Rename folder, add basic README.md

5 files changed, 291 insertions, 0 deletions

diff --git a/lid_driven_cavity/cpp/CMakeLists.txt b/lid_driven_cavity/cpp/CMakeLists.txt
new file mode 100644
index 0000000..5b5fb90
--- /dev/null
+++ b/lid_driven_cavity/cpp/CMakeLists.txt
@@ -0,0 +1,44 @@
+cmake_minimum_required(VERSION 3.10)
+project(ldc LANGUAGES CXX)
+
+if(NOT CMAKE_BUILD_TYPE)
+  set(CMAKE_BUILD_TYPE Release)
+endif()
+
+set(CMAKE_CXX_FLAGS_RELEASE "-O3")
+
+add_custom_command(
+	OUTPUT
+		kernel.h
+		ldc.cpp
+	COMMAND
+		${CMAKE_CURRENT_SOURCE_DIR}/generate.py --output ${CMAKE_CURRENT_BINARY_DIR}
+	WORKING_DIRECTORY
+		${CMAKE_CURRENT_SOURCE_DIR}
+	DEPENDS
+		generate.py config.py ldc.cpp.mako
+)
+
+include_directories(
+	${CMAKE_BINARY_DIR}
+)
+
+add_executable(
+	ldc
+		${CMAKE_CURRENT_BINARY_DIR}/ldc.cpp
+)
+
+target_compile_features(
+	ldc
+	PUBLIC
+		cxx_std_17
+)
+
+find_package(OpenMP)
+if(OpenMP_CXX_FOUND)
+	target_link_libraries(
+		ldc
+		PUBLIC
+			OpenMP::OpenMP_CXX
+	)
+endif()
diff --git a/lid_driven_cavity/cpp/README.md b/lid_driven_cavity/cpp/README.md
new file mode 100644
index 0000000..8e84c83
--- /dev/null
+++ b/lid_driven_cavity/cpp/README.md
@@ -0,0 +1,18 @@
+# Lid driven cavity
+
+This example models the common lid driven cavity example.
+Note that the actual optimized C++ implementation is generated using the _boltzgen_ library.
+
+See `config.py` for various configuration options. Both 2D and 3D are supported.
+
+## Build instructions
+
+```
+mkdir build
+cd build
+cmake ..
+make
+./ldc
+```
+
+This should result in some summarizing CLI output in addition to a `test.vtk` file for visualization in Paraview.
diff --git a/lid_driven_cavity/cpp/config.py b/lid_driven_cavity/cpp/config.py
new file mode 100644
index 0000000..3ed8bd5
--- /dev/null
+++ b/lid_driven_cavity/cpp/config.py
@@ -0,0 +1,13 @@
+from boltzgen.lbm.model import *
+from boltzgen import Geometry
+
+descriptor = D2Q9
+geometry   = Geometry(512, 512)
+tau        = 0.52
+precision  = 'single'
+
+## 3D LDC
+#descriptor = D3Q19
+#geometry   = Geometry(64, 64, 64)
+#tau        = 0.52
+#precision  = 'single'
diff --git a/lid_driven_cavity/cpp/generate.py b/lid_driven_cavity/cpp/generate.py
new file mode 100755
index 0000000..4222e98
--- /dev/null
+++ b/lid_driven_cavity/cpp/generate.py
@@ -0,0 +1,40 @@
+#!/usr/bin/env python
+
+import argparse
+
+from boltzgen import LBM, Generator, Geometry
+from boltzgen.lbm.model import D2Q9
+
+import config
+
+argparser = argparse.ArgumentParser(
+    description = 'Generate a C++ implementation of a lid driven cavity simulation using LBM')
+argparser.add_argument(
+    '--output', required = False, help = 'Target directory for the generated sources')
+
+args = argparser.parse_args()
+
+lbm = LBM(config.descriptor)
+generator = Generator(
+    descriptor = config.descriptor,
+    moments    = lbm.moments(),
+    collision  = lbm.bgk(f_eq = lbm.equilibrium(), tau = config.tau),
+    target     = 'cpp',
+    precision  = config.precision,
+    index      = 'XYZ',
+    layout     = 'AOS')
+
+if args.output is None:
+    args.output = '.'
