Dangling

8 files changed, 779 insertions, 35 deletions

diff --git a/examples/moving_poiseulle_particles_2d_fplbm_gpu/.nix/derivation.nix b/examples/moving_poiseulle_particles_2d_fplbm_gpu/.nix/derivation.nix
new file mode 100644
index 0000000..672927c
--- /dev/null
+++ b/examples/moving_poiseulle_particles_2d_fplbm_gpu/.nix/derivation.nix
@@ -0,0 +1,41 @@
+{ lib
+, stdenv
+, scons
+, clang-tools
+, forstio
+, python3
+, pname
+, version
+, adaptive-cpp
+, kel
+}:
+
+stdenv.mkDerivation {
+	pname = pname + "-examples-" + "moving_poiseulle_particles_2d_fplbm_gpu";
+	inherit version;
+	src = ./..;
+
+	nativeBuildInputs = [
+		scons
+		clang-tools
+		python3
+	];
+
+	buildInputs = [
+		forstio.core
+		forstio.async
+		forstio.codec
+		forstio.codec-unit
+		forstio.io
+    forstio.remote
+    forstio.remote-filesystem
+		forstio.codec-json
+		adaptive-cpp
+		kel.lbm.core
+		kel.lbm.sycl
+	];
+
+	preferLocalBuild = true;
+
+	outputs = [ "out" "dev" ];
+}
diff --git a/examples/moving_poiseulle_particles_2d_fplbm_gpu/SConscript b/examples/moving_poiseulle_particles_2d_fplbm_gpu/SConscript
new file mode 100644
index 0000000..5e19236
--- /dev/null
+++ b/examples/moving_poiseulle_particles_2d_fplbm_gpu/SConscript
@@ -0,0 +1,34 @@
+#!/bin/false
+
+import os
+import os.path
+import glob
+
+
+Import('env')
+
+dir_path = Dir('.').abspath
+
+# Environment for base library
+examples_env = env.Clone();
+examples_env['CXX'] = 'syclcc-clang';
+examples_env['CXXFLAGS'] += ['-O3'];
+
+examples_env.sources = sorted(glob.glob(dir_path + "/*.cpp"))
+examples_env.headers = sorted(glob.glob(dir_path + "/*.hpp"))
+
+env.sources += examples_env.sources;
+env.headers += examples_env.headers;
+
+# Cavity2D
+examples_objects = [];
+examples_env.add_source_files(examples_objects, ['sim.cpp'], shared=False);
+examples_env.poiseulle_2d_gpu = examples_env.Program('#bin/moving_poiseulle_particles_2d_fplbm_gpu', [examples_objects]);
+
+# Set Alias
+env.examples = [
+	examples_env.poiseulle_2d_gpu
+];
+env.Alias('examples', env.examples);
+env.targets += ['examples'];
+env.Install('$prefix/bin/', env.examples);
diff --git a/examples/moving_poiseulle_particles_2d_fplbm_gpu/SConstruct b/examples/moving_poiseulle_particles_2d_fplbm_gpu/SConstruct
new file mode 100644
index 0000000..0611b67
--- /dev/null
+++ b/examples/moving_poiseulle_particles_2d_fplbm_gpu/SConstruct
@@ -0,0 +1,81 @@
+#!/usr/bin/env python3
+
+import sys
+import os
+import os.path
+import glob
+import re
+
+
+if sys.version_info < (3,):
+    def isbasestring(s):
+        return isinstance(s,basestring)
+else:
+    def isbasestring(s):
+        return isinstance(s, (str,bytes))
+
+def add_kel_source_files(self, sources, filetype, lib_env=None, shared=False, target_post=""):
+
+    if isbasestring(filetype):
+        dir_path = self.Dir('.').abspath
+        filetype = sorted(glob.glob(dir_path+"/"+filetype))
+
+    for path in filetype:
+        target_name = re.sub( r'(.*?)(\.cpp|\.c\+\+)', r'\1' + target_post, path )
+        if shared:
+            target_name+='.os'
+            sources.append( self.SharedObject( target=target_name, source=path ) )
+        else:
+            target_name+='.o'
+            sources.append( self.StaticObject( target=target_name, source=path ) )
+    pass
+
+def isAbsolutePath(key, dirname, env):
+	assert os.path.isabs(dirname), "%r must have absolute path syntax" % (key,)
+
+env_vars = Variables(
+	args=ARGUMENTS
+)
+
+env_vars.Add('prefix',
+	help='Installation target location of build results and headers',
+	default='/usr/local/',
+	validator=isAbsolutePath
+)
+
+env_vars.Add('build_examples',
+	help='If examples should be built',
+	default="true"
+)
+
+env=Environment(ENV=os.environ, variables=env_vars, CPPPATH=[],
+    CPPDEFINES=['SAW_UNIX'],
+    CXXFLAGS=[
+        '-std=c++20',
+        '-g',
+        '-Wall',
