Merge branch 'dev'

1 files changed, 375 insertions, 0 deletions

diff --git a/examples/settling_cubes_2d_ibm_gpu/sim.cpp b/examples/settling_cubes_2d_ibm_gpu/sim.cpp
new file mode 100644
index 0000000..05a4a09
--- /dev/null
+++ b/examples/settling_cubes_2d_ibm_gpu/sim.cpp
@@ -0,0 +1,375 @@
+#include <kel/lbm/sycl/lbm.hpp>
+#include <kel/lbm/lbm.hpp>
+#include <kel/lbm/particle.hpp>
+
+#include <forstio/io/io.hpp>
+#include <forstio/remote/filesystem/easy.hpp>
+#include <forstio/codec/json/json.hpp>
+#include <forstio/codec/simple.hpp>
+
+namespace kel {
+namespace lbm {
+
+constexpr uint64_t dim_x = 512ul;
+constexpr uint64_t dim_y = dim_x * 2ul;
+
+constexpr uint64_t particle_amount = 16ul;
+
+namespace sch {
+using namespace saw::schema;
+
+using InfoChunk = Chunk<UInt8, 0u, dim_x, dim_y>;
+
+template<typename T, typename Desc>
+using DfChunk = Chunk<FixedArray<T,Desc::Q>, 1u, dim_x, dim_y>;
+
+template<typename T, typename Desc>
+using ScalarChunk = Chunk<Scalar<T>, 0u, dim_x, dim_y>;
+
+template<typename T, typename Desc>
+using VectorChunk = Chunk<Vector<T,Desc::D>, 0u, dim_x, dim_y>;
+
+template<typename T, typename Desc>
+using ChunkStruct = Struct<
+	Member<InfoChunk, "info">,
+	Member<DfChunk<T,Desc>, "dfs">,
+	Member<DfChunk<T,Desc>, "dfs_old">
+>;
+
+template<typename T, typename Desc>
+using VelChunk = Chunk<Vector<T,Desc::D>, 0u, dim_x, dim_y>;
+
+template<typename T>
+using RhoChunk = Chunk<Scalar<T>, 0u, dim_x, dim_y>;
+
+template<typename T, typename Desc>
+using MacroStruct = Struct<
+	Member<VelChunk<T,Desc>, "velocity">,
+	Member<RhoChunk<T>, "density">,
+	Member<VectorChunk<T,Desc>, "force">
+>;
+
+//template<typename T, typename Desc>
+//using ParticleArray = Array<
+//	Particle<T,Desc::D>
+//>;
+}
+
+template<typename T, typename Desc>
+saw::error_or<void> setup_initial_conditions(
+		saw::data<sch::ChunkStruct<T,Desc>>& fields,
+		saw::data<sch::MacroStruct<T,Desc>>& macros,
+		saw::data<sch::FixedArray<sch::Particle<T,Desc::D>, particle_amount>>& particles
+){
+	auto& info_f = fields.template get<"info">();
+	// Set everything as walls
+	iterator<Desc::D>::apply(
+		[&](auto& index){
+			info_f.at(index).set(1u);
+		},
+		{},
+		info_f.get_dims(),
+		{}
+	);
+	// Fluid
+	iterator<Desc::D>::apply(
+		[&](auto& index){
+			info_f.at(index).set(2u);
+		},
+		{},
+		info_f.get_dims(),
+		{{1u,1u}}
+	);
+	//
+	auto& df_f = fields.template get<"dfs_old">();
+	auto& rho_f = macros.template get<"density">();
+	auto& vel_f = macros.template get<"velocity">();
+	
+	iterator<Desc::D>::apply(
+		[&](auto& index){
+			auto& df = df_f.at(index);
+			auto& rho = rho_f.at(index);
+			rho.at({}) = {1};
+			auto& vel = vel_f.at(index);
+			auto eq = equilibrium<T,Desc>(rho,vel);
+
+			df = eq;
+		},
+		{},// 0-index
+		df_f.get_dims()
+	);
+
+	return saw::make_void();
+}
+
+template<typename T, typename Desc>
+void add_particles(saw::data<sch::Array<sch::Particle<T,Desc::D>>>& particles){
+	////// TODO
+}
+
+template<typename T, typename Desc>
+saw::error_or<void> step(
+		saw::data<sch::Ptr<sch::ChunkStruct<T,Desc>>,encode::Sycl<saw::encode::Native>>& fields,
+		saw::data<sch::Ptr<sch::MacroStruct<T,Desc>>,encode::Sycl<saw::encode::Native>>& macros,
+		saw::data<sch::FixedArray<sch::Particle<T,Desc::D>, particle_amount>>& particles,
