├── .gitignore ├── CMakeLists.txt ├── LICENSE ├── README.html ├── README.md ├── README.pdf ├── config ├── config_convert.yaml └── config_infer.yaml ├── img ├── image-20240723161128904.png ├── image-20240723161407539.png ├── image-20240723171056775.png └── image-20240724152920430.png ├── src ├── application │ ├── app_yolo_obb.cpp │ ├── app_yolo_obb │ │ ├── yolo_obb.cpp │ │ ├── yolo_obb.hpp │ │ └── yolo_obb_decode.cu │ ├── common │ │ ├── face_detector.hpp │ │ └── object_detector.hpp │ └── tools │ │ ├── Eigen │ │ ├── CMakeLists.txt │ │ ├── Cholesky │ │ ├── CholmodSupport │ │ ├── Core │ │ ├── Dense │ │ ├── Eigen │ │ ├── Eigenvalues │ │ ├── Geometry │ │ ├── Householder │ │ ├── IterativeLinearSolvers │ │ ├── Jacobi │ │ ├── LU │ │ ├── MetisSupport │ │ ├── OrderingMethods │ │ ├── PaStiXSupport │ │ ├── PardisoSupport │ │ ├── QR │ │ ├── QtAlignedMalloc │ │ ├── SPQRSupport │ │ ├── SVD │ │ ├── Sparse │ │ ├── SparseCholesky │ │ ├── SparseCore │ │ ├── SparseLU │ │ ├── SparseQR │ │ ├── StdDeque │ │ ├── StdList │ │ ├── StdVector │ │ ├── SuperLUSupport │ │ ├── UmfPackSupport │ │ └── src │ │ │ ├── Cholesky │ │ │ ├── LDLT.h │ │ │ ├── LLT.h │ │ │ └── LLT_LAPACKE.h │ │ │ ├── CholmodSupport │ │ │ └── CholmodSupport.h │ │ │ ├── Core │ │ │ ├── Array.h │ │ │ ├── ArrayBase.h │ │ │ ├── ArrayWrapper.h │ │ │ ├── Assign.h │ │ │ ├── AssignEvaluator.h │ │ │ ├── Assign_MKL.h │ │ │ ├── BandMatrix.h │ │ │ ├── Block.h │ │ │ ├── BooleanRedux.h │ │ │ ├── CommaInitializer.h │ │ │ ├── ConditionEstimator.h │ │ │ ├── CoreEvaluators.h │ │ │ ├── CoreIterators.h │ │ │ ├── CwiseBinaryOp.h │ │ │ ├── CwiseNullaryOp.h │ │ │ ├── CwiseTernaryOp.h │ │ │ ├── CwiseUnaryOp.h │ │ │ ├── CwiseUnaryView.h │ │ │ ├── DenseBase.h │ │ │ ├── DenseCoeffsBase.h │ │ │ ├── DenseStorage.h │ │ │ ├── Diagonal.h │ │ │ ├── DiagonalMatrix.h │ │ │ ├── DiagonalProduct.h │ │ │ ├── Dot.h │ │ │ ├── EigenBase.h │ │ │ ├── ForceAlignedAccess.h │ │ │ ├── Fuzzy.h │ │ │ ├── GeneralProduct.h │ │ │ ├── GenericPacketMath.h │ │ │ ├── GlobalFunctions.h │ │ │ ├── IO.h │ │ │ ├── Inverse.h │ │ │ ├── Map.h │ │ │ ├── MapBase.h │ │ │ ├── MathFunctions.h │ │ │ ├── MathFunctionsImpl.h │ │ │ ├── Matrix.h │ │ │ ├── MatrixBase.h │ │ │ ├── NestByValue.h │ │ │ ├── NoAlias.h │ │ │ ├── NumTraits.h │ │ │ ├── PermutationMatrix.h │ │ │ ├── PlainObjectBase.h │ │ │ ├── Product.h │ │ │ ├── ProductEvaluators.h │ │ │ ├── Random.h │ │ │ ├── Redux.h │ │ │ ├── Ref.h │ │ │ ├── Replicate.h │ │ │ ├── ReturnByValue.h │ │ │ ├── Reverse.h │ │ │ ├── Select.h │ │ │ ├── SelfAdjointView.h │ │ │ ├── SelfCwiseBinaryOp.h │ │ │ ├── Solve.h │ │ │ ├── SolveTriangular.h │ │ │ ├── SolverBase.h │ │ │ ├── StableNorm.h │ │ │ ├── Stride.h │ │ │ ├── Swap.h │ │ │ ├── Transpose.h │ │ │ ├── Transpositions.h │ │ │ ├── TriangularMatrix.h │ │ │ ├── VectorBlock.h │ │ │ ├── VectorwiseOp.h │ │ │ ├── Visitor.h │ │ │ ├── arch │ │ │ │ ├── AVX │ │ │ │ │ ├── Complex.h │ │ │ │ │ ├── MathFunctions.h │ │ │ │ │ ├── PacketMath.h │ │ │ │ │ └── TypeCasting.h │ │ │ │ ├── AVX512 │ │ │ │ │ ├── MathFunctions.h │ │ │ │ │ └── PacketMath.h │ │ │ │ ├── AltiVec │ │ │ │ │ ├── Complex.h │ │ │ │ │ ├── MathFunctions.h │ │ │ │ │ └── PacketMath.h │ │ │ │ ├── CUDA │ │ │ │ │ ├── Complex.h │ │ │ │ │ ├── Half.h │ │ │ │ │ ├── MathFunctions.h │ │ │ │ │ ├── PacketMath.h │ │ │ │ │ ├── PacketMathHalf.h │ │ │ │ │ └── TypeCasting.h │ │ │ │ ├── Default │ │ │ │ │ └── Settings.h │ │ │ │ ├── NEON │ │ │ │ │ ├── Complex.h │ │ │ │ │ ├── MathFunctions.h │ │ │ │ │ └── PacketMath.h │ │ │ │ ├── SSE │ │ │ │ │ ├── Complex.h │ │ │ │ │ ├── MathFunctions.h │ │ │ │ │ ├── PacketMath.h │ │ │ │ │ └── TypeCasting.h │ │ │ │ └── ZVector │ │ │ │ │ ├── Complex.h │ │ │ │ │ ├── MathFunctions.h │ │ │ │ │ └── PacketMath.h │ │ │ ├── functors │ │ │ │ ├── AssignmentFunctors.h │ │ │ │ ├── BinaryFunctors.h │ │ │ │ ├── NullaryFunctors.h │ │ │ │ ├── StlFunctors.h │ │ │ │ ├── TernaryFunctors.h │ │ │ │ └── UnaryFunctors.h │ │ │ ├── products │ │ │ │ ├── GeneralBlockPanelKernel.h │ │ │ │ ├── GeneralMatrixMatrix.h │ │ │ │ ├── GeneralMatrixMatrixTriangular.h │ │ │ │ ├── GeneralMatrixMatrixTriangular_BLAS.h │ │ │ │ ├── GeneralMatrixMatrix_BLAS.h │ │ │ │ ├── GeneralMatrixVector.h │ │ │ │ ├── GeneralMatrixVector_BLAS.h │ │ │ │ ├── Parallelizer.h │ │ │ │ ├── SelfadjointMatrixMatrix.h │ │ │ │ ├── SelfadjointMatrixMatrix_BLAS.h │ │ │ │ ├── SelfadjointMatrixVector.h │ │ │ │ ├── SelfadjointMatrixVector_BLAS.h │ │ │ │ ├── SelfadjointProduct.h │ │ │ │ ├── SelfadjointRank2Update.h │ │ │ │ ├── TriangularMatrixMatrix.h │ │ │ │ ├── TriangularMatrixMatrix_BLAS.h │ │ │ │ ├── TriangularMatrixVector.h │ │ │ │ ├── TriangularMatrixVector_BLAS.h │ │ │ │ ├── TriangularSolverMatrix.h │ │ │ │ ├── TriangularSolverMatrix_BLAS.h │ │ │ │ └── TriangularSolverVector.h │ │ │ └── util │ │ │ │ ├── BlasUtil.h │ │ │ │ ├── Constants.h │ │ │ │ ├── DisableStupidWarnings.h │ │ │ │ ├── ForwardDeclarations.h │ │ │ │ ├── MKL_support.h │ │ │ │ ├── Macros.h │ │ │ │ ├── Memory.h │ │ │ │ ├── Meta.h │ │ │ │ ├── NonMPL2.h │ │ │ │ ├── ReenableStupidWarnings.h │ │ │ │ ├── StaticAssert.h │ │ │ │ └── XprHelper.h │ │ │ ├── Eigenvalues │ │ │ ├── ComplexEigenSolver.h │ │ │ ├── ComplexSchur.h │ │ │ ├── ComplexSchur_LAPACKE.h │ │ │ ├── EigenSolver.h │ │ │ ├── GeneralizedEigenSolver.h │ │ │ ├── GeneralizedSelfAdjointEigenSolver.h │ │ │ ├── HessenbergDecomposition.h │ │ │ ├── MatrixBaseEigenvalues.h │ │ │ ├── RealQZ.h │ │ │ ├── RealSchur.h │ │ │ ├── RealSchur_LAPACKE.h │ │ │ ├── SelfAdjointEigenSolver.h │ │ │ ├── SelfAdjointEigenSolver_LAPACKE.h │ │ │ └── Tridiagonalization.h │ │ │ ├── Geometry │ │ │ ├── AlignedBox.h │ │ │ ├── AngleAxis.h │ │ │ ├── EulerAngles.h │ │ │ ├── Homogeneous.h │ │ │ ├── Hyperplane.h │ │ │ ├── OrthoMethods.h │ │ │ ├── ParametrizedLine.h │ │ │ ├── Quaternion.h │ │ │ ├── Rotation2D.h │ │ │ ├── RotationBase.h │ │ │ ├── Scaling.h │ │ │ ├── Transform.h │ │ │ ├── Translation.h │ │ │ ├── Umeyama.h │ │ │ └── arch │ │ │ │ └── Geometry_SSE.h │ │ │ ├── Householder │ │ │ ├── BlockHouseholder.h │ │ │ ├── Householder.h │ │ │ └── HouseholderSequence.h │ │ │ ├── IterativeLinearSolvers │ │ │ ├── BasicPreconditioners.h │ │ │ ├── BiCGSTAB.h │ │ │ ├── ConjugateGradient.h │ │ │ ├── IncompleteCholesky.h │ │ │ ├── IncompleteLUT.h │ │ │ ├── IterativeSolverBase.h │ │ │ ├── LeastSquareConjugateGradient.h │ │ │ └── SolveWithGuess.h │ │ │ ├── Jacobi │ │ │ └── Jacobi.h │ │ │ ├── LU │ │ │ ├── Determinant.h │ │ │ ├── FullPivLU.h │ │ │ ├── InverseImpl.h │ │ │ ├── PartialPivLU.h │ │ │ ├── PartialPivLU_LAPACKE.h │ │ │ └── arch │ │ │ │ └── Inverse_SSE.h │ │ │ ├── MetisSupport │ │ │ └── MetisSupport.h │ │ │ ├── OrderingMethods │ │ │ ├── Amd.h │ │ │ ├── Eigen_Colamd.h │ │ │ └── Ordering.h │ │ │ ├── PaStiXSupport │ │ │ └── PaStiXSupport.h │ │ │ ├── PardisoSupport │ │ │ └── PardisoSupport.h │ │ │ ├── QR │ │ │ ├── ColPivHouseholderQR.h │ │ │ ├── ColPivHouseholderQR_LAPACKE.h │ │ │ ├── CompleteOrthogonalDecomposition.h │ │ │ ├── FullPivHouseholderQR.h │ │ │ ├── HouseholderQR.h │ │ │ └── HouseholderQR_LAPACKE.h │ │ │ ├── SPQRSupport │ │ │ └── SuiteSparseQRSupport.h │ │ │ ├── SVD │ │ │ ├── BDCSVD.h │ │ │ ├── JacobiSVD.h │ │ │ ├── JacobiSVD_LAPACKE.h │ │ │ ├── SVDBase.h │ │ │ └── UpperBidiagonalization.h │ │ │ ├── SparseCholesky │ │ │ ├── SimplicialCholesky.h │ │ │ └── SimplicialCholesky_impl.h │ │ │ ├── SparseCore │ │ │ ├── AmbiVector.h │ │ │ ├── CompressedStorage.h │ │ │ ├── ConservativeSparseSparseProduct.h │ │ │ ├── MappedSparseMatrix.h │ │ │ ├── SparseAssign.h │ │ │ ├── SparseBlock.h │ │ │ ├── SparseColEtree.h │ │ │ ├── SparseCompressedBase.h │ │ │ ├── SparseCwiseBinaryOp.h │ │ │ ├── SparseCwiseUnaryOp.h │ │ │ ├── SparseDenseProduct.h │ │ │ ├── SparseDiagonalProduct.h │ │ │ ├── SparseDot.h │ │ │ ├── SparseFuzzy.h │ │ │ ├── SparseMap.h │ │ │ ├── SparseMatrix.h │ │ │ ├── SparseMatrixBase.h │ │ │ ├── SparsePermutation.h │ │ │ ├── SparseProduct.h │ │ │ ├── SparseRedux.h │ │ │ ├── SparseRef.h │ │ │ ├── SparseSelfAdjointView.h │ │ │ ├── SparseSolverBase.h │ │ │ ├── SparseSparseProductWithPruning.h │ │ │ ├── SparseTranspose.h │ │ │ ├── SparseTriangularView.h │ │ │ ├── SparseUtil.h │ │ │ ├── SparseVector.h │ │ │ ├── SparseView.h │ │ │ └── TriangularSolver.h │ │ │ ├── SparseLU │ │ │ ├── SparseLU.h │ │ │ ├── SparseLUImpl.h │ │ │ ├── SparseLU_Memory.h │ │ │ ├── SparseLU_Structs.h │ │ │ ├── SparseLU_SupernodalMatrix.h │ │ │ ├── SparseLU_Utils.h │ │ │ ├── SparseLU_column_bmod.h │ │ │ ├── SparseLU_column_dfs.h │ │ │ ├── SparseLU_copy_to_ucol.h │ │ │ ├── SparseLU_gemm_kernel.h │ │ │ ├── SparseLU_heap_relax_snode.h │ │ │ ├── SparseLU_kernel_bmod.h │ │ │ ├── SparseLU_panel_bmod.h │ │ │ ├── SparseLU_panel_dfs.h │ │ │ ├── SparseLU_pivotL.h │ │ │ ├── SparseLU_pruneL.h │ │ │ └── SparseLU_relax_snode.h │ │ │ ├── SparseQR │ │ │ └── SparseQR.h │ │ │ ├── StlSupport │ │ │ ├── StdDeque.h │ │ │ ├── StdList.h │ │ │ ├── StdVector.h │ │ │ └── details.h │ │ │ ├── SuperLUSupport │ │ │ └── SuperLUSupport.h │ │ │ ├── UmfPackSupport │ │ │ └── UmfPackSupport.h │ │ │ ├── misc │ │ │ ├── Image.h │ │ │ ├── Kernel.h │ │ │ ├── RealSvd2x2.h │ │ │ ├── blas.h │ │ │ ├── lapack.h │ │ │ ├── lapacke.h │ │ │ └── lapacke_mangling.h │ │ │ └── plugins │ │ │ ├── ArrayCwiseBinaryOps.h │ │ │ ├── ArrayCwiseUnaryOps.h │ │ │ ├── BlockMethods.h │ │ │ ├── CommonCwiseBinaryOps.h │ │ │ ├── CommonCwiseUnaryOps.h │ │ │ ├── MatrixCwiseBinaryOps.h │ │ │ └── MatrixCwiseUnaryOps.h │ │ ├── auto_download.cpp │ │ ├── deepsort.cpp │ │ ├── deepsort.hpp │ │ ├── pybind11.hpp │ │ ├── zmq_remote_show.cpp │ │ ├── zmq_remote_show.hpp │ │ ├── zmq_u.cpp │ │ └── zmq_u.hpp ├── main.cpp └── tensorRT │ ├── builder │ ├── trt_builder.cpp │ └── trt_builder.hpp │ ├── common │ ├── cuda_tools.cpp │ ├── cuda_tools.hpp │ ├── ilogger.cpp │ ├── ilogger.hpp │ ├── infer_controller.hpp │ ├── json.cpp │ ├── json.hpp │ ├── monopoly_allocator.hpp │ ├── preprocess_kernel.cu │ ├── preprocess_kernel.cuh │ ├── trt_tensor.cpp │ └── trt_tensor.hpp │ ├── import_lib.cpp │ ├── infer │ ├── trt_infer.cpp │ └── trt_infer.hpp │ ├── onnx │ ├── onnx-ml.pb.cpp │ ├── onnx-ml.pb.h │ ├── onnx-operators-ml.pb.cpp │ ├── onnx-operators-ml.pb.h │ ├── onnx_pb.h │ ├── onnxifi.h │ └── readme.md │ ├── onnx_parser │ ├── ImporterContext.hpp │ ├── LoopHelpers.cpp │ ├── LoopHelpers.hpp │ ├── ModelImporter.cpp │ ├── ModelImporter.hpp │ ├── NvOnnxParser.cpp │ ├── NvOnnxParser.h │ ├── OnnxAttrs.cpp │ ├── OnnxAttrs.hpp │ ├── RNNHelpers.cpp │ ├── RNNHelpers.hpp │ ├── ShapeTensor.cpp │ ├── ShapeTensor.hpp │ ├── ShapedWeights.cpp │ ├── ShapedWeights.hpp │ ├── Status.hpp │ ├── TensorOrWeights.hpp │ ├── builtin_op_importers.cpp │ ├── builtin_op_importers.hpp │ ├── onnx2trt.hpp │ ├── onnx2trt_common.hpp │ ├── onnx2trt_runtime.hpp │ ├── onnx2trt_utils.cpp │ ├── onnx2trt_utils.hpp │ ├── onnxErrorRecorder.cpp │ ├── onnxErrorRecorder.hpp │ ├── onnx_utils.hpp │ ├── readme.md │ ├── toposort.hpp │ ├── trt_utils.hpp │ └── utils.hpp │ └── onnxplugin │ ├── onnxplugin.cpp │ ├── onnxplugin.hpp │ ├── plugin_binary_io.cpp │ ├── plugin_binary_io.hpp │ └── plugins │ ├── DCNv2.cu │ ├── HSigmoid.cu │ ├── HSwish.cu │ ├── ScatterND.cu │ └── custom_layernorm.cu └── yolov8_obb芯片引脚缺陷检测.pdf /.gitignore: -------------------------------------------------------------------------------- 1 | 2 | # temp tensor and data 3 | *.tensor 4 | *.data 5 | 6 | 7 | # Prerequisites 8 | *.d 9 | 10 | # Compiled Object files 11 | *.slo 12 | *.lo 13 | *.o 14 | *.obj 15 | 16 | # Precompiled Headers 17 | *.gch 18 | *.pch 19 | 20 | # Compiled Dynamic libraries 21 | *.so 22 | *.dylib 23 | *.dll 24 | 25 | # Compiled Static libraries 26 | *.lai 27 | *.la 28 | *.a 29 | *.lib 30 | 31 | # Executables 32 | *.exe 33 | *.out 34 | *.app 35 | 36 | /objs 37 | 38 | *.trtmodel 39 | *.onnx 40 | /workspace/pro 41 | /build 42 | 43 | /.vs 44 | *.pyd 45 | *.zip 46 | *.pdb 47 | *.ilk 48 | *.lib 49 | *.exp 50 | 51 | /workspace/* 52 | /.vscode/ -------------------------------------------------------------------------------- /CMakeLists.txt: -------------------------------------------------------------------------------- 1 | cmake_minimum_required(VERSION 2.6) 2 | project(pro) 3 | 4 | option(CUDA_USE_STATIC_CUDA_RUNTIME OFF) 5 | set(CMAKE_CXX_STANDARD 11) 6 | set(CMAKE_BUILD_TYPE Debug) 7 | set(EXECUTABLE_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/workspace) 8 | 9 | # 如果你是不同显卡，请设置为显卡对应的号码参考这里：https://developer.nvidia.com/zh-cn/cuda-gpus#compute 10 | set(CUDA_GEN_CODE "-gencode=arch=compute_87,code=sm_87") 11 | 12 | # 如果你的opencv找不到，可以自己指定目录 13 | set(OpenCV_DIR "/usr/local/include/opencv4") 14 | 15 | set(CUDA_TOOLKIT_ROOT_DIR "/usr/local/cuda-11.4") 16 | set(CUDNN_DIR "/usr/local/cudnn8.4.0.27-cuda11.6") 17 | 18 | # 因为protobuf，需要用特定版本，所以这里指定路径 19 | set(PROTOBUF_DIR "/usr/include/google/protobuf") 20 | 21 | find_package(CUDA REQUIRED) 22 | find_package(OpenCV) 23 | find_package(yaml-cpp REQUIRED) 24 | include_directories( 25 | ${PROJECT_SOURCE_DIR}/src 26 | ${PROJECT_SOURCE_DIR}/src/application 27 | ${PROJECT_SOURCE_DIR}/src/tensorRT 28 | ${PROJECT_SOURCE_DIR}/src/tensorRT/common 29 | ${OpenCV_INCLUDE_DIRS} 30 | ${CUDA_TOOLKIT_ROOT_DIR}/include 31 | ${PROTOBUF_DIR}/include 32 | ${CUDNN_DIR}/include 33 | ${OpenCV_DIR}/include/opencv4 34 | ) 35 | 36 | # 切记，protobuf的lib目录一定要比tensorRT目录前面，因为tensorRTlib下带有protobuf的so文件 37 | # 这可能带来错误 38 | link_directories( 39 | ${PROTOBUF_DIR}/lib 40 | ${CUDA_TOOLKIT_ROOT_DIR}/lib64 41 | ${CUDNN_DIR}/lib 42 | ${OpenCV_DIR}/lib 43 | ) 44 | 45 | set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -O0 -Wfatal-errors -pthread -w -g") 46 | set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -std=c++11 -O0 -Xcompiler -fPIC -g -w ${CUDA_GEN_CODE}") 47 | file(GLOB_RECURSE cpp_srcs ${PROJECT_SOURCE_DIR}/src/*.cpp) 48 | file(GLOB_RECURSE cuda_srcs ${PROJECT_SOURCE_DIR}/src/*.cu) 49 | cuda_add_library(plugin_list SHARED ${cuda_srcs}) 50 | target_link_libraries(plugin_list nvinfer nvinfer_plugin) 51 | target_link_libraries(plugin_list cuda cublas cudart cudnn) 52 | target_link_libraries(plugin_list protobuf pthread) 53 | target_link_libraries(plugin_list ${OpenCV_LIBS}) 54 | target_link_libraries(plugin_list opencv_core opencv_imgproc opencv_videoio opencv_highgui opencv_imgcodecs) 55 | 56 | ########################## custom_layernorm.so ################################ 57 | cuda_add_library(custom_layernorm SHARED 58 | src/tensorRT/onnxplugin/plugins/custom_layernorm.cu 59 | ) 60 | 61 | target_link_libraries(custom_layernorm 62 | libnvinfer.so 63 | libnvinfer_plugin.so 64 | ) 65 | 66 | add_executable(pro ${cpp_srcs}) 67 | 68 | # 如果提示插件找不到，请使用dlopen(xxx.so, NOW)的方式手动加载可以解决插件找不到问题 69 | target_link_libraries(pro nvinfer nvinfer_plugin) 70 | target_link_libraries(pro cuda cublas cudart cudnn) 71 | target_link_libraries(pro protobuf pthread plugin_list) 72 | target_link_libraries(pro ${OpenCV_LIBS}) 73 | target_link_libraries(pro opencv_core opencv_imgproc opencv_videoio opencv_highgui opencv_imgcodecs) 74 | target_link_libraries(pro yaml-cpp) 75 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2023 WenGuang Zhou 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. -------------------------------------------------------------------------------- /README.