+        '-Wextra'
+    ],
+    LIBS=[
+        'forstio-core',
+        'forstio-async',
+        'forstio-io'
+    ]
+);
+env.__class__.add_source_files = add_kel_source_files
+env.Tool('compilation_db');
+env.cdb = env.CompilationDatabase('compile_commands.json');
+
+env.objects = [];
+env.sources = [];
+env.headers = [];
+env.targets = [];
+
+Export('env')
+SConscript('SConscript')
+
+env.Alias('cdb', env.cdb);
+env.Alias('all', [env.targets]);
+env.Default('all');
+
+env.Alias('install', '$prefix')
diff --git a/examples/moving_poiseulle_particles_2d_fplbm_gpu/sim.cpp b/examples/moving_poiseulle_particles_2d_fplbm_gpu/sim.cpp
new file mode 100644
index 0000000..3164945
--- /dev/null
+++ b/examples/moving_poiseulle_particles_2d_fplbm_gpu/sim.cpp
@@ -0,0 +1,472 @@
+#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_y = 256ul;
+constexpr uint64_t dim_x = dim_y * 20ul;
+
+constexpr uint64_t particle_amount = 1ul;
+
+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">,
+	Member<VectorChunk<T,Desc>, "particle_N">,
+	Member<ScalarChunk<T,Desc>, "particle_D">
+>;
+
+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<ScalarChunk<T,Desc>, "porosity">
+>;
+
+//template<typename T, typename Desc>
+//using ParticleArray = Array<
+//	Particle<T,Desc::D>
+//>;
+
+template<typename T, typename Desc>
+using ParticleSpheroidGroup = ParticleGroup<T,Desc::D,1u,coll::Spheroid<T>>;
+}
+
+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::ParticleSpheroidGroup<T,Desc>>& particles
+){
+	auto& info_f = fields.template get<"info">();
+	auto& porous_f = macros.template get<"porosity">();
+	// 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}}
+	);
+	
+	// Inflow
+	iterator<Desc::D>::apply(
+		[&](auto& index){
+			info_f.at(index).set(3u);
+		},
+		{{0u,0u}},
+		{{1u,dim_y}},
+		{{0u,1u}}
+	);
+	
+	// Outflow
+	iterator<Desc::D>::apply(
+		[&](auto& index){
+			info_f.at(index).set(4u);
+		},
+		{{dim_x-1u,0u}},
+		{{dim_x, dim_y}},
+		{{0u,1u}}
+	);
+	//
+	auto& df_f = fields.template get<"dfs_old">();
+	auto& rho_f = macros.template get<"density">();
+	auto& vel_f = macros.template get<"velocity">();
+	auto& por_f = macros.template get<"porosity">();
+	
+	iterator<Desc::D>::apply(
+		[&](auto& index){
+			auto& df = df_f.at(index);
+			auto& rho = rho_f.at(index);
+			por_f.at(index).at({}) = {1};
+			rho.at({}) = {1};
+			auto& vel = vel_f.at(index);
+			auto eq = equilibrium<T,Desc>(rho,vel);
+
+			df = eq;
+		},
+		{},// 0-index
+		df_f.get_dims()
+	);
+
+	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);
+			if(info_f.at(index).get() == 2u){
+				vel.at({{0u}}) = 0.0;
+			}
+			auto eq = equilibrium<T,Desc>(rho,vel);
+
+			df = eq;
+		},
+		{},// 0-index
+		df_f.get_dims(),
+		{{1u,1u}}
+	);
+
+	saw::data<sch::Scalar<T>> rad;
+	rad.at({}) = dim_y * 0.1f;
+	saw::data<sch::Scalar<T>> dense;
+	dense.at({}) = 1.0f;
+	particles = create_spheroid_particle_group<T,Desc::D,particle_amount>(rad,dense,{16u});
+	{
+		auto& rb = particles.template get<"particles">().at({0u}).template get<"rigid_body">();
+		auto& rbp = rb.template get<"position">();
+		rbp.at({{0u}}) = 0.25 * dim_x;
+		rbp.at({{1u}}) = 0.5 * dim_y;
+		rb.template get<"position_old">() = rbp;
+		saw::data<sch::Vector<T,Desc::D>> zero;
+		zero.at({{0u}}) = 0.0f;
+		zero.at({{1u}}) = 0.0f;
+		rb.template get<"acceleration">() = zero;
+		rb.template get<"rotation">() = {};
+		rb.template get<"rotation_old">() = {};
+		rb.template get<"angular_acceleration">() = {};
+	}
+
+	return saw::make_void();
+}
+
+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::Ptr<sch::ParticleSpheroidGroup<T,Desc>>,encode::Sycl<saw::encode::Native>>& particles,
+		saw::data<sch::UInt64> t_i,
+		device& dev
+){
+	auto& q = dev.get_handle();
+	auto& info_f = fields.template get<"info">();
+	auto& porous_f = macros.template get<"porosity">();
