Merge pull request #658 from Xiangyu-Hu/xiangyu/revise_bidirectional_…

…buffer Xiangyu/revise bidirectional buffer
Xiangyu-Hu · Sep 16, 2024 · 29a4c04 · 29a4c04
2 parents a0ef616 + 18320c9
commit 29a4c04
Show file tree

Hide file tree

Showing 473 changed files with 13,275 additions and 6,301 deletions.
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -27,7 +27,7 @@ env:
 # A workflow is made up of one or more jobs that can run sequentially or in parallel
 jobs:
   ###############################################################################
-  Linux-float:
+  Linux-sycl:
     if: ${{ github.event_name != 'workflow_dispatch' }}
     runs-on: ubuntu-22.04
     env:
@@ -51,6 +51,11 @@ jobs:
             libfontconfig1-dev `# From here required for vcpkg opencascade`\
             libx11-dev \
             libgl-dev
+          wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | sudo tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null
+          echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list
+          sudo apt update
+          sudo apt install intel-basekit \
+            intel-oneapi-runtime-opencl
 
       - uses: hendrikmuhs/ccache-action@v1.2
         with:
@@ -59,7 +64,7 @@ jobs:
       - uses: friendlyanon/setup-vcpkg@v1 # Setup vcpkg into ${{github.workspace}}
         with:
           committish: ${{ env.VCPKG_VERSION }}
-          cache: false
+          cache-version: ${{env.VCPKG_VERSION}}
 
       - name: Install dependencies
         run: |
@@ -73,24 +78,70 @@ jobs:
             boost-geometry \
             simbody \
             gtest \
-            xsimd \
-            pybind11 \
-            opencascade
+            pybind11
 
-      - name: Generate buildsystem using float (No test)
+      - name: Generate buildsystem
         run: |
+          # sycl environment has to be setup for each session
+          source /opt/intel/oneapi/setvars.sh --include-intel-llvm
           cmake -G Ninja \
             -D CMAKE_BUILD_TYPE=Release \
+            -D CMAKE_C_COMPILER=icx -D CMAKE_CXX_COMPILER=icpx \
             -D CMAKE_TOOLCHAIN_FILE="${{github.workspace}}/vcpkg/scripts/buildsystems/vcpkg.cmake" \
             -D CMAKE_C_COMPILER_LAUNCHER=ccache -D CMAKE_CXX_COMPILER_LAUNCHER=ccache \
-            -D SPHINXSYS_USE_FLOAT=ON \
-            -D SPHINXSYS_MODULE_OPENCASCADE=ON \
             -D SPHINXSYS_CI=ON \
+            -D SPHINXSYS_USE_SYCL=ON \
+            -D TEST_STATE_RECORDING=OFF \
             -S ${{github.workspace}} \
             -B ${{github.workspace}}/build
 
-      - name: Build using float
-        run: cmake --build build --config Release --verbose
+      - name: Build libraries and tests which are using SYCL
+        run: |
+          # sycl environment has to be setup for each session        
+          source /opt/intel/oneapi/setvars.sh --include-intel-llvm
+          cmake --build build --config Release --verbose
+
+      - name: Test with the first try
+        id: first-try
+        run: |
+          source /opt/intel/oneapi/setvars.sh --include-intel-llvm
+          cd build/tests/tests_sycl
+          ctest  --output-on-failure --timeout 1000
+        continue-on-error: true
+
+      - name: Test with the second try for failed cases
+        id: second-try
+        if: ${{ steps.first-try.outcome == 'failure' }}
+        run: |
+          source /opt/intel/oneapi/setvars.sh --include-intel-llvm        
+          cd build/tests/tests_sycl
+          ctest --rerun-failed --output-on-failure --timeout 1000
+        continue-on-error: true
+
+      - name: Test with the third try for failed cases
+        id: third-try
+        if: ${{ steps.second-try.outcome == 'failure' }}
+        run: |
+          source /opt/intel/oneapi/setvars.sh --include-intel-llvm
+          cd build/tests/tests_sycl
+          ctest --rerun-failed --output-on-failure --timeout 1000
+        continue-on-error: true
+
+      - name: Test with the fourth try for failed cases
+        id: fourth-try
+        if: ${{ steps.third-try.outcome == 'failure' }}
+        run: |
+          source /opt/intel/oneapi/setvars.sh --include-intel-llvm
+          cd build/tests/tests_sycl
+          ctest --rerun-failed --output-on-failure --timeout 1000
+        continue-on-error: true
+
+      - name: Test with the last try for failed cases
+        if: ${{ steps.fourth-try.outcome == 'failure' }}
+        run: |
+          source /opt/intel/oneapi/setvars.sh --include-intel-llvm
+          cd build/tests/tests_sycl
+          ctest --rerun-failed --output-on-failure --timeout 1000
 