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/xjhaz/yolov8_obb_ChipPinDefectDetection/11aa8aa73472a2436809f2f21208127a5836a912/README.pdf -------------------------------------------------------------------------------- /config/config_convert.yaml: -------------------------------------------------------------------------------- 1 | mode: "FP16" # FP32 or FP16 or INT8 2 | model: "best" # Your model name -------------------------------------------------------------------------------- /config/config_infer.yaml: -------------------------------------------------------------------------------- 1 | source_mode: 1 # Source mode, '1' for using the video file specified in 'video_path'. '0' for using video input from devices like 'video0', 'video1', etc. 2 | video: 0 # Video input device setting, '0' refer to device '/dev/video0'. 3 | video_path: "./Video_00001.mp4" # Path to the video file used for processing. 4 | engine_file: "./three.FP32.trtmodel" # Path to the TensorRT engine file. 5 | gpu_id: 0 # GPU ID, '0' refers to the first GPU to be used for processing. 6 | confidence_threshold: 0.75 # Confidence threshold, filters detections having a confidence score less than 0.75. 7 | nms_threshold: 0.3 # Non-maximum suppression (NMS) threshold, used to resolve overlapping bounding boxes. 8 | nms_method: "FastGPU" # NMS method, options include 'FastGPU' for GPU-accelerated computation or 'CPU' for CPU-based computation. 9 | max_objects: 1024 # Maximum number of objects to detect, limiting the detection to 1024 objects. 10 | preprocess_multi_stream: false # Preprocess multi stream flag, 'false' indicates that preprocessing for multiple streams is not enabled. 11 | -------------------------------------------------------------------------------- /img/image-20240723161128904.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/xjhaz/yolov8_obb_ChipPinDefectDetection/11aa8aa73472a2436809f2f21208127a5836a912/img/image-20240723161128904.png -------------------------------------------------------------------------------- /img/image-20240723161407539.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/xjhaz/yolov8_obb_ChipPinDefectDetection/11aa8aa73472a2436809f2f21208127a5836a912/img/image-20240723161407539.png -------------------------------------------------------------------------------- /img/image-20240723171056775.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/xjhaz/yolov8_obb_ChipPinDefectDetection/11aa8aa73472a2436809f2f21208127a5836a912/img/image-20240723171056775.png -------------------------------------------------------------------------------- /img/image-20240724152920430.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/xjhaz/yolov8_obb_ChipPinDefectDetection/11aa8aa73472a2436809f2f21208127a5836a912/img/image-20240724152920430.png -------------------------------------------------------------------------------- /src/application/app_yolo_obb/yolo_obb.hpp: -------------------------------------------------------------------------------- 1 | #ifndef YOLO_OBB_HPP 2 | #define YOLO_OBB_HPP 3 | 4 | #include 5 | #include 6 | #include 7 | #include 8 | #include 9 | #include 10 | 11 | namespace YoloOBB{ 12 | 13 | using namespace std; 14 | 15 | struct Box{ 16 | float center_x, center_y, width, height, angle, confidence; 17 | int class_label; 18 | 19 | Box() = default; 20 | 21 | Box(float center_x, float center_y, float width, float height, float angle, float confidence, int class_label) 22 | :center_x(center_x), center_y(center_y), width(width), height(height), angle(angle), confidence(confidence), class_label(class_label){} 23 | }; 24 | typedef vector BoxArray; 25 | 26 | enum class NMSMethod : int{ 27 | CPU = 0, // General, for estimate mAP 28 | FastGPU = 1 // Fast NMS with a small loss of accuracy in corner cases 29 | }; 30 | 31 | void image_to_tensor(const cv::Mat& image, shared_ptr& tensor, int ibatch); 32 | 33 | class Infer{ 34 | public: 35 | virtual shared_future commit(const cv::Mat& image) = 0; 36 | virtual vector> commits(const vector& images) = 0; 37 | }; 38 | 39 | shared_ptr create_infer( 40 | const string& engine_file, int gpuid, 41 | float confidence_threshold=0.25f, float nms_threshold=0.5f, 42 | NMSMethod nms_method = NMSMethod::FastGPU, int max_objects = 1024, 43 | bool use_multi_preprocess_stream = false 44 | ); 45 | 46 | }; // namespace YoloOBB 47 | 48 | #endif // YOLO_OBB_HPP -------------------------------------------------------------------------------- /src/application/common/face_detector.hpp: -------------------------------------------------------------------------------- 1 | #ifndef FACE_DETECTOR_HPP 2 | #define FACE_DETECTOR_HPP 3 | 4 | #include 5 | #include 6 | 7 | namespace FaceDetector{ 8 | 9 | struct Box{ 10 | float left, top, right, bottom, confidence; 11 | float landmark[10]; 12 | 13 | cv::Rect cvbox() const{return cv::Rect(left, top, right-left, bottom-top);} 14 | float width() const{return std::max(0.0f, right-left);} 15 | float height() const{return std::max(0.0f, bottom-top);} 16 | float area() const{return width() * height();} 17 | float get_left() {return left;} 18 | void set_left(float value) {left = value;} 19 | float get_top() {return top;} 20 | void set_top(float value) {top = value;} 21 | float get_right() {return right;} 22 | void set_right(float value) {right = value;} 23 | float get_bottom() {return bottom;} 24 | void set_bottom(float value) {bottom = value;} 25 | float get_confidence() {return confidence;} 26 | void set_confidence(float value){confidence = value;} 27 | }; 28 | 29 | typedef std::vector BoxArray; 30 | }; 31 | 32 | #endif // FACE_DETECTOR_HPP -------------------------------------------------------------------------------- /src/application/common/object_detector.hpp: -------------------------------------------------------------------------------- 1 | #ifndef OBJECT_DETECTOR_HPP 2 | #define OBJECT_DETECTOR_HPP 3 | 4 | #include 5 | 6 | namespace ObjectDetector{ 7 | 8 | struct Box{ 9 | float left, top, right, bottom, confidence; 10 | int class_label; 11 | 12 | Box() = default; 13 | 14 | Box(float left, float top, float right, float bottom, float confidence, int class_label) 15 | :left(left), top(top), right(right), bottom(bottom), confidence(confidence), class_label(class_label){} 16 | }; 17 | 18 | typedef std::vector BoxArray; 19 | }; 20 | 21 | 22 | #endif // OBJECT_DETECTOR_HPP -------------------------------------------------------------------------------- /src/application/tools/Eigen/CMakeLists.txt: -------------------------------------------------------------------------------- 1 | include(RegexUtils) 2 | test_escape_string_as_regex() 3 | 4 | file(GLOB Eigen_directory_files "*") 5 | 6 | escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}") 7 | 8 | foreach(f ${Eigen_directory_files}) 9 | if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/src") 10 | list(APPEND Eigen_directory_files_to_install ${f}) 11 | endif() 12 | endforeach(f ${Eigen_directory_files}) 13 | 14 | install(FILES 15 | ${Eigen_directory_files_to_install} 16 | DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel 17 | ) 18 | 19 | install(DIRECTORY src DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel FILES_MATCHING PATTERN "*.h") 20 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/Cholesky: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_CHOLESKY_MODULE_H 9 | #define EIGEN_CHOLESKY_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | /** \defgroup Cholesky_Module Cholesky module 16 | * 17 | * 18 | * 19 | * This module provides two variants of the Cholesky decomposition for selfadjoint (hermitian) matrices. 