+		saw::data<sch::UInt64> t_i,
+		device& dev
+){
+	auto& q = dev.get_handle();
+	auto& info_f = fields.template get<"info">();
+
+	{
+		auto& p = particles.at({{0u}});
+
+		auto& p_coll = p.template get<"collision">();
+		auto& p_rad = p_coll.template get<"radius">();
+	}
+
+
+	// auto coll_ev = 
+	q.submit([&](acpp::sycl::handler& h){
+		// Need nicer things to handle the flow. I see improvement here
+		component<T,Desc,cmpt::BGKGuo, encode::Sycl<saw::encode::Native>> collision{0.8};
+		component<T,Desc,cmpt::BounceBack,encode::Sycl<saw::encode::Native>> bb;
+
+		h.parallel_for(acpp::sycl::range<Desc::D>{dim_x,dim_y}, [=](acpp::sycl::id<Desc::D> idx){
+			saw::data<sch::FixedArray<sch::UInt64,Desc::D>> index;
+			for(uint64_t i = 0u; i < Desc::D; ++i){
+				index.at({{i}}).set(idx[i]);
+			}
+
+			auto info = info_f.at(index);
+			
+			switch(info.get()){
+				case 0u:
+				break;
+				case 1u:
+					bb.apply(fields,index,t_i);
+				break;
+				case 2u:
+					collision.apply(fields,macros,index,t_i);
+					break;
+				default:
+				break;
+			}
+		});
+
+
+	}).wait();
+
+	
+	// Step
+	/*
+	q.submit([&](acpp::sycl::handler& h){
+		// h.depends_on(collision_ev);
+	}).wait();
+	*/
+
+	return saw::make_void();
+}
+}
+}
+
+template<typename T, typename Desc>
+saw::error_or<void> lbm_main(int argc, char** argv){
+	using namespace kel::lbm;
+
+	using dfi = df_info<T,Desc>;
+
+	auto eo_lbm_env = lbm_directory();
+	if(eo_lbm_env.is_error()){
+		return std::move(eo_lbm_env.get_error());
+	}
+	auto& lbm_env = eo_lbm_env.get_value();
+
+	auto out_dir = lbm_env.data_dir / "settling_cubes_2d_ibm_gpu";
+
+	{
+		std::error_code ec;
+		std::filesystem::create_directories(out_dir,ec);
+		if(ec != std::errc{}){
+			return saw::make_error<saw::err::critical>("Could not create output directory");
+		}
+	}
+
+	converter<T> conv {
+		// delta_x
+		{{1.0}},
+		// delta_t
+		{{1.0}}
+	};
+	
+	print_lbm_meta<T,Desc>(conv,{0.05},{0.01},{0.4 * dim_y});
+
+	// saw::data<sch::FixedArray<sch::UInt64,Desc::D>> meta{{dim_x,dim_y}};
+	auto lbm_data_ptr = saw::heap<saw::data<sch::ChunkStruct<T,Desc>>>();
+	auto lbm_macro_data_ptr = saw::heap<saw::data<sch::MacroStruct<T,Desc>>>();
+	auto lbm_particle_data_ptr = saw::heap<saw::data<sch::FixedArray<sch::Particle<T,Desc::D>, particle_amount>>>();
+	
+	std::cout<<"Estimated Bytes: "<<memory_estimate<sch::ChunkStruct<T,Desc>,sch::MacroStruct<T,Desc>>().get()<<std::endl;
+
+	auto eo_aio = saw::setup_async_io();
+	if(eo_aio.is_error()){
+		return std::move(eo_aio.get_error());
+	}
+	auto& aio = eo_aio.get_value();
+	saw::wait_scope wait{aio.event_loop};
+
+	bool krun = true;
+	bool print_status = false;
+	aio.event_port.on_signal(saw::Signal::Terminate).then([&](){
+		krun = false;
+	}).detach();
+	aio.event_port.on_signal(saw::Signal::User1).then([&](){
+		print_status = true;
+	}).detach();
+
+	device dev;
+
+	auto& sycl_q = dev.get_handle();
+
+	sycl_q.wait();
+	{
+		auto eov = setup_initial_conditions<T,Desc>(*lbm_data_ptr,*lbm_macro_data_ptr,*lbm_particle_data_ptr);
+		if(eov.is_error()){
+			return eov;
+		}
+	}
+	{
+		auto eov = write_vtk_file(out_dir,"initial_state",0u,*lbm_data_ptr);
+		if(eov.is_error()){
+			return eov;
+		}
+	}
+
+	saw::data<sch::ChunkStruct<T,Desc>, encode::Sycl<saw::encode::Native>> lbm_sycl_data{sycl_q};