+
+	q.submit([&](acpp::sycl::handler& h){
+		component<T,Desc,cmpt::FplbmReset,encode::Sycl<saw::encode::Native>> hlbm_reset;
+		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]);
+			}
+
+			hlbm_reset.apply(fields,macros,index,t_i);
+		});
+	}).wait();
+
+	q.submit([&](acpp::sycl::handler& h){
+		component<T,Desc,cmpt::FplbmOneParticle,encode::Sycl<saw::encode::Native>> hlbm_one_part;
+
+		h.parallel_for(acpp::sycl::range<1u>{particle_amount}, [=](acpp::sycl::id<1u> idx){
+			saw::data<sch::FixedArray<sch::UInt64,1u>> index;
+			for(uint64_t i = 0u; i < 1u; ++i){
+				index.at({{i}}).set(idx[i]);
+			}
+
+			hlbm_one_part.apply(fields,macros,particles,index,t_i,{16u});
+		});
+	}).wait();
+
+	// auto coll_ev = 
+	q.submit([&](acpp::sycl::handler& h){
+		component<T,Desc,cmpt::Fplbm,encode::Sycl<saw::encode::Native>> collision{0.8};
+		component<T,Desc,cmpt::BounceBack,encode::Sycl<saw::encode::Native>> bb;
+
+		component<T,Desc,cmpt::ZouHeHorizontal<true>,encode::Sycl<saw::encode::Native>> flow_in{
+			[&](){
+				uint64_t target_t_i = 8u;
+				if(t_i.get() < target_t_i){
+					return 1.0 + (0.01 / target_t_i) * t_i.get();
+				}
+				return 1.01;
+			}()
+		};
+		component<T,Desc,cmpt::ZouHeHorizontal<false>,encode::Sycl<saw::encode::Native>> flow_out{1.0};
+
+
+		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;
+				case 3u:
+					flow_in.apply(fields,index,t_i);
+					collision.apply(fields,macros,index,t_i);
+					break;
+				case 4u:
+					flow_out.apply(fields,index,t_i);
+					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_dir = output_directory();
+	if(eo_lbm_dir.is_error()){
+		return std::move(eo_lbm_dir.get_error());
+	}
+	auto& lbm_dir = eo_lbm_dir.get_value();
+
+	auto out_dir = lbm_dir / "moving_poiseulle_particles_2d_hlbm_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,{0.1},{1e-4},{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_parts_data_ptr = saw::heap<saw::data<sch::ParticleSpheroidGroup<T,Desc>>>();
+	//saw::data<sch::Ptr<sch::ParticleSpheroidGroup<T,Desc>>,encode::Sycl<saw::encode::Native>>& particles
+	
+	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_parts_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::ParticleSpheroidGroup<T,Desc>, encode::Sycl<saw::encode::Native>> lbm_sycl_parts_data{sycl_q};
+	sycl_q.wait();
+
+	{
+		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_parts_data_ptr,lbm_sycl_parts_data);
+		if(eov.is_error()){
+			return eov;
+		}
+	}
+	sycl_q.wait();
+	auto lsd_view = make_view(lbm_sycl_data);
+	auto lsdm_view = make_view(lbm_sycl_macro_data);
+	auto lsdp_view = make_view(lbm_sycl_parts_data);
+
+	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,lsdp_view,i,dev);
+			if(eov.is_error()){
+				return eov;
+			}
+		}
+		sycl_q.wait();
+
+		if( i.get() % 32u == 0u){
+			{
+				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_parts_data,*lbm_parts_data_ptr);
+			if(eov.is_error()){
+				return eov;
+			}
+			auto& p = lbm_parts_data_ptr->template get<"particles">().at({0u}).template get<"rigid_body">();
+			auto& ppos = p.template get<"position">();
+			auto& pacc = p.template get<"acceleration">();
+			std::cout<<"Particle: "<<ppos.at({{0u}}).get()<<", "<<ppos.at({{1u}}).get()<<"; "<<pacc.at({{0u}}).get()<<", "<<pacc.at({{1u}}).get()<<";"<<i.get()<<std::endl;
+		}
+		// 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;
+		}
+	}
+	{
+		auto eov = write_csv_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::Float64;
+
+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;
+}
diff --git a/examples/poiseulle_particles_2d_gpu/init.hpp b/examples/poiseulle_particles_2d_gpu/init.hpp
index 617b296..cdce25e 100644
--- a/examples/poiseulle_particles_2d_gpu/init.hpp
+++ b/examples/poiseulle_particles_2d_gpu/init.hpp
@@ -5,8 +5,8 @@
 namespace kel {
 namespace lbm {
 