20 | * Those decompositions are also accessible via the following methods: 21 | * - MatrixBase::llt() 22 | * - MatrixBase::ldlt() 23 | * - SelfAdjointView::llt() 24 | * - SelfAdjointView::ldlt() 25 | * 26 | * \code 27 | * #include 28 | * \endcode 29 | */ 30 | 31 | #include "src/Cholesky/LLT.h" 32 | #include "src/Cholesky/LDLT.h" 33 | #ifdef EIGEN_USE_LAPACKE 34 | #include "src/misc/lapacke.h" 35 | #include "src/Cholesky/LLT_LAPACKE.h" 36 | #endif 37 | 38 | #include "src/Core/util/ReenableStupidWarnings.h" 39 | 40 | #endif // EIGEN_CHOLESKY_MODULE_H 41 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 42 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/CholmodSupport: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_CHOLMODSUPPORT_MODULE_H 9 | #define EIGEN_CHOLMODSUPPORT_MODULE_H 10 | 11 | #include "SparseCore" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | extern "C" { 16 | #include 17 | } 18 | 19 | /** \ingroup Support_modules 20 | * \defgroup CholmodSupport_Module CholmodSupport module 21 | * 22 | * This module provides an interface to the Cholmod library which is part of the suitesparse package. 23 | * It provides the two following main factorization classes: 24 | * - class CholmodSupernodalLLT: a supernodal LLT Cholesky factorization. 25 | * - class CholmodDecomposiiton: a general L(D)LT Cholesky factorization with automatic or explicit runtime selection of the underlying factorization method (supernodal or simplicial). 26 | * 27 | * For the sake of completeness, this module also propose the two following classes: 28 | * - class CholmodSimplicialLLT 29 | * - class CholmodSimplicialLDLT 30 | * Note that these classes does not bring any particular advantage compared to the built-in 31 | * SimplicialLLT and SimplicialLDLT factorization classes. 32 | * 33 | * \code 34 | * #include 35 | * \endcode 36 | * 37 | * In order to use this module, the cholmod headers must be accessible from the include paths, and your binary must be linked to the cholmod library and its dependencies. 38 | * The dependencies depend on how cholmod has been compiled. 39 | * For a cmake based project, you can use our FindCholmod.cmake module to help you in this task. 40 | * 41 | */ 42 | 43 | #include "src/CholmodSupport/CholmodSupport.h" 44 | 45 | #include "src/Core/util/ReenableStupidWarnings.h" 46 | 47 | #endif // EIGEN_CHOLMODSUPPORT_MODULE_H 48 | 49 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/Dense: -------------------------------------------------------------------------------- 1 | #include "Core" 2 | #include "LU" 3 | #include "Cholesky" 4 | #include "QR" 5 | #include "SVD" 6 | #include "Geometry" 7 | #include "Eigenvalues" 8 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/Eigen: -------------------------------------------------------------------------------- 1 | #include "Dense" 2 | #include "Sparse" 3 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/Eigenvalues: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_EIGENVALUES_MODULE_H 9 | #define EIGEN_EIGENVALUES_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | #include "Cholesky" 16 | #include "Jacobi" 17 | #include "Householder" 18 | #include "LU" 19 | #include "Geometry" 20 | 21 | /** \defgroup Eigenvalues_Module Eigenvalues module 22 | * 23 | * 24 | * 25 | * This module mainly provides various eigenvalue solvers. 26 | * This module also provides some MatrixBase methods, including: 27 | * - MatrixBase::eigenvalues(), 28 | * - MatrixBase::operatorNorm() 29 | * 30 | * \code 31 | * #include 32 | * \endcode 33 | */ 34 | 35 | #include "src/misc/RealSvd2x2.h" 36 | #include "src/Eigenvalues/Tridiagonalization.h" 37 | #include "src/Eigenvalues/RealSchur.h" 38 | #include "src/Eigenvalues/EigenSolver.h" 39 | #include "src/Eigenvalues/SelfAdjointEigenSolver.h" 40 | #include "src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h" 41 | #include "src/Eigenvalues/HessenbergDecomposition.h" 42 | #include "src/Eigenvalues/ComplexSchur.h" 43 | #include "src/Eigenvalues/ComplexEigenSolver.h" 44 | #include "src/Eigenvalues/RealQZ.h" 45 | #include "src/Eigenvalues/GeneralizedEigenSolver.h" 46 | #include "src/Eigenvalues/MatrixBaseEigenvalues.h" 47 | #ifdef EIGEN_USE_LAPACKE 48 | #include "src/misc/lapacke.h" 49 | #include "src/Eigenvalues/RealSchur_LAPACKE.h" 50 | #include "src/Eigenvalues/ComplexSchur_LAPACKE.h" 51 | #include "src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h" 52 | #endif 53 | 54 | #include "src/Core/util/ReenableStupidWarnings.h" 55 | 56 | #endif // EIGEN_EIGENVALUES_MODULE_H 57 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 58 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/Geometry: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_GEOMETRY_MODULE_H 9 | #define EIGEN_GEOMETRY_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | #include "SVD" 16 | #include "LU" 17 | #include 18 | 19 | /** \defgroup Geometry_Module Geometry module 20 | * 21 | * This module provides support for: 22 | * - fixed-size homogeneous transformations 23 | * - translation, scaling, 2D and 3D rotations 24 | * - \link Quaternion quaternions \endlink 25 | * - cross products (\ref MatrixBase::cross, \ref MatrixBase::cross3) 26 | * - orthognal vector generation (\ref MatrixBase::unitOrthogonal) 27 | * - some linear components: \link ParametrizedLine parametrized-lines \endlink and \link Hyperplane hyperplanes \endlink 28 | * - \link AlignedBox axis aligned bounding boxes \endlink 29 | * - \link umeyama least-square transformation fitting \endlink 30 | * 31 | * \code 32 | * #include 33 | * \endcode 34 | */ 35 | 36 | #include "src/Geometry/OrthoMethods.h" 37 | #include "src/Geometry/EulerAngles.h" 38 | 39 | #include "src/Geometry/Homogeneous.h" 40 | #include "src/Geometry/RotationBase.h" 41 | #include "src/Geometry/Rotation2D.h" 42 | #include "src/Geometry/Quaternion.h" 43 | #include "src/Geometry/AngleAxis.h" 44 | #include "src/Geometry/Transform.h" 45 | #include "src/Geometry/Translation.h" 46 | #include "src/Geometry/Scaling.h" 47 | #include "src/Geometry/Hyperplane.h" 48 | #include "src/Geometry/ParametrizedLine.h" 49 | #include "src/Geometry/AlignedBox.h" 50 | #include "src/Geometry/Umeyama.h" 51 | 52 | // Use the SSE optimized version whenever possible. At the moment the 53 | // SSE version doesn't compile when AVX is enabled 54 | #if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX 55 | #include "src/Geometry/arch/Geometry_SSE.h" 56 | #endif 57 | 58 | #include "src/Core/util/ReenableStupidWarnings.h" 59 | 60 | #endif // EIGEN_GEOMETRY_MODULE_H 61 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 62 | 63 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/Householder: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_HOUSEHOLDER_MODULE_H 9 | #define EIGEN_HOUSEHOLDER_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | /** \defgroup Householder_Module Householder module 16 | * This module provides Householder transformations. 