   ###############################################################################
   Linux-build:

diff --git a/.gitignore b/.gitignore
@@ -1,6 +1,8 @@
 ##### VS Code
 # VS Code setting up
 .vscode
+*.json
+CMakePresets.json
 
 ##### Windows
 # Windows thumbnail cache files

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -14,6 +14,7 @@ option(SPHINXSYS_DEVELOPER_MODE "Developer mode has more flags active for code q
 option(SPHINXSYS_USE_FLOAT "Build using float (single-precision floating-point format) as primary type" OFF)
 option(SPHINXSYS_USE_SIMD "Build using SIMD instructions" OFF)
 option(SPHINXSYS_MODULE_OPENCASCADE "Build extension relying on OpenCASCADE" OFF)
+option(SPHINXSYS_USE_SYCL "Build using SYCL acceleration or not" OFF)
 
 # ------ Global properties (Some cannot be set on INTERFACE targets)
 set(CMAKE_VERBOSE_MAKEFILE OFF CACHE BOOL "Enable verbose compilation commands for Makefile and Ninja" FORCE) # Extra fluff needed for Ninja: https://github.com/ninja-build/ninja/issues/900
@@ -49,6 +50,14 @@ elseif(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
     # target_compile_options(sphinxsys_core INTERFACE $<$<BOOL:${SPHINXSYS_DEVELOPER_MODE}>:-Wpedantic>) # For strict C++ standard compliance # TODO:
 endif()
 
+if(SPHINXSYS_USE_SYCL)
+    if(NOT SPHINXSYS_USE_FLOAT)
+    set(SPHINXSYS_USE_FLOAT ON)
+    message("-- Float is used as required by SPHinXsysSYCL.")
+    endif()
+endif()
+
+target_compile_definitions(sphinxsys_core INTERFACE SPHINXSYS_USE_SYCL=$<BOOL:${SPHINXSYS_USE_SYCL}>)
 target_compile_definitions(sphinxsys_core INTERFACE SPHINXSYS_USE_FLOAT=$<BOOL:${SPHINXSYS_USE_FLOAT}>)
 
 # ------ Dependencies
@@ -98,6 +107,21 @@ if(TARGET Boost::program_options)
     target_link_libraries(sphinxsys_core INTERFACE Boost::program_options)
 endif()
 
+if(SPHINXSYS_USE_SYCL)
+    set(SPHINXSYS_USE_SYCL ON)
+    if(NOT CMAKE_CXX_COMPILER_ID STREQUAL "IntelLLVM")
+        message(FATAL_ERROR "-- SPHinXsysSYCL is only supported with IntelLLVM compiler.")
+    endif()
+
+    if(NOT SPHINXSYS_SYCL_TARGETS)
+        set(SPHINXSYS_SYCL_TARGETS spir64_x86_64)
+    endif()
+
+    message("-- Set SPHinXsysSYCL target as ${SPHINXSYS_SYCL_TARGETS}")
+    target_compile_options(sphinxsys_core INTERFACE -fsycl -fsycl-targets=${SPHINXSYS_SYCL_TARGETS} -Wno-unknown-cuda-version)
+    target_link_options(sphinxsys_core INTERFACE -fsycl -fsycl-targets=${SPHINXSYS_SYCL_TARGETS} -Wno-unknown-cuda-version)
+endif()
+
 # ------ Setup the concrete libraries
 add_subdirectory(src)
 add_subdirectory(modules)

diff --git a/README.md b/README.md
@@ -1,11 +1,12 @@
 # ![SPHinXsys Logo](assets/logo.png) SPHinXsys
 