-template<typename T, typename Desc>
-saw::error_or<void> setup_initial_conditions(
+template<typename T, typename Desc, typename Coll>
+saw::error_or<void> init(
 		const converter<T>& conv,
 		saw::data<sch::ChunkStruct<T,Desc>>& fields,
 		saw::data<sch::MacroStruct<T,Desc>>& macros,
@@ -117,7 +117,7 @@ saw::error_or<void> setup_initial_conditions(
 		radius_p.at({}).set(2);
 		saw::data<sch::Scalar<T>> dense_p;
 		dense_p.at({}).set(1);
-		particles = create_spheroid_particle_group<T,Desc::D>(radius_p, dense_p, {{16u}});
+		particles = create_spheroid_particle_group<T,Desc::D,1u>(radius_p, dense_p, {{16u}});
 	}
 
 	return saw::make_void();
diff --git a/examples/poiseulle_particles_2d_gpu/sim.cpp b/examples/poiseulle_particles_2d_gpu/sim.cpp
index a4f08f0..fd6cdca 100644
--- a/examples/poiseulle_particles_2d_gpu/sim.cpp
+++ b/examples/poiseulle_particles_2d_gpu/sim.cpp
@@ -81,7 +81,7 @@ saw::error_or<void> lbm_main(const saw::data<args::LbmArgs>& args){
 
