From 25c85bf962e0646f8e03f67fd4982450f41ee6a6 Mon Sep 17 00:00:00 2001 From: "Claudius \"keldu\" Holeksa" Date: Fri, 6 Mar 2026 21:05:36 +0100 Subject: Work finished for this week --- .../.nix/derivation.nix | 41 ++ examples/poiseulle_particles_2d_bgk_gpu/SConscript | 34 ++ examples/poiseulle_particles_2d_bgk_gpu/SConstruct | 81 ++++ examples/poiseulle_particles_2d_bgk_gpu/sim.cpp | 455 +++++++++++++++++++++ 4 files changed, 611 insertions(+) create mode 100644 examples/poiseulle_particles_2d_bgk_gpu/.nix/derivation.nix create mode 100644 examples/poiseulle_particles_2d_bgk_gpu/SConscript create mode 100644 examples/poiseulle_particles_2d_bgk_gpu/SConstruct create mode 100644 examples/poiseulle_particles_2d_bgk_gpu/sim.cpp (limited to 'examples/poiseulle_particles_2d_bgk_gpu') diff --git a/examples/poiseulle_particles_2d_bgk_gpu/.nix/derivation.nix b/examples/poiseulle_particles_2d_bgk_gpu/.nix/derivation.nix new file mode 100644 index 0000000..517b6b8 --- /dev/null +++ b/examples/poiseulle_particles_2d_bgk_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-" + "poiseulle_2d_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/poiseulle_particles_2d_bgk_gpu/SConscript b/examples/poiseulle_particles_2d_bgk_gpu/SConscript new file mode 100644 index 0000000..fb22d18 --- /dev/null +++ b/examples/poiseulle_particles_2d_bgk_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/poiseulle_particles_2d_bgk_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/poiseulle_particles_2d_bgk_gpu/SConstruct b/examples/poiseulle_particles_2d_bgk_gpu/SConstruct new file mode 100644 index 0000000..0611b67 --- /dev/null +++ b/examples/poiseulle_particles_2d_bgk_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/poiseulle_particles_2d_bgk_gpu/sim.cpp b/examples/poiseulle_particles_2d_bgk_gpu/sim.cpp new file mode 100644 index 0000000..f5d3f72 --- /dev/null +++ b/examples/poiseulle_particles_2d_bgk_gpu/sim.cpp @@ -0,0 +1,455 @@ +#include +#include +#include + +#include +#include +#include +#include + +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; + +template +using DfChunk = Chunk, 1u, dim_x, dim_y>; + +template +using ScalarChunk = Chunk, 0u, dim_x, dim_y>; + +template +using VectorChunk = Chunk, 0u, dim_x, dim_y>; + +template +using ChunkStruct = Struct< + Member, + Member, "dfs">, + Member, "dfs_old"> +>; + +template +using VelChunk = Chunk, 0u, dim_x, dim_y>; + +template +using RhoChunk = Chunk, 0u, dim_x, dim_y>; + +template +using MacroStruct = Struct< + Member, "velocity">, + Member, "density"> +>; + +//template +//using ParticleArray = Array< +// Particle +//>; +} + +template +saw::error_or setup_initial_conditions( + saw::data>& fields, + saw::data>& macros, + saw::data, particle_amount>>& particles +){ + auto& info_f = fields.template get<"info">(); + // Set everything as walls + iterator::apply( + [&](auto& index){ + info_f.at(index).set(1u); + }, + {}, + info_f.get_dims(), + {} + ); + // Fluid + iterator::apply( + [&](auto& index){ + info_f.at(index).set(2u); + }, + {}, + info_f.get_dims(), + {{1u,1u}} + ); + + // Inflow + iterator::apply( + [&](auto& index){ + info_f.at(index).set(3u); + }, + {{0u,0u}}, + {{1u,dim_y}}, + {{0u,1u}} + ); + + // Outflow + iterator::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">(); + + iterator::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(rho,vel); + + df = eq; + }, + {},// 0-index + df_f.get_dims() + ); + + iterator::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(rho,vel); + + df = eq; + }, + {},// 0-index + df_f.get_dims(), + {{1u,1u}} + ); + + iterator::apply( + [&](auto& index){ + saw::data> middle, ind_vec; + middle.at({{0u}}) = dim_x * 0.25; + middle.at({{1u}}) = dim_y * 0.5; + + ind_vec.at({{0u}}) = index.at({{0u}}).template cast_to(); + ind_vec.at({{1u}}) = index.at({{1u}}).template cast_to(); + + auto dist = middle - ind_vec; + auto dist_2 = saw::math::dot(dist,dist); + if(dist_2.at({}).get() < dim_y*dim_y*0.01){ + info_f.at(index) = 1u; + } + }, + {},// 0-index + df_f.get_dims() + ); + + return saw::make_void(); +} + +template +saw::error_or step( + saw::data>,encode::Sycl>& fields, + saw::data>,encode::Sycl>& macros, + saw::data, particle_amount>>& particles, + saw::data 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> collision{0.65}; + // component> collision{0.65}; + // component> collision{0.65}; + component> bb; + + saw::data> rho_b; + rho_b.at({}) = 1.0; + saw::data> vel_b; + vel_b.at({{0u}}) = 0.015; + + component> equi{rho_b,vel_b}; + + component,encode::Sycl> flow_in{ + [&](){ + uint64_t target_t_i = 64u; + if(t_i.get() < target_t_i){ + return 1.0 + (0.0015 / target_t_i) * t_i.get(); + } + return 1.0015; + }() + }; + component,encode::Sycl> flow_out{1.0}; + + + h.parallel_for(acpp::sycl::range{dim_x,dim_y}, [=](acpp::sycl::id idx){ + saw::data> 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); + 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: + break; + } + }); + }).wait(); + + + // Step + /* + q.submit([&](acpp::sycl::handler& h){ + // h.depends_on(collision_ev); + }).wait(); + */ + + return saw::make_void(); +} +} +} + +template +saw::error_or lbm_main(int argc, char** argv){ + using namespace kel::lbm; + + using dfi = df_info; + + 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 / "poiseulle_particles_2d_bgk_gpu"; + + { + std::error_code ec; + std::filesystem::create_directories(out_dir,ec); + if(ec != std::errc{}){ + return saw::make_error("Could not create output directory"); + } + } + + converter conv { + // delta_x + {{1.0}}, + // delta_t + {{1.0}} + }; + + print_lbm_meta(conv,{0.05},{0.01},{0.4 * dim_y}); + + // saw::data> meta{{dim_x,dim_y}}; + auto lbm_data_ptr = saw::heap>>(); + auto lbm_macro_data_ptr = saw::heap>>(); + auto lbm_particle_data_ptr = saw::heap, particle_amount>>>(); + + std::cout<<"Estimated Bytes: "<,sch::MacroStruct>().get()<(*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, encode::Sycl> lbm_sycl_data{sycl_q}; + saw::data, encode::Sycl> lbm_sycl_macro_data{sycl_q}; + saw::data, particle_amount>, encode::Sycl> 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 time_steps{16u*4096ul}; + + auto& info_f = lsd_view.template get<"info">(); + + for(saw::data i{0u}; i < time_steps and krun; ++i){ + // BC + Collision + { + auto eov = step(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> stream; + + h.parallel_for(acpp::sycl::range{dim_x,dim_y}, [=](acpp::sycl::id idx){ + saw::data> 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: "<().get() * 100 / time_steps.get())<<"%"<(argc, argv); + if(eov.is_error()){ + auto& err = eov.get_error(); + std::cerr<<"[Error] "< 0u){ + std::cerr<<" - "<