-**Notice on repository transfer to SPHinXsys team** 
+## Notice on the transformation to heterogeneous computational platform
 
-In order to promoting open-source democratization,
-this repository will be transferred to the SPHinXsys team 
-in the next few weeks.
-After the transfer, the decision-making process for SPHinXsys will be made by a number of project leaders from different institutions.
+SPHinXsys is set to undergo a major transformation, moving from traditional CPU parallelism to a new era of heterogeneous parallelism, where compute-intensive tasks can harness the power of both CPUs and GPUs. This evolution will be driven by SYCL (via Intel's DPC++), enabling us to leverage heterogeneous architectures with standard C++. Importantly, this transformation will be achieved with minimal disruption to the existing codebase, ensuring seamless continuity for current users.
+
+The groundwork for this shift is already laid out. A 2D dambreak test case has been added to the `test/test_sycl` folder, showcasing the capabilities of our specially designed framework. What's unique about this framework is that it allows for the development and testing of numerical methods even in environments without GPUs or DPC++ installed. If these methods are crafted following our specified guidelines and prove functional, they will seamlessly operate in environments equipped with DPC++ and GPU support.
+
+By embracing this new paradigm, SPHinXsys is positioning itself at the forefront of multi-physics modeling, where performance meets versatility.
 
 **Project status**  
 [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
@@ -19,7 +20,14 @@ After the transfer, the decision-making process for SPHinXsys will be made by a
 [![YouTube](https://img.shields.io/badge/YouTube-FF0000.svg?style=flat&logo=YouTube&logoColor=white)](https://www.youtube.com/channel/UCexdJbxOn9dvim6Jg1dnCFQ)
 [![Bilibili](https://img.shields.io/badge/bilibili-%E5%93%94%E5%93%A9%E5%93%94%E5%93%A9-critical)](https://space.bilibili.com/1761273682/video)
 
-## Description
+## Master and stable branches
+
+The most active development (or default) branch of this repository is `master`.
+This branch gives the updated development of SPHinXsys using heterogeneous (using TBB and SYCL) parallelism.
+
+The most stable branch of the repository is `version1.0`, which is based on CPU (using TBB) parallelism.
+
+## Repository Description
 
 SPHinXsys (pronunciation: s'fink-sis) is an acronym from **S**moothed **P**article **H**ydrodynamics for **in**dustrial comple**X** **sys**tems.
 The multi-physics library uses SPH (smoothed particle hydrodynamics) as the underlying numerical method
@@ -46,6 +54,8 @@ Here, we present several short examples in flow, solid dynamics, fluid structure
 <img src="https://github.com/Xiangyu-Hu/SPHinXsys-public-files/blob/master/videos/fish-swimming.gif" height="168px"></a>
 <a href="https://github.com/Xiangyu-Hu/SPHinXsys/blob/master/tests/2d_examples/test_2d_column_collapse/column_collapse.cpp">
 <img src="https://github.com/Xiangyu-Hu/SPHinXsys-public-files/blob/master/videos/2d_column_collapse.gif" height="168px"></a>
+<a href="https://github.com/Xiangyu-Hu/SPHinXsys/blob/master/tests/extra_source_and_tests/test_2d_T_pipe_VIPO_shell/T_pipe_VIPO_shell.cpp">
+<img src="https://github.com/Xiangyu-Hu/SPHinXsys-public-files/blob/master/videos/fluid-shell-interaction.gif" height="168px"></a>
 
 ## Fully compatible to classical FVM method
 
@@ -61,36 +71,28 @@ The only difference is that SPHinXsys reads a predefined mesh, other than genera
 
 ## Target-driven optimization
 
-The unique target-driven optimization is able to achieve the optimization target and physical solution all-in-once, 
+The unique target-driven optimization is able to achieve the optimization target and physical solution all-in-once,
 which is able to accelerate optimization process greatly.
-The following gives an example of optimizing the conductivity distribution 
+The following gives an example of optimizing the conductivity distribution
 for a thermal domain problem targeting minimum average temperature.
 