 	sycl_q.wait();
 	{
-		auto eov = setup_initial_conditions<T,Desc>(conv,*lbm_data_ptr,*lbm_macro_data_ptr,*lbm_particle_data_ptr);
+		auto eov = init<T,Desc,Coll>(conv,*lbm_data_ptr,*lbm_macro_data_ptr,*lbm_particle_data_ptr);
 		if(eov.is_error()){
 			return eov;
 		}
@@ -115,21 +115,21 @@ saw::error_or<void> lbm_main(const saw::data<args::LbmArgs>& args){
 	auto lsdm_view = make_view(lbm_sycl_macro_data);
 	auto lsdp_view = make_view(lbm_sycl_particle_data);
 	
-	saw::data<sch::UInt64> time_steps{16u*4096ul};
+	saw::data<sch::UInt64> time_steps{64u*1024ul};
 
 	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>(conv,lsd_view,lsdm_view,lsdp_view,i,dev);
+			auto eov = step<T,Desc,Coll>(conv,lsd_view,lsdm_view,lsdp_view,i,dev);
 			if(eov.is_error()){
 				return eov;
 			}
 		}
 
 		sycl_q.wait();
-		if(i.get() % 32u == 0u){
+		if(i.get() % 64u == 0u){
 			{
 				auto eov = dev.copy_to_host(lbm_sycl_macro_data,*lbm_macro_data_ptr);
 				if(eov.is_error()){
diff --git a/examples/poiseulle_particles_2d_gpu/step.hpp b/examples/poiseulle_particles_2d_gpu/step.hpp
index c1b148f..37eb4e7 100644
--- a/examples/poiseulle_particles_2d_gpu/step.hpp
+++ b/examples/poiseulle_particles_2d_gpu/step.hpp
@@ -5,7 +5,7 @@
 namespace kel {
 namespace lbm {
 
-template<typename T, typename Desc>
+template<typename T, typename Desc, typename Coll>
 saw::error_or<void> step(
 		const converter<T>& conv,
 		saw::data<sch::Ptr<sch::ChunkStruct<T,Desc>>,encode::Sycl<saw::encode::Native>>& fields,
@@ -17,28 +17,44 @@ saw::error_or<void> step(
 	auto& q = dev.get_handle();
 	auto& info_f = fields.template get<"info">();
 	auto& porous_f = macros.template get<"porosity">();
+	
+	q.submit([&](acpp::sycl::handler& h){
+		component<T,Desc,cmpt::HlbmReset,encode::Sycl<saw::encode::Native>> hlbm_reset;
+		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]);
+			}
+
+			hlbm_reset.apply(fields,macros,index,t_i);
+		});
+	}).wait();
+
+	q.submit([&](acpp::sycl::handler& h){
+		component<T,Desc,cmpt::HlbmOneParticle,encode::Sycl<saw::encode::Native>> hlbm_one_part;
+
+		h.parallel_for(acpp::sycl::range<1u>{particle_amount}, [=](acpp::sycl::id<1u> idx){
+			saw::data<sch::FixedArray<sch::UInt64,1u>> index;
+			for(uint64_t i = 0u; i < 1u; ++i){
+				index.at({{i}}).set(idx[i]);
+			}
 
-	// component<T,Desc,cmpt::HlbmOneParticle,encode::Sycl<saw::encode::Native>> particle;
+			hlbm_one_part.apply(fields,macros,particles,index,t_i,{16u});
+		});
+	}).wait();
 
 	// auto coll_ev = 
 	q.submit([&](acpp::sycl::handler& h){
 		component<T,Desc,cmpt::Hlbm,encode::Sycl<saw::encode::Native>> collision{0.8};
 		component<T,Desc,cmpt::BounceBack,encode::Sycl<saw::encode::Native>> bb;
 