17 | * 18 | * \code 19 | * #include 20 | * \endcode 21 | */ 22 | 23 | #include "src/Householder/Householder.h" 24 | #include "src/Householder/HouseholderSequence.h" 25 | #include "src/Householder/BlockHouseholder.h" 26 | 27 | #include "src/Core/util/ReenableStupidWarnings.h" 28 | 29 | #endif // EIGEN_HOUSEHOLDER_MODULE_H 30 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 31 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/IterativeLinearSolvers: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_ITERATIVELINEARSOLVERS_MODULE_H 9 | #define EIGEN_ITERATIVELINEARSOLVERS_MODULE_H 10 | 11 | #include "SparseCore" 12 | #include "OrderingMethods" 13 | 14 | #include "src/Core/util/DisableStupidWarnings.h" 15 | 16 | /** 17 | * \defgroup IterativeLinearSolvers_Module IterativeLinearSolvers module 18 | * 19 | * This module currently provides iterative methods to solve problems of the form \c A \c x = \c b, where \c A is a squared matrix, usually very large and sparse. 20 | * Those solvers are accessible via the following classes: 21 | * - ConjugateGradient for selfadjoint (hermitian) matrices, 22 | * - LeastSquaresConjugateGradient for rectangular least-square problems, 23 | * - BiCGSTAB for general square matrices. 24 | * 25 | * These iterative solvers are associated with some preconditioners: 26 | * - IdentityPreconditioner - not really useful 27 | * - DiagonalPreconditioner - also called Jacobi preconditioner, work very well on diagonal dominant matrices. 28 | * - IncompleteLUT - incomplete LU factorization with dual thresholding 29 | * 30 | * Such problems can also be solved using the direct sparse decomposition modules: SparseCholesky, CholmodSupport, UmfPackSupport, SuperLUSupport. 31 | * 32 | \code 33 | #include 34 | \endcode 35 | */ 36 | 37 | #include "src/IterativeLinearSolvers/SolveWithGuess.h" 38 | #include "src/IterativeLinearSolvers/IterativeSolverBase.h" 39 | #include "src/IterativeLinearSolvers/BasicPreconditioners.h" 40 | #include "src/IterativeLinearSolvers/ConjugateGradient.h" 41 | #include "src/IterativeLinearSolvers/LeastSquareConjugateGradient.h" 42 | #include "src/IterativeLinearSolvers/BiCGSTAB.h" 43 | #include "src/IterativeLinearSolvers/IncompleteLUT.h" 44 | #include "src/IterativeLinearSolvers/IncompleteCholesky.h" 45 | 46 | #include "src/Core/util/ReenableStupidWarnings.h" 47 | 48 | #endif // EIGEN_ITERATIVELINEARSOLVERS_MODULE_H 49 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/Jacobi: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_JACOBI_MODULE_H 9 | #define EIGEN_JACOBI_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | /** \defgroup Jacobi_Module Jacobi module 16 | * This module provides Jacobi and Givens rotations. 17 | * 18 | * \code 19 | * #include 20 | * \endcode 21 | * 22 | * In addition to listed classes, it defines the two following MatrixBase methods to apply a Jacobi or Givens rotation: 23 | * - MatrixBase::applyOnTheLeft() 24 | * - MatrixBase::applyOnTheRight(). 25 | */ 26 | 27 | #include "src/Jacobi/Jacobi.h" 28 | 29 | #include "src/Core/util/ReenableStupidWarnings.h" 30 | 31 | #endif // EIGEN_JACOBI_MODULE_H 32 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 33 | 34 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/LU: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_LU_MODULE_H 9 | #define EIGEN_LU_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | /** \defgroup LU_Module LU module 16 | * This module includes %LU decomposition and related notions such as matrix inversion and determinant. 17 | * This module defines the following MatrixBase methods: 18 | * - MatrixBase::inverse() 19 | * - MatrixBase::determinant() 20 | * 21 | * \code 22 | * #include 23 | * \endcode 24 | */ 25 | 26 | #include "src/misc/Kernel.h" 27 | #include "src/misc/Image.h" 28 | #include "src/LU/FullPivLU.h" 29 | #include "src/LU/PartialPivLU.h" 30 | #ifdef EIGEN_USE_LAPACKE 31 | #include "src/misc/lapacke.h" 32 | #include "src/LU/PartialPivLU_LAPACKE.h" 33 | #endif 34 | #include "src/LU/Determinant.h" 35 | #include "src/LU/InverseImpl.h" 36 | 37 | // Use the SSE optimized version whenever possible. At the moment the 38 | // SSE version doesn't compile when AVX is enabled 39 | #if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX 40 | #include "src/LU/arch/Inverse_SSE.h" 41 | #endif 42 | 43 | #include "src/Core/util/ReenableStupidWarnings.h" 44 | 45 | #endif // EIGEN_LU_MODULE_H 46 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 47 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/MetisSupport: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_METISSUPPORT_MODULE_H 9 | #define EIGEN_METISSUPPORT_MODULE_H 10 | 11 | #include "SparseCore" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | extern "C" { 16 | #include 17 | } 18 | 19 | 20 | /** \ingroup Support_modules 21 | * \defgroup MetisSupport_Module MetisSupport module 22 | * 23 | * \code 24 | * #include 25 | * \endcode 26 | * This module defines an interface to the METIS reordering package (http://glaros.dtc.umn.edu/gkhome/views/metis). 27 | * It can be used just as any other built-in method as explained in \link OrderingMethods_Module here. \endlink 28 | */ 29 | 30 | 31 | #include "src/MetisSupport/MetisSupport.h" 32 | 33 | #include "src/Core/util/ReenableStupidWarnings.h" 34 | 35 | #endif // EIGEN_METISSUPPORT_MODULE_H 36 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/OrderingMethods: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_ORDERINGMETHODS_MODULE_H 9 | #define EIGEN_ORDERINGMETHODS_MODULE_H 10 | 11 | #include "SparseCore" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | /** 16 | * \defgroup OrderingMethods_Module OrderingMethods module 17 | * 18 | * This module is currently for internal use only 19 | * 20 | * It defines various built-in and external ordering methods for sparse matrices. 21 | * They are typically used to reduce the number of elements during 22 | * the sparse matrix decomposition (LLT, LU, QR). 23 | * Precisely, in a preprocessing step, a permutation matrix P is computed using 24 | * those ordering methods and applied to the columns of the matrix. 25 | * Using for instance the sparse Cholesky decomposition, it is expected that 26 | * the nonzeros elements in LLT(A*P) will be much smaller than that in LLT(A). 27 | * 28 | * 29 | * Usage : 30 | * \code 31 | * #include 32 | * \endcode 33 | * 34 | * A simple usage is as a template parameter in the sparse decomposition classes : 35 | * 36 | * \code 37 | * SparseLU > solver; 38 | * \endcode 39 | * 40 | * \code 41 | * SparseQR > solver; 42 | * \endcode 43 | * 44 | * It is possible as well to call directly a particular ordering method for your own purpose, 45 | * \code 46 | * AMDOrdering ordering; 47 | * PermutationMatrix perm; 48 | * SparseMatrix A; 49 | * //Fill the matrix ... 50 | * 51 | * ordering(A, perm); // Call AMD 52 | * \endcode 53 | * 54 | * \note Some of these methods (like AMD or METIS), need the sparsity pattern 55 | * of the input matrix to be symmetric. When the matrix is structurally unsymmetric, 56 | * Eigen computes internally the pattern of \f$A^T*A\f$ before calling the method. 