 <a href="https://github.com/Xiangyu-Hu/SPHinXsys/blob/master/tests/optimization/test_2d_VP_heat_flux_optimization/VP_heat_flux_optimization.cpp">
 <img src="https://github.com/Xiangyu-Hu/SPHinXsys-public-files/blob/master/videos/optimization.gif" height="192px"></a>
 
 Note that the physical solution of the thermal domain (right) and the optimal distribution of conductivity (left)
-are obtained at the same time when optimization is finished. 
-Also note that the entire optimization process is very fast and 
+are obtained at the same time when optimization is finished.
+Also note that the entire optimization process is very fast and
 only several times slower than that for a single physical solution with given conductivity distribution.  
 
 ## Python interface
 
-While SPHinXsys is written in C++, it provides a python interface for users to write python scripts to control the simulation, 
+While SPHinXsys is written in C++, it provides a python interface for users to write python scripts to control the simulation,
 including carry out regression tests for continuous integration (CI) and other tasks.
 One example is given below for the dambreak case.
-Please check the source code of 
-[2D Dambreak case with python interface](https://github.com/Xiangyu-Hu/SPHinXsys/tree/master/tests/2d_examples/test_2d_dambreak_python) 
+Please check the source code of
+[2D Dambreak case with python interface](https://github.com/Xiangyu-Hu/SPHinXsys/tree/master/tests/2d_examples/test_2d_dambreak_python)
 for the usage.
 
-## Heterogenous computing
-
-Recently, we have a preview release for the heterogeneous computing version of SPHinXsys. 
-By using SYCL, a royalty-free open standard developed by the Khronos Group that allows developers
-to program heterogeneous architectures in standard C++, SPHinXsys is able to utilize the power of GPU.
-Please check the [Preview Release](https://github.com/Xiangyu-Hu/SPHinXsys/releases/tag/v1.0-beta.08-sycl)
-and the [SYCL branch](https://github.com/Xiangyu-Hu/SPHinXsys/tree/sycl) for details.
-
 ## Publications
 
 Main publication on the library:
@@ -114,10 +116,10 @@ For a Docker image, check <https://hub.docker.com/r/toshev/sphinxsys>.
 
 ## Interaction with SPHinXsys and the team
 
-Thank you for using and supporting our open-source project! 
+Thank you for using and supporting our open-source project!
 We value each feedback.
 
-#### For SPHinXsys users:
+#### For SPHinXsys users
 
 Your input is crucial to us. We encourage you to report any issues you encounter with the library, including:
 
@@ -128,7 +130,7 @@ Your input is crucial to us. We encourage you to report any issues you encounter
 
 We particularly appreciate feedback stemming from practical simulations or projects, as these insights are essential for improving SPHinXsys.
 
-#### For SPHinXsys developers:
+#### For SPHinXsys developers
 
 If you don't have a GitHub account yet, please register for one. Fork the SPHinXsys repository to add new features or improve existing ones. Once your changes are ready, commit them and initiate a pull request to have your contributions merged into the main repository.
 