-		saw::data<sch::Scalar<T>> rho_b;
-		rho_b.at({}) = 1.0;
-		saw::data<sch::Vector<T,Desc::D>> vel_b;
-		vel_b.at({{0u}}) = 0.015;
-
-		component<T,Desc,cmpt::Equilibrium,encode::Sycl<saw::encode::Native>> equi{rho_b,vel_b};
-
 		component<T,Desc,cmpt::ZouHeHorizontal<true>,encode::Sycl<saw::encode::Native>> flow_in{
 			[&](){
 				uint64_t target_t_i = 16u;
 				if(t_i.get() < target_t_i){
-					return 1.0 + (0.0002 / target_t_i) * t_i.get();
+					return 1.0 + (0.01 / target_t_i) * t_i.get();
 				}
-				return 1.0002;
+				return 1.01;
 			}()
 		};
 		component<T,Desc,cmpt::ZouHeHorizontal<false>,encode::Sycl<saw::encode::Native>> flow_out{1.0};
@@ -62,12 +78,10 @@ saw::error_or<void> step(
 					break;
 				case 3u:
 					flow_in.apply(fields,index,t_i);
-					// equi.apply(fields,index,t_i);
 					collision.apply(fields,macros,index,t_i);
 					break;
 				case 4u:
 					flow_out.apply(fields,index,t_i);
-					// equi.apply(fields,index,t_i);
 					collision.apply(fields,macros,index,t_i);
 				break;
 				default:
@@ -76,21 +90,6 @@ saw::error_or<void> step(
 		});
 	}).wait();
 
-	/*
-	q.submit([&](acpp::sycl::handler& h){
-		h.parallel_for(acpp::sycl::range<1u>{1u}, [=](acpp::sycl::id<1u> idx){
-			particle.apply(fields,macros,particles,{{0u}},t_i);
-		});
-	}).wait();
-	*/
-	
-	// Step
-	/*
-	q.submit([&](acpp::sycl::handler& h){
-		// h.depends_on(collision_ev);
-	}).wait();
-	*/
-
 	return saw::make_void();
 }
 
diff --git a/modules/core/c++/fplbm.hpp b/modules/core/c++/fplbm.hpp
index c974472..34806aa 100644
--- a/modules/core/c++/fplbm.hpp
+++ b/modules/core/c++/fplbm.hpp
@@ -5,9 +5,31 @@
 namespace kel {
 namespace lbm {
 namespace cmpt {
+struct FpLbmReset{};
 struct FpLbm {};
+struct FpLbmOneParticle {};
 struct FpLbmOneParticleNoVelocity{};
 }
+namespace method {
+struct FpLbm {};
+}
+
+template<typename T, typename Descriptor, typename Encode>
+class component<T, Descriptor, cmpt::FpLbmReset, Encode> final {
+public:
+	using Component = cmpt::FpLbmReset;
+private:
+public:
+	component() = default;
+
+	template<typename CellFieldSchema, typename MacroFieldSchema>
+	void apply(const saw::data<CellFieldSchema, Encode>& field, const saw::data<MacroFieldSchema,Encode>& macros, saw::data<sch::FixedArray<sch::UInt64,Descriptor::D>> index, saw::data<sch::UInt64> time_step) const {
+		auto& por_f = macros.template get<"porosity">();
+
+		auto& por = por_f.at(index);
+		por.at({}) = 1.0;
+	}
+};
 
 template<typename T, typename Descriptor, typename Encode>
 class component<T, Descriptor, cmpt::FpLbm, Encode> final {
@@ -85,6 +107,101 @@ public:
 };
 