57 | * If your matrix is already symmetric (at leat in structure), you can avoid that 58 | * by calling the method with a SelfAdjointView type. 59 | * 60 | * \code 61 | * // Call the ordering on the pattern of the lower triangular matrix A 62 | * ordering(A.selfadjointView(), perm); 63 | * \endcode 64 | */ 65 | 66 | #ifndef EIGEN_MPL2_ONLY 67 | #include "src/OrderingMethods/Amd.h" 68 | #endif 69 | 70 | #include "src/OrderingMethods/Ordering.h" 71 | #include "src/Core/util/ReenableStupidWarnings.h" 72 | 73 | #endif // EIGEN_ORDERINGMETHODS_MODULE_H 74 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/PaStiXSupport: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_PASTIXSUPPORT_MODULE_H 9 | #define EIGEN_PASTIXSUPPORT_MODULE_H 10 | 11 | #include "SparseCore" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | extern "C" { 16 | #include 17 | #include 18 | } 19 | 20 | #ifdef complex 21 | #undef complex 22 | #endif 23 | 24 | /** \ingroup Support_modules 25 | * \defgroup PaStiXSupport_Module PaStiXSupport module 26 | * 27 | * This module provides an interface to the PaSTiX library. 28 | * PaSTiX is a general \b supernodal, \b parallel and \b opensource sparse solver. 29 | * It provides the two following main factorization classes: 30 | * - class PastixLLT : a supernodal, parallel LLt Cholesky factorization. 31 | * - class PastixLDLT: a supernodal, parallel LDLt Cholesky factorization. 32 | * - class PastixLU : a supernodal, parallel LU factorization (optimized for a symmetric pattern). 33 | * 34 | * \code 35 | * #include 36 | * \endcode 37 | * 38 | * In order to use this module, the PaSTiX headers must be accessible from the include paths, and your binary must be linked to the PaSTiX library and its dependencies. 39 | * The dependencies depend on how PaSTiX has been compiled. 40 | * For a cmake based project, you can use our FindPaSTiX.cmake module to help you in this task. 41 | * 42 | */ 43 | 44 | #include "src/PaStiXSupport/PaStiXSupport.h" 45 | 46 | #include "src/Core/util/ReenableStupidWarnings.h" 47 | 48 | #endif // EIGEN_PASTIXSUPPORT_MODULE_H 49 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/PardisoSupport: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_PARDISOSUPPORT_MODULE_H 9 | #define EIGEN_PARDISOSUPPORT_MODULE_H 10 | 11 | #include "SparseCore" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | #include 16 | 17 | /** \ingroup Support_modules 18 | * \defgroup PardisoSupport_Module PardisoSupport module 19 | * 20 | * This module brings support for the Intel(R) MKL PARDISO direct sparse solvers. 21 | * 22 | * \code 23 | * #include 24 | * \endcode 25 | * 26 | * In order to use this module, the MKL headers must be accessible from the include paths, and your binary must be linked to the MKL library and its dependencies. 27 | * See this \ref TopicUsingIntelMKL "page" for more information on MKL-Eigen integration. 28 | * 29 | */ 30 | 31 | #include "src/PardisoSupport/PardisoSupport.h" 32 | 33 | #include "src/Core/util/ReenableStupidWarnings.h" 34 | 35 | #endif // EIGEN_PARDISOSUPPORT_MODULE_H 36 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/QR: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_QR_MODULE_H 9 | #define EIGEN_QR_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | #include "Cholesky" 16 | #include "Jacobi" 17 | #include "Householder" 18 | 19 | /** \defgroup QR_Module QR module 20 | * 21 | * 22 | * 23 | * This module provides various QR decompositions 24 | * This module also provides some MatrixBase methods, including: 25 | * - MatrixBase::householderQr() 26 | * - MatrixBase::colPivHouseholderQr() 27 | * - MatrixBase::fullPivHouseholderQr() 28 | * 29 | * \code 30 | * #include 31 | * \endcode 32 | */ 33 | 34 | #include "src/QR/HouseholderQR.h" 35 | #include "src/QR/FullPivHouseholderQR.h" 36 | #include "src/QR/ColPivHouseholderQR.h" 37 | #include "src/QR/CompleteOrthogonalDecomposition.h" 38 | #ifdef EIGEN_USE_LAPACKE 39 | #include "src/misc/lapacke.h" 40 | #include "src/QR/HouseholderQR_LAPACKE.h" 41 | #include "src/QR/ColPivHouseholderQR_LAPACKE.h" 42 | #endif 43 | 44 | #include "src/Core/util/ReenableStupidWarnings.h" 45 | 46 | #endif // EIGEN_QR_MODULE_H 47 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 48 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/QtAlignedMalloc: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_QTMALLOC_MODULE_H 9 | #define EIGEN_QTMALLOC_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #if (!EIGEN_MALLOC_ALREADY_ALIGNED) 14 | 15 | #include "src/Core/util/DisableStupidWarnings.h" 16 | 17 | void *qMalloc(std::size_t size) 18 | { 19 | return Eigen::internal::aligned_malloc(size); 20 | } 21 | 22 | void qFree(void *ptr) 23 | { 24 | Eigen::internal::aligned_free(ptr); 25 | } 26 | 27 | void *qRealloc(void *ptr, std::size_t size) 28 | { 29 | void* newPtr = Eigen::internal::aligned_malloc(size); 30 | memcpy(newPtr, ptr, size); 31 | Eigen::internal::aligned_free(ptr); 32 | return newPtr; 33 | } 34 | 35 | #include "src/Core/util/ReenableStupidWarnings.h" 36 | 37 | #endif 38 | 39 | #endif // EIGEN_QTMALLOC_MODULE_H 40 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 41 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/SPQRSupport: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_SPQRSUPPORT_MODULE_H 9 | #define EIGEN_SPQRSUPPORT_MODULE_H 10 | 11 | #include "SparseCore" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | #include "SuiteSparseQR.hpp" 16 | 17 | /** \ingroup Support_modules 18 | * \defgroup SPQRSupport_Module SuiteSparseQR module 19 | * 20 | * This module provides an interface to the SPQR library, which is part of the suitesparse package. 21 | * 22 | * \code 23 | * #include 24 | * \endcode 25 | * 26 | * In order to use this module, the SPQR headers must be accessible from the include paths, and your binary must be linked to the SPQR library and its dependencies (Cholmod, AMD, COLAMD,...). 27 | * For a cmake based project, you can use our FindSPQR.cmake and FindCholmod.Cmake modules 28 | * 29 | */ 30 | 31 | #include "src/CholmodSupport/CholmodSupport.h" 32 | #include "src/SPQRSupport/SuiteSparseQRSupport.h" 33 | 34 | #endif 35 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/SVD: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_SVD_MODULE_H 9 | #define EIGEN_SVD_MODULE_H 10 | 11 | #include "QR" 12 | #include "Householder" 13 | #include "Jacobi" 14 | 15 | #include "src/Core/util/DisableStupidWarnings.h" 16 | 17 | /** \defgroup SVD_Module SVD module 18 | * 19 | * 20 | * 21 | * This module provides SVD decomposition for matrices (both real and complex). 22 | * Two decomposition algorithms are provided: 23 | * - JacobiSVD implementing two-sided Jacobi iterations is numerically very accurate, fast for small matrices, but very slow for larger ones. 24 | * - BDCSVD implementing a recursive divide & conquer strategy on top of an upper-bidiagonalization which remains fast for large problems. 