diff --git a/modules/opencascade/opencascade/relax_dynamics_surface.cpp b/modules/opencascade/opencascade/relax_dynamics_surface.cpp
@@ -11,8 +11,8 @@ namespace relax_dynamics
 {
 //=================================================================================================//
 ShapeSurfaceBounding2::ShapeSurfaceBounding2(RealBody &real_body_)
-    : LocalDynamics(real_body_), DataDelegateSimple(real_body_),
-      pos_(*particles_->getVariableDataByName<Vecd>("Position"))
+    : LocalDynamics(real_body_),
+      pos_(particles_->getVariableDataByName<Vecd>("Position"))
 {
     shape_ = &real_body_.getInitialShape();
 }
@@ -24,7 +24,7 @@ void ShapeSurfaceBounding2::update(size_t index_i, Real dt)
 //=================================================================================================//
 RelaxationStepInnerFirstHalf::
     RelaxationStepInnerFirstHalf(BaseInnerRelation &inner_relation)
-    : BaseDynamics<void>(inner_relation.getSPHBody()), real_body_(inner_relation.real_body_),
+    : BaseDynamics<void>(), real_body_(inner_relation.real_body_),
       inner_relation_(inner_relation), relaxation_acceleration_inner_(inner_relation) {}
 //=================================================================================================//
 void RelaxationStepInnerFirstHalf::exec(Real dt)
@@ -37,7 +37,7 @@ void RelaxationStepInnerFirstHalf::exec(Real dt)
 //=================================================================================================//
 RelaxationStepInnerSecondHalf::
     RelaxationStepInnerSecondHalf(BaseInnerRelation &inner_relation)
-    : BaseDynamics<void>(inner_relation.getSPHBody()), real_body_(inner_relation.real_body_),
+    : BaseDynamics<void>(), real_body_(inner_relation.real_body_),
       get_time_step_square_(*real_body_), update_particle_position_(*real_body_),
       surface_bounding_(*real_body_)
 {
@@ -52,10 +52,10 @@ void RelaxationStepInnerSecondHalf::exec(Real dt)
 
 //=================================================================================================//
 SurfaceNormalDirection::SurfaceNormalDirection(SPHBody &sph_body)
-    : DataDelegateSimple(sph_body), LocalDynamics(sph_body),
+    : LocalDynamics(sph_body),
       surface_shape_(DynamicCast<SurfaceShape>(this, &sph_body.getInitialShape())),
-      pos_(*particles_->getVariableDataByName<Vecd>("Position")),
-      n_(*particles_->registerSharedVariable<Vecd>("NormalDirection")) {}
+      pos_(particles_->getVariableDataByName<Vecd>("Position")),
+      n_(particles_->registerStateVariable<Vecd>("NormalDirection")) {}
 
 //=================================================================================================//
 void SurfaceNormalDirection::update(size_t index_i, Real dt)

diff --git a/modules/opencascade/opencascade/relax_dynamics_surface.h b/modules/opencascade/opencascade/relax_dynamics_surface.h
@@ -41,16 +41,15 @@ class SurfaceShape;
 
 namespace relax_dynamics
 {
-class ShapeSurfaceBounding2 : public LocalDynamics,
-                              public DataDelegateSimple
+class ShapeSurfaceBounding2 : public LocalDynamics
 {
   public:
     ShapeSurfaceBounding2(RealBody &real_body_);
     virtual ~ShapeSurfaceBounding2(){};
     void update(size_t index_i, Real dt = 0.0);
 
   protected:
-    StdLargeVec<Vecd> &pos_;
+    Vecd *pos_;
     Shape *shape_;
 };
 
@@ -86,7 +85,7 @@ class RelaxationStepInnerSecondHalf : public BaseDynamics<void>
  * @class SurfaceNormalDirection
  * @brief get the normal direction of surface particles.
  */
-class SurfaceNormalDirection : public DataDelegateSimple, public LocalDynamics
+class SurfaceNormalDirection : public LocalDynamics
 {
   public:
     explicit SurfaceNormalDirection(SPHBody &sph_body);
@@ -95,7 +94,7 @@ class SurfaceNormalDirection : public DataDelegateSimple, public LocalDynamics
 
   protected:
     SurfaceShape *surface_shape_;
-    StdLargeVec<Vecd> &pos_, &n_;
+    Vecd *pos_, *n_;
 };
 
 } // namespace relax_dynamics