+	saw::data<sch::MacroStruct<T,Desc>, encode::Sycl<saw::encode::Native>> lbm_sycl_macro_data{sycl_q};
+	saw::data<sch::FixedArray<sch::Particle<T,Desc::D>, particle_amount>, encode::Sycl<saw::encode::Native>> lbm_sycl_particle_data{sycl_q};
+	sycl_q.wait();
+
+	auto lsd_view = make_chunk_struct_view(lbm_sycl_data);
+	auto lsdm_view = make_chunk_struct_view(lbm_sycl_macro_data);
+	{
+		auto eov = dev.copy_to_device(*lbm_data_ptr,lbm_sycl_data);
+		if(eov.is_error()){
+			return eov;
+		}
+	}
+	{
+		auto eov = dev.copy_to_device(*lbm_macro_data_ptr,lbm_sycl_macro_data);
+		if(eov.is_error()){
+			return eov;
+		}
+	}
+	{
+		auto eov = dev.copy_to_device(*lbm_particle_data_ptr,lbm_sycl_particle_data);
+		if(eov.is_error()){
+			return eov;
+		}
+	}
+	sycl_q.wait();
+	saw::data<sch::UInt64> time_steps{16u*4096ul};
+
+	auto& info_f = lsd_view.template get<"info">();
+
+	for(saw::data<sch::UInt64> i{0u}; i < time_steps and krun; ++i){
+		// BC + Collision
+		{
+			auto eov = step<T,Desc>(lsd_view,lsdm_view,*lbm_particle_data_ptr,i,dev);
+			if(eov.is_error()){
+				return eov;
+			}
+		}
+		sycl_q.wait();
+		/*
+		{
+			{
+				auto eov = dev.copy_to_host(lbm_sycl_macro_data,*lbm_macro_data_ptr);
+				if(eov.is_error()){
+					return eov;
+				}
+			}
+			{
+				auto eov = write_vtk_file(out_dir,"m",i.get(), *lbm_macro_data_ptr);
+				if(eov.is_error()){
+					return eov;
+				}
+			}
+		}
+		*/
+		/*
+		{
+			{
+				auto eov = dev.copy_to_host(lbm_sycl_data,*lbm_data_ptr);
+				if(eov.is_error()){
+					return eov;
+				}
+			}
+			{
+				auto eov = write_csv_file(out_dir,"lbm",i.get(), *lbm_data_ptr);
+				if(eov.is_error()){
+					return eov;
+				}
+			}
+		}
+		*/
+		// Stream
+		sycl_q.submit([&](acpp::sycl::handler& h){
+			component<T,Desc,cmpt::Stream,encode::Sycl<saw::encode::Native>> stream;
+
+			h.parallel_for(acpp::sycl::range<Desc::D>{dim_x,dim_y}, [=](acpp::sycl::id<Desc::D> idx){
+				saw::data<sch::FixedArray<sch::UInt64,Desc::D>> index;
+				for(uint64_t i = 0u; i < Desc::D; ++i){
+					index.at({{i}}).set(idx[i]);
+				}
+				
+				auto info = info_f.at(index);
+				
+				if(info.get() > 0u){
+					stream.apply(lsd_view,index,i);
+				}
+			});
+		}).wait();
+		wait.poll();
+		if(print_status){
+			std::cout<<"Status: "<<i.get()<<" of "<<time_steps.get()<<" - "<<(i.template cast_to<sch::Float64>().get() * 100 / time_steps.get())<<"%"<<std::endl;
+			print_status = false;
+		}
+		print_progress_bar(i.get(), time_steps.get()-1u);
+	}
+
+	// After Loop
+	sycl_q.wait();
+	{
+		auto eov = dev.copy_to_host(lbm_sycl_macro_data,*lbm_macro_data_ptr);
+		if(eov.is_error()){
+			return eov;
+		}
+	}
+	{
+		auto eov = write_vtk_file(out_dir,"m",time_steps.get(), *lbm_macro_data_ptr);
+		if(eov.is_error()){
+			return eov;
+		}
+	}
+
+	sycl_q.wait();
+	return saw::make_void();
+}
+
+using FloatT = kel::lbm::sch::Float32;
+
+int main(int argc, char** argv){
+	auto eov = lbm_main<FloatT,kel::lbm::sch::D2Q9>(argc, argv);
+	if(eov.is_error()){
+		auto& err = eov.get_error();
+		std::cerr<<"[Error] "<<err.get_category();
+		auto err_msg = err.get_message();
+		if(err_msg.size() > 0u){
+			std::cerr<<" - "<<err_msg;
+		}
+		std::cerr<<std::endl;
+		return err.get_id();
+	}
+	return 0;
+}