@
		C73BBE6F29F72BC50027BB7F /* trigonometry_test */,
		C73BBE7B29F816E50027BB7F /* piecewise_test */,
		C73690312A38C498001733B0 /* erfi_test */,
		C74DF4572AA8BC7300319113 /* graph_benchmark */,
		);
		name = Products;
		sourceTree = "<group>";
		@@ -481,6 +612,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF47C2AA8BD6600319113 /* PBXTargetDependency */,
		);
		name = erfi_test;
		productName = erfi_test;
		@@ -498,6 +630,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF4782AA8BD5F00319113 /* PBXTargetDependency */,
		);
		name = trigonometry_test;
		productName = trigonometry_test;
		@@ -515,12 +648,31 @@
		buildRules = (
		);
		dependencies = (
		C74DF47A2AA8BD6200319113 /* PBXTargetDependency */,
		);
		name = piecewise_test;
		productName = piecewise_test;
		productReference = C73BBE7B29F816E50027BB7F /* piecewise_test */;
		productType = "com.apple.product-type.tool";
		};
		C74DF4562AA8BC7300319113 /* graph_benchmark */ = {
		isa = PBXNativeTarget;
		buildConfigurationList = C74DF45D2AA8BC7300319113 /* Build configuration list for PBXNativeTarget "graph_benchmark" */;
		buildPhases = (
		C74DF4532AA8BC7300319113 /* Sources */,
		C74DF4542AA8BC7300319113 /* Frameworks */,
		C74DF4552AA8BC7300319113 /* CopyFiles */,
		);
		buildRules = (
		);
		dependencies = (
		C74DF4622AA8BD2600319113 /* PBXTargetDependency */,
		);
		name = graph_benchmark;
		productName = graph_benchmark;
		productReference = C74DF4572AA8BC7300319113 /* graph_benchmark */;
		productType = "com.apple.product-type.tool";
		};
		C79141A522DA9BF200E0BA0D /* graph_framework */ = {
		isa = PBXNativeTarget;
		buildConfigurationList = C79141AA22DA9BF200E0BA0D /* Build configuration list for PBXNativeTarget "graph_framework" */;
		@@ -549,6 +701,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF4642AA8BD3400319113 /* PBXTargetDependency */,
		);
		name = graph_driver;
		productName = graph_driver;
		@@ -566,6 +719,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF4662AA8BD3900319113 /* PBXTargetDependency */,
		);
		name = backend_test;
		productName = graph_tests;
		@@ -583,6 +737,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF4682AA8BD3E00319113 /* PBXTargetDependency */,
		);
		name = dispersion_test;
		productName = graph_tests;
		@@ -600,6 +755,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF46A2AA8BD4200319113 /* PBXTargetDependency */,
		);
		name = solver_test;
		productName = graph_tests;
		@@ -617,6 +773,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF46C2AA8BD4600319113 /* PBXTargetDependency */,
		);
		name = arithmetic_test;
		productName = graph_tests;
		@@ -634,6 +791,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF46E2AA8BD4B00319113 /* PBXTargetDependency */,
		);
		name = math_test;
		productName = graph_tests;
		@@ -651,6 +809,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF4702AA8BD4F00319113 /* PBXTargetDependency */,
		);
		name = node_test;
		productName = graph_tests;
		@@ -668,6 +827,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF4722AA8BD5300319113 /* PBXTargetDependency */,
		);
		name = vector_test;
		productName = graph_tests;
		@@ -685,6 +845,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF4742AA8BD5700319113 /* PBXTargetDependency */,
		);
		name = physics_test;
		productName = graph_tests;
		@@ -702,6 +863,7 @@
		buildRules = (
		);
		dependencies = (
		C74DF4762AA8BD5B00319113 /* PBXTargetDependency */,
		);
		name = jit_test;
		productName = jit_test;
		@@ -727,6 +889,9 @@
		C73BBE7A29F816E50027BB7F = {
		CreatedOnToolsVersion = 14.3;
		};
		C74DF4562AA8BC7300319113 = {
		CreatedOnToolsVersion = 14.3.1;
		};
		C79141A522DA9BF200E0BA0D = {
		CreatedOnToolsVersion = 10.2.1;
		};
		@@ -789,6 +954,7 @@
		C73BBE6E29F72BC50027BB7F /* trigonometry_test */,
		C73BBE7A29F816E50027BB7F /* piecewise_test */,
		C73690302A38C498001733B0 /* erfi_test */,
		C74DF4562AA8BC7300319113 /* graph_benchmark */,
		);
		};
		/* End PBXProject section */
		@@ -818,6 +984,13 @@
		);
		runOnlyForDeploymentPostprocessing = 0;
		};
		C74DF4532AA8BC7300319113 /* Sources */ = {
		isa = PBXSourcesBuildPhase;
		buildActionMask = 2147483647;
		files = (
		);
		runOnlyForDeploymentPostprocessing = 0;
		};
		C79141A322DA9BF200E0BA0D /* Sources */ = {
		isa = PBXSourcesBuildPhase;
		buildActionMask = 2147483647;
		@@ -907,6 +1080,79 @@
		};
		/* End PBXSourcesBuildPhase section */