+
+functions = ['collide_and_stream', 'equilibrilize', 'collect_moments', 'momenta_boundary']
+
+with open('%s/kernel.h' % args.output, 'w') as kernel:
+    kernel.write(generator.kernel(config.geometry, functions))
+
+ldc_src = ''
+with open('ldc.cpp.mako', 'r') as template:
+    ldc_src = template.read()
+
+with open('%s/ldc.cpp' % args.output, 'w') as app:
+    app.write(generator.custom(config.geometry, ldc_src))
diff --git a/lid_driven_cavity/cpp/ldc.cpp.mako b/lid_driven_cavity/cpp/ldc.cpp.mako
new file mode 100644
index 0000000..5d480de
--- /dev/null
+++ b/lid_driven_cavity/cpp/ldc.cpp.mako
@@ -0,0 +1,176 @@
+#include <cstdint>
+#include <memory>
+#include <vector>
+#include <chrono>
+#include <iostream>
+#include <fstream>
+
+#include "kernel.h"
+
+void collect_moments_to_vtk(const std::string& path, ${float_type}* f) {
+    std::ofstream fout;
+    fout.open(path.c_str());
+
+    fout << "# vtk DataFile Version 3.0\n";
+    fout << "lbm_output\n";
+    fout << "ASCII\n";
+    fout << "DATASET RECTILINEAR_GRID\n";
+% if descriptor.d == 2:
+    fout << "DIMENSIONS " << ${geometry.size_x-2} << " " << ${geometry.size_y-2} << " 1" << "\n";
+% else:
+    fout << "DIMENSIONS " << ${geometry.size_x-2} << " " << ${geometry.size_y-2} << " " << ${geometry.size_z-2} << "\n";
+% endif
+
+    fout << "X_COORDINATES " << ${geometry.size_x-2} << " float\n";
+    for( std::size_t x = 1; x < ${geometry.size_x-1}; ++x ) {
+        fout << x << " ";
+    }
+
+    fout << "\nY_COORDINATES " << ${geometry.size_y-2} << " float\n";
+    for( std::size_t y = 1; y < ${geometry.size_y-1}; ++y ) {
+        fout << y << " ";
+    }
+
+% if descriptor.d == 2:
+    fout << "\nZ_COORDINATES " << 1 << " float\n";
+    fout << 0 << "\n";
+    fout << "POINT_DATA " << ${(geometry.size_x-2) * (geometry.size_y-2)} << "\n";
+% else:
+    fout << "\nZ_COORDINATES " << ${geometry.size_z-2} << " float\n";
+    for( std::size_t z = 1; z < ${geometry.size_z-1}; ++z ) {
+        fout << z << " ";
+    }
+    fout << "\nPOINT_DATA " << ${(geometry.size_x-2) * (geometry.size_y-2) * (geometry.size_z-2)} << "\n";
+% endif
+
+    ${float_type} rho;
+    ${float_type} u[${descriptor.d}];
+
+    fout << "VECTORS velocity float\n";
+% if descriptor.d == 2:
+    for ( std::size_t y = 1; y < ${geometry.size_y-1}; ++y ) {
+        for ( std::size_t x = 1; x < ${geometry.size_x-1}; ++x ) {
+            collect_moments(f, x*${geometry.size_y}+y, rho, u);
+            fout << u[0] << " " << u[1] << " 0\n";
+        }
+    }
+% else:
+    for ( std::size_t z = 1; z < ${geometry.size_z-1}; ++z ) {
+        for ( std::size_t y = 1; y < ${geometry.size_y-1}; ++y ) {
+            for ( std::size_t x = 1; x < ${geometry.size_x-1}; ++x ) {
+                collect_moments(f, x*${geometry.size_y*geometry.size_z}+y*${geometry.size_z}+z, rho, u);
+                fout << u[0] << " " << u[1] << " " << u[2] << "\n";
+            }
+        }
+    }
+% endif
+
+    fout << "SCALARS density float 1\n";
+    fout << "LOOKUP_TABLE default\n";
+% if descriptor.d == 2:
+    for ( std::size_t y = 1; y < ${geometry.size_y-1}; ++y ) {
+        for ( std::size_t x = 1; x < ${geometry.size_x-1}; ++x ) {
+            collect_moments(f, x*${geometry.size_y}+y, rho, u);
+            fout << rho << "\n";
+        }
+    }
+% else:
+    for ( std::size_t z = 1; z < ${geometry.size_z-1}; ++z ) {
+        for ( std::size_t y = 1; y < ${geometry.size_y-1}; ++y ) {