25 | * These decompositions are accessible via the respective classes and following MatrixBase methods: 26 | * - MatrixBase::jacobiSvd() 27 | * - MatrixBase::bdcSvd() 28 | * 29 | * \code 30 | * #include 31 | * \endcode 32 | */ 33 | 34 | #include "src/misc/RealSvd2x2.h" 35 | #include "src/SVD/UpperBidiagonalization.h" 36 | #include "src/SVD/SVDBase.h" 37 | #include "src/SVD/JacobiSVD.h" 38 | #include "src/SVD/BDCSVD.h" 39 | #if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT) 40 | #include "src/misc/lapacke.h" 41 | #include "src/SVD/JacobiSVD_LAPACKE.h" 42 | #endif 43 | 44 | #include "src/Core/util/ReenableStupidWarnings.h" 45 | 46 | #endif // EIGEN_SVD_MODULE_H 47 | /* vim: set filetype=cpp et sw=2 ts=2 ai: */ 48 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/Sparse: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_SPARSE_MODULE_H 9 | #define EIGEN_SPARSE_MODULE_H 10 | 11 | /** \defgroup Sparse_Module Sparse meta-module 12 | * 13 | * Meta-module including all related modules: 14 | * - \ref SparseCore_Module 15 | * - \ref OrderingMethods_Module 16 | * - \ref SparseCholesky_Module 17 | * - \ref SparseLU_Module 18 | * - \ref SparseQR_Module 19 | * - \ref IterativeLinearSolvers_Module 20 | * 21 | \code 22 | #include 23 | \endcode 24 | */ 25 | 26 | #include "SparseCore" 27 | #include "OrderingMethods" 28 | #ifndef EIGEN_MPL2_ONLY 29 | #include "SparseCholesky" 30 | #endif 31 | #include "SparseLU" 32 | #include "SparseQR" 33 | #include "IterativeLinearSolvers" 34 | 35 | #endif // EIGEN_SPARSE_MODULE_H 36 | 37 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/SparseCholesky: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008-2013 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_SPARSECHOLESKY_MODULE_H 11 | #define EIGEN_SPARSECHOLESKY_MODULE_H 12 | 13 | #include "SparseCore" 14 | #include "OrderingMethods" 15 | 16 | #include "src/Core/util/DisableStupidWarnings.h" 17 | 18 | /** 19 | * \defgroup SparseCholesky_Module SparseCholesky module 20 | * 21 | * This module currently provides two variants of the direct sparse Cholesky decomposition for selfadjoint (hermitian) matrices. 22 | * Those decompositions are accessible via the following classes: 23 | * - SimplicialLLt, 24 | * - SimplicialLDLt 25 | * 26 | * Such problems can also be solved using the ConjugateGradient solver from the IterativeLinearSolvers module. 27 | * 28 | * \code 29 | * #include 30 | * \endcode 31 | */ 32 | 33 | #ifdef EIGEN_MPL2_ONLY 34 | #error The SparseCholesky module has nothing to offer in MPL2 only mode 35 | #endif 36 | 37 | #include "src/SparseCholesky/SimplicialCholesky.h" 38 | 39 | #ifndef EIGEN_MPL2_ONLY 40 | #include "src/SparseCholesky/SimplicialCholesky_impl.h" 41 | #endif 42 | 43 | #include "src/Core/util/ReenableStupidWarnings.h" 44 | 45 | #endif // EIGEN_SPARSECHOLESKY_MODULE_H 46 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/SparseCore: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_SPARSECORE_MODULE_H 9 | #define EIGEN_SPARSECORE_MODULE_H 10 | 11 | #include "Core" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | #include 16 | #include 17 | #include 18 | #include 19 | #include 20 | 21 | /** 22 | * \defgroup SparseCore_Module SparseCore module 23 | * 24 | * This module provides a sparse matrix representation, and basic associated matrix manipulations 25 | * and operations. 26 | * 27 | * See the \ref TutorialSparse "Sparse tutorial" 28 | * 29 | * \code 30 | * #include 31 | * \endcode 32 | * 33 | * This module depends on: Core. 34 | */ 35 | 36 | #include "src/SparseCore/SparseUtil.h" 37 | #include "src/SparseCore/SparseMatrixBase.h" 38 | #include "src/SparseCore/SparseAssign.h" 39 | #include "src/SparseCore/CompressedStorage.h" 40 | #include "src/SparseCore/AmbiVector.h" 41 | #include "src/SparseCore/SparseCompressedBase.h" 42 | #include "src/SparseCore/SparseMatrix.h" 43 | #include "src/SparseCore/SparseMap.h" 44 | #include "src/SparseCore/MappedSparseMatrix.h" 45 | #include "src/SparseCore/SparseVector.h" 46 | #include "src/SparseCore/SparseRef.h" 47 | #include "src/SparseCore/SparseCwiseUnaryOp.h" 48 | #include "src/SparseCore/SparseCwiseBinaryOp.h" 49 | #include "src/SparseCore/SparseTranspose.h" 50 | #include "src/SparseCore/SparseBlock.h" 51 | #include "src/SparseCore/SparseDot.h" 52 | #include "src/SparseCore/SparseRedux.h" 53 | #include "src/SparseCore/SparseView.h" 54 | #include "src/SparseCore/SparseDiagonalProduct.h" 55 | #include "src/SparseCore/ConservativeSparseSparseProduct.h" 56 | #include "src/SparseCore/SparseSparseProductWithPruning.h" 57 | #include "src/SparseCore/SparseProduct.h" 58 | #include "src/SparseCore/SparseDenseProduct.h" 59 | #include "src/SparseCore/SparseSelfAdjointView.h" 60 | #include "src/SparseCore/SparseTriangularView.h" 61 | #include "src/SparseCore/TriangularSolver.h" 62 | #include "src/SparseCore/SparsePermutation.h" 63 | #include "src/SparseCore/SparseFuzzy.h" 64 | #include "src/SparseCore/SparseSolverBase.h" 65 | 66 | #include "src/Core/util/ReenableStupidWarnings.h" 67 | 68 | #endif // EIGEN_SPARSECORE_MODULE_H 69 | 70 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/SparseLU: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2012 Désiré Nuentsa-Wakam 5 | // Copyright (C) 2012 Gael Guennebaud 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_SPARSELU_MODULE_H 12 | #define EIGEN_SPARSELU_MODULE_H 13 | 14 | #include "SparseCore" 15 | 16 | /** 17 | * \defgroup SparseLU_Module SparseLU module 18 | * This module defines a supernodal factorization of general sparse matrices. 19 | * The code is fully optimized for supernode-panel updates with specialized kernels. 20 | * Please, see the documentation of the SparseLU class for more details. 21 | */ 22 | 23 | // Ordering interface 24 | #include "OrderingMethods" 25 | 26 | #include "src/SparseLU/SparseLU_gemm_kernel.h" 27 | 28 | #include "src/SparseLU/SparseLU_Structs.h" 29 | #include "src/SparseLU/SparseLU_SupernodalMatrix.h" 30 | #include "src/SparseLU/SparseLUImpl.h" 31 | #include "src/SparseCore/SparseColEtree.h" 32 | #include "src/SparseLU/SparseLU_Memory.h" 33 | #include "src/SparseLU/SparseLU_heap_relax_snode.h" 34 | #include "src/SparseLU/SparseLU_relax_snode.h" 35 | #include "src/SparseLU/SparseLU_pivotL.h" 36 | #include "src/SparseLU/SparseLU_panel_dfs.h" 37 | #include "src/SparseLU/SparseLU_kernel_bmod.h" 38 | #include "src/SparseLU/SparseLU_panel_bmod.h" 39 | #include "src/SparseLU/SparseLU_column_dfs.h" 40 | #include "src/SparseLU/SparseLU_column_bmod.h" 41 | #include "src/SparseLU/SparseLU_copy_to_ucol.h" 42 | #include "src/SparseLU/SparseLU_pruneL.h" 43 | #include "src/SparseLU/SparseLU_Utils.h" 44 | #include "src/SparseLU/SparseLU.h" 45 | 46 | #endif // EIGEN_SPARSELU_MODULE_H 47 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/SparseQR: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_SPARSEQR_MODULE_H 9 | #define EIGEN_SPARSEQR_MODULE_H 10 | 11 | #include "SparseCore" 12 | #include "OrderingMethods" 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | /** \defgroup SparseQR_Module SparseQR module 16 | * \brief Provides QR decomposition for sparse matrices 17 | * 18 | * This module provides a simplicial version of the left-looking Sparse QR decomposition. 19 | * The columns of the input matrix should be reordered to limit the fill-in during the 20 | * decomposition. Built-in methods (COLAMD, AMD) or external methods (METIS) can be used to this end. 21 | * See the \link OrderingMethods_Module OrderingMethods\endlink module for the list 22 | * of built-in and external ordering methods. 