		/* Begin PBXTargetDependency section */
		C74DF4622AA8BD2600319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4612AA8BD2600319113 /* PBXContainerItemProxy */;
		};
		C74DF4642AA8BD3400319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4632AA8BD3400319113 /* PBXContainerItemProxy */;
		};
		C74DF4662AA8BD3900319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4652AA8BD3900319113 /* PBXContainerItemProxy */;
		};
		C74DF4682AA8BD3E00319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4672AA8BD3E00319113 /* PBXContainerItemProxy */;
		};
		C74DF46A2AA8BD4200319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4692AA8BD4200319113 /* PBXContainerItemProxy */;
		};
		C74DF46C2AA8BD4600319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF46B2AA8BD4600319113 /* PBXContainerItemProxy */;
		};
		C74DF46E2AA8BD4B00319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF46D2AA8BD4B00319113 /* PBXContainerItemProxy */;
		};
		C74DF4702AA8BD4F00319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF46F2AA8BD4F00319113 /* PBXContainerItemProxy */;
		};
		C74DF4722AA8BD5300319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4712AA8BD5300319113 /* PBXContainerItemProxy */;
		};
		C74DF4742AA8BD5700319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4732AA8BD5700319113 /* PBXContainerItemProxy */;
		};
		C74DF4762AA8BD5B00319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4752AA8BD5B00319113 /* PBXContainerItemProxy */;
		};
		C74DF4782AA8BD5F00319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4772AA8BD5F00319113 /* PBXContainerItemProxy */;
		};
		C74DF47A2AA8BD6200319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF4792AA8BD6200319113 /* PBXContainerItemProxy */;
		};
		C74DF47C2AA8BD6600319113 /* PBXTargetDependency */ = {
		isa = PBXTargetDependency;
		target = C79141A522DA9BF200E0BA0D /* graph_framework */;
		targetProxy = C74DF47B2AA8BD6600319113 /* PBXContainerItemProxy */;
		};
		/* End PBXTargetDependency section */

		/* Begin XCBuildConfiguration section */
		C73690362A38C498001733B0 /* Debug */ = {
		isa = XCBuildConfiguration;
		@@ -1004,6 +1250,32 @@
		};
		name = Release;
		};
		C74DF45B2AA8BC7300319113 /* Debug */ = {
		isa = XCBuildConfiguration;
		buildSettings = {
		CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
		"CODE_SIGN_IDENTITY[sdk=macosx*]" = "-";
		CODE_SIGN_STYLE = Automatic;
		GCC_PREPROCESSOR_DEFINITIONS = (
		"DEBUG=1",
		"$(inherited)",
		);
		MACOSX_DEPLOYMENT_TARGET = 13.3;
		PRODUCT_NAME = "$(TARGET_NAME)";
		};
		name = Debug;
		};
		C74DF45C2AA8BC7300319113 /* Release */ = {
		isa = XCBuildConfiguration;
		buildSettings = {
		CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
		"CODE_SIGN_IDENTITY[sdk=macosx*]" = "-";
		CODE_SIGN_STYLE = Automatic;
		MACOSX_DEPLOYMENT_TARGET = 13.3;
		PRODUCT_NAME = "$(TARGET_NAME)";
		};
		name = Release;
		};
		C79141A822DA9BF200E0BA0D /* Debug */ = {
		isa = XCBuildConfiguration;
		buildSettings = {
		@@ -1492,6 +1764,15 @@
		defaultConfigurationIsVisible = 0;
		defaultConfigurationName = Release;
		};
		C74DF45D2AA8BC7300319113 /* Build configuration list for PBXNativeTarget "graph_benchmark" */ = {
		isa = XCConfigurationList;
		buildConfigurations = (
		C74DF45B2AA8BC7300319113 /* Debug */,
		C74DF45C2AA8BC7300319113 /* Release */,
		);
		defaultConfigurationIsVisible = 0;
		defaultConfigurationName = Release;
		};
		C79141A122DA9BF200E0BA0D /* Build configuration list for PBXProject "graph_framework" */ = {
		isa = XCConfigurationList;
		buildConfigurations = (

graph_framework/cpu_context.hpp

+18 −5

Original line number	Diff line number	Diff line
		@@ -43,6 +43,13 @@ namespace gpu {
		return std::thread::hardware_concurrency();
		}

		//------------------------------------------------------------------------------
		/// @brief Device discription.
		//------------------------------------------------------------------------------
		static std::string device_type() {
		return "CPU";
		}

		//------------------------------------------------------------------------------
		/// @brief Construct a cpu context.
		///
		@@ -106,8 +113,10 @@ namespace gpu {
		#endif
		temp_stream << filename << " -o " << library_name;

		if (jit::verbose) {
		std::cout << "CPU info." << std::endl;
		std::cout << " Command Line : " << temp_stream.str() << std::endl;
		}
		int error = system(temp_stream.str().c_str());
		if (error) {
		std::cerr << "Failed to compile cpu kernel. Check source code in "
		@@ -130,9 +139,11 @@ namespace gpu {
		exit(1);
		}

		if (jit::verbose) {
		std::cout << " Library name : " << library_name << std::endl;
		std::cout << " Library handle : " << reinterpret_cast<size_t> (lib_handle) << std::endl;
		}
		}

		//------------------------------------------------------------------------------
		/// @brief Create a kernel calling function.
		@@ -173,7 +184,9 @@ namespace gpu {
		buffers[jit::to_string('o', output.get())] = kernel_arguments[output.get()].data();
		}

		if (jit::verbose) {
		std::cout << " Function pointer: " << reinterpret_cast<size_t> (kernel) << std::endl;
		}

		return [kernel, buffers] () mutable {
		((void ()(std::map<std::string, T > &))kernel)(buffers);

graph_framework/cuda_context.hpp

+29 −10

File changed.

Preview size limit exceeded, changes collapsed.