 template<typename T, typename Descriptor, typename Encode>
+class component<T, Descriptor, cmpt::FpLbmOneParticle, Encode> final {
+public:
+	using Component = cmpt::FpLbmOneParticle;
+private:
+public:
+	component() = default;
+
+	template<typename CellFieldSchema, typename MacroFieldSchema>
+	void apply(const saw::data<CellFieldSchema, Encode>& field, const saw::data<MacroFieldSchema,Encode>& macros, saw::data<sch::FixedArray<sch::UInt64,Descriptor::D>> index, saw::data<sch::UInt64> time_step) const {
+	// void apply(saw::data<CellFieldSchema, Encode>& field, saw::data<sch::FixedArray<sch::UInt64, Descriptor::D>> index, saw::data<sch::UInt64> time_step){
+		bool is_even = ((time_step.get() % 2) == 0);
+
+		auto& dfs_old_f = (is_even) ? field.template get<"dfs_old">() : field.template get<"dfs">();
+		auto& dfs = dfs_old_f.at(index);
+
+		auto& rho_f = macros.template get<"density">();
+		auto& vel_f = macros.template get<"velocity">();
+		auto& por_f = macros.template get<"porosity">();
+
+		// Temporary find a better way
+		auto& force_f = macros.template get<"force">();
+
+
+		/**
+		 * The other methods all work with a flipped porosity.
+		 * For compat reasons this is also flipped
+		 */
+		auto porosity = por_f.at(index);
+		saw::data<sch::Scalar<T>> one;
+		one.at({}) = 1.0;
+		auto flip_porosity = one - porosity;
+
+		saw::data<sch::Scalar<T>>& rho = rho_f.at(index);
+
+		saw::data<sch::Scalar<T>> half;
+		half.at({}).set(0.5);
+		saw::data<sch::Vector<T,Descriptor::D>>& vel = vel_f.at(index);// + total_force * ( half / rho );
+
+		compute_rho_u<T,Descriptor>(dfs_old_f.at(index),rho,vel);
+		auto eq = equilibrium<T,Descriptor>(rho,vel);
+
+		using dfi = df_info<T,Descriptor>;
+
+		saw::data<sch::Scalar<T>> min_two;
+		min_two.at({}).set(-2);
+
+		// Maybe ? 
+		auto& force = force_f.at(index);
+		force = vel * rho * min_two * flip_porosity;
+
+		saw::data<sch::Scalar<T>> dfi_inv_cs2;
+		dfi_inv_cs2.at({}).set(dfi::inv_cs2);
+
+		// auto vel = vel_f.at(index);
+
+		for(uint64_t i = 0u; i < Descriptor::Q; ++i){
+			// saw::data<T> ci_min_u{0};
+			saw::data<sch::Vector<T,Descriptor::D>> ci;
+			for(uint64_t d = 0u; d < Descriptor::D; ++d){
+				ci.at({{d}}).set(static_cast<typename saw::native_data_type<T>::type>(dfi::directions[i][d]));
+			}
+			auto ci_dot_u = saw::math::dot(ci,vel);
+
+			// saw::data<sch::Vector<T,Descriptor::D>> F_i;
+			// F_i = f * ((c_i - u) * ics2  + <c_i,u> * c_i * ics2 * ics2) * w_i;
+			saw::data<sch::Scalar<T>> w;
+			w.at({}).set(dfi::weights[i]);
+
+			/*
+			saw::data<sch::Scalar<T>> F_i_sum;
+			for(uint64_t d = 0u; d < Descriptor::D; ++d){
+				saw::data<sch::Scalar<T>> F_i_d;
+				F_i_d.at({}) = F_i.at({{d}});
+				F_i_sum = F_i_sum + F_i_d;
+			}
+			*/
+			auto term1 = (ci-vel) * dfi_inv_cs2;
+			auto term2 = ci * (ci_dot_u * dfi_inv_cs2 * dfi_inv_cs2);
+
+			auto force_projection = saw::math::dot(term1 + term2, force);
+
+			auto F_i = w * force_projection;
+
+			dfs.at({i}) = dfs.at({i})
+									+ frequency_
+										* (eq.at(i) - dfs.at({i}) )
+									+ F_i.at({})
+										* (saw::data<T>{1} - saw::data<T>{0.5f} * frequency_);
+		}
+	}
+
+
+};
+
+template<typename T, typename Descriptor, typename Encode>
 class component<T, Descriptor, cmpt::FpLbmOneParticleNoVelocity, Encode> final {
 private:
 	saw::data<T> relaxation_;