+            for ( std::size_t x = 1; x < ${geometry.size_x-1}; ++x ) {
+                collect_moments(f, x*${geometry.size_y*geometry.size_z}+y*${geometry.size_z}+z, rho, u);
+                fout << rho << "\n";
+            }
+        }
+    }
+% endif
+
+    fout.close();
+}
+
+void simulate(std::size_t nStep)
+{
+    auto f_a = std::make_unique<${float_type}[]>(${geometry.volume*descriptor.q});
+    auto f_b = std::make_unique<${float_type}[]>(${geometry.volume*descriptor.q});
+
+    ${float_type}* f_prev = f_a.get();
+    ${float_type}* f_next = f_b.get();
+
+    std::vector<std::size_t> bulk;
+    std::vector<std::size_t> lid_bc;
+    std::vector<std::size_t> box_bc;
+
+    for (int iX = 1; iX < ${geometry.size_x-1}; ++iX) {
+        for (int iY = 1; iY < ${geometry.size_y-1}; ++iY) {
+% if descriptor.d == 2:
+            const std::size_t iCell = iX*${geometry.size_y} + iY;
+            if (iY == ${geometry.size_y-2}) {
+                lid_bc.emplace_back(iCell);
+            } else if (iX == 1 || iX == ${geometry.size_x-2} || iY == 1) {
+                box_bc.emplace_back(iCell);
+            } else {
+                bulk.emplace_back(iCell);
+            }
+% elif descriptor.d == 3:
+            for (int iZ = 0; iZ < ${geometry.size_z}; ++iZ) {
+                const std::size_t iCell = iX*${geometry.size_y*geometry.size_z} + iY*${geometry.size_z} + iZ;
+                if (iZ == ${geometry.size_z-2}) {
+                    lid_bc.emplace_back(iCell);
+                } else if (iX == 1 || iX == ${geometry.size_x-2} || iY == 1 || iY == ${geometry.size_y-2} || iZ == 1) {
+                    box_bc.emplace_back(iCell);
+                } else {
+                    bulk.emplace_back(iCell);
+                }
+            }
+% endif
+        }
+    }
+
+    std::cout << "#bulk   : " << bulk.size()   << std::endl;
+    std::cout << "#lid    : " << lid_bc.size() << std::endl;
+    std::cout << "#wall   : " << box_bc.size() << std::endl;
+    std::cout << "#steps  : " << nStep         << std::endl;
+    std::cout << std::endl;
+
+    for (std::size_t iCell = 0; iCell < ${geometry.volume}; ++iCell) {
+        equilibrilize(f_prev, f_next, iCell);
+    }
+
+    const auto start = std::chrono::high_resolution_clock::now();
+
+    for (std::size_t iStep = 0; iStep < nStep; ++iStep) {
+        if (iStep % 2 == 0) {
+            f_next = f_a.get();
+            f_prev = f_b.get();
+        } else {
+            f_next = f_b.get();
+            f_prev = f_a.get();
+        }
+
+#pragma omp parallel for
+        for (std::size_t i = 0; i < bulk.size(); ++i) {
+            collide_and_stream(f_next, f_prev, bulk[i]);
+        }
+        ${float_type} u[${descriptor.d}] { 0. };
+#pragma omp parallel for
+        for (std::size_t i = 0; i < box_bc.size(); ++i) {
+            velocity_momenta_boundary(f_next, f_prev, box_bc[i], u);
+        }
+        u[0] = 0.05;
+#pragma omp parallel for
+        for (std::size_t i = 0; i < lid_bc.size(); ++i) {
+            velocity_momenta_boundary(f_next, f_prev, lid_bc[i], u);
+        }
+    }
+
+    auto duration = std::chrono::duration_cast<std::chrono::duration<double>>(
+        std::chrono::high_resolution_clock::now() - start);
+
+    std::cout << "MLUPS   : " << nStep*${geometry.volume}/(1e6*duration.count()) << std::endl;
+
+    collect_moments_to_vtk("test.vtk", f_next);
+}
+
+int main() {
+    simulate(10000);
+}