23 | * 24 | * \code 25 | * #include 26 | * \endcode 27 | * 28 | * 29 | */ 30 | 31 | #include "OrderingMethods" 32 | #include "src/SparseCore/SparseColEtree.h" 33 | #include "src/SparseQR/SparseQR.h" 34 | 35 | #include "src/Core/util/ReenableStupidWarnings.h" 36 | 37 | #endif 38 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/StdDeque: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009 Gael Guennebaud 5 | // Copyright (C) 2009 Hauke Heibel 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_STDDEQUE_MODULE_H 12 | #define EIGEN_STDDEQUE_MODULE_H 13 | 14 | #include "Core" 15 | #include 16 | 17 | #if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */ 18 | 19 | #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...) 20 | 21 | #else 22 | 23 | #include "src/StlSupport/StdDeque.h" 24 | 25 | #endif 26 | 27 | #endif // EIGEN_STDDEQUE_MODULE_H 28 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/StdList: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009 Hauke Heibel 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_STDLIST_MODULE_H 11 | #define EIGEN_STDLIST_MODULE_H 12 | 13 | #include "Core" 14 | #include 15 | 16 | #if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */ 17 | 18 | #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...) 19 | 20 | #else 21 | 22 | #include "src/StlSupport/StdList.h" 23 | 24 | #endif 25 | 26 | #endif // EIGEN_STDLIST_MODULE_H 27 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/StdVector: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009 Gael Guennebaud 5 | // Copyright (C) 2009 Hauke Heibel 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_STDVECTOR_MODULE_H 12 | #define EIGEN_STDVECTOR_MODULE_H 13 | 14 | #include "Core" 15 | #include 16 | 17 | #if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */ 18 | 19 | #define EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION(...) 20 | 21 | #else 22 | 23 | #include "src/StlSupport/StdVector.h" 24 | 25 | #endif 26 | 27 | #endif // EIGEN_STDVECTOR_MODULE_H 28 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/SuperLUSupport: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_SUPERLUSUPPORT_MODULE_H 9 | #define EIGEN_SUPERLUSUPPORT_MODULE_H 10 | 11 | #include "SparseCore" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | #ifdef EMPTY 16 | #define EIGEN_EMPTY_WAS_ALREADY_DEFINED 17 | #endif 18 | 19 | typedef int int_t; 20 | #include 21 | #include 22 | #include 23 | 24 | // slu_util.h defines a preprocessor token named EMPTY which is really polluting, 25 | // so we remove it in favor of a SUPERLU_EMPTY token. 26 | // If EMPTY was already defined then we don't undef it. 27 | 28 | #if defined(EIGEN_EMPTY_WAS_ALREADY_DEFINED) 29 | # undef EIGEN_EMPTY_WAS_ALREADY_DEFINED 30 | #elif defined(EMPTY) 31 | # undef EMPTY 32 | #endif 33 | 34 | #define SUPERLU_EMPTY (-1) 35 | 36 | namespace Eigen { struct SluMatrix; } 37 | 38 | /** \ingroup Support_modules 39 | * \defgroup SuperLUSupport_Module SuperLUSupport module 40 | * 41 | * This module provides an interface to the SuperLU library. 42 | * It provides the following factorization class: 43 | * - class SuperLU: a supernodal sequential LU factorization. 44 | * - class SuperILU: a supernodal sequential incomplete LU factorization (to be used as a preconditioner for iterative methods). 45 | * 46 | * \warning This wrapper requires at least versions 4.0 of SuperLU. The 3.x versions are not supported. 47 | * 48 | * \warning When including this module, you have to use SUPERLU_EMPTY instead of EMPTY which is no longer defined because it is too polluting. 49 | * 50 | * \code 51 | * #include 52 | * \endcode 53 | * 54 | * In order to use this module, the superlu headers must be accessible from the include paths, and your binary must be linked to the superlu library and its dependencies. 55 | * The dependencies depend on how superlu has been compiled. 56 | * For a cmake based project, you can use our FindSuperLU.cmake module to help you in this task. 57 | * 58 | */ 59 | 60 | #include "src/SuperLUSupport/SuperLUSupport.h" 61 | 62 | #include "src/Core/util/ReenableStupidWarnings.h" 63 | 64 | #endif // EIGEN_SUPERLUSUPPORT_MODULE_H 65 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/UmfPackSupport: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | #ifndef EIGEN_UMFPACKSUPPORT_MODULE_H 9 | #define EIGEN_UMFPACKSUPPORT_MODULE_H 10 | 11 | #include "SparseCore" 12 | 13 | #include "src/Core/util/DisableStupidWarnings.h" 14 | 15 | extern "C" { 16 | #include 17 | } 18 | 19 | /** \ingroup Support_modules 20 | * \defgroup UmfPackSupport_Module UmfPackSupport module 21 | * 22 | * This module provides an interface to the UmfPack library which is part of the suitesparse package. 23 | * It provides the following factorization class: 24 | * - class UmfPackLU: a multifrontal sequential LU factorization. 25 | * 26 | * \code 27 | * #include 28 | * \endcode 29 | * 30 | * In order to use this module, the umfpack headers must be accessible from the include paths, and your binary must be linked to the umfpack library and its dependencies. 31 | * The dependencies depend on how umfpack has been compiled. 32 | * For a cmake based project, you can use our FindUmfPack.cmake module to help you in this task. 33 | * 34 | */ 35 | 36 | #include "src/UmfPackSupport/UmfPackSupport.h" 37 | 38 | #include "src/Core/util/ReenableStupidWarnings.h" 39 | 40 | #endif // EIGEN_UMFPACKSUPPORT_MODULE_H 41 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/Assign.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2007 Michael Olbrich 5 | // Copyright (C) 2006-2010 Benoit Jacob 6 | // Copyright (C) 2008 Gael Guennebaud 7 | // 8 | // This Source Code Form is subject to the terms of the Mozilla 9 | // Public License v. 2.0. If a copy of the MPL was not distributed 10 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 11 | 12 | #ifndef EIGEN_ASSIGN_H 13 | #define EIGEN_ASSIGN_H 14 | 15 | namespace Eigen { 16 | 17 | template 18 | template 19 | EIGEN_STRONG_INLINE Derived& DenseBase 20 | ::lazyAssign(const DenseBase& other) 21 | { 22 | enum{ 23 | SameType = internal::is_same::value 24 | }; 25 | 26 | EIGEN_STATIC_ASSERT_LVALUE(Derived) 27 | EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived) 28 | EIGEN_STATIC_ASSERT(SameType,YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) 29 | 30 | eigen_assert(rows() == other.rows() && cols() == other.cols()); 31 | internal::call_assignment_no_alias(derived(),other.derived()); 32 | 33 | return derived(); 34 | } 35 | 36 | template 37 | template 38 | EIGEN_DEVICE_FUNC 39 | EIGEN_STRONG_INLINE Derived& DenseBase::operator=(const DenseBase& other) 40 | { 41 | internal::call_assignment(derived(), other.derived()); 42 | return derived(); 43 | } 44 | 45 | template 46 | EIGEN_DEVICE_FUNC 47 | EIGEN_STRONG_INLINE Derived& DenseBase::operator=(const DenseBase& other) 48 | { 49 | internal::call_assignment(derived(), other.derived()); 50 | return derived(); 51 | } 52 | 53 | template 54 | EIGEN_DEVICE_FUNC 55 | EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const MatrixBase& other) 56 | { 57 | internal::call_assignment(derived(), other.derived()); 58 | return derived(); 59 | } 60 | 61 | template 62 | template 63 | EIGEN_DEVICE_FUNC 64 | EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const DenseBase& other) 65 | { 66 | internal::call_assignment(derived(), other.derived()); 67 | return derived(); 68 | } 69 | 70 | template 71 | template 72 | EIGEN_DEVICE_FUNC 73 | EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const EigenBase& other) 74 | { 75 | internal::call_assignment(derived(), other.derived()); 76 | return derived(); 77 | } 78 | 79 | template 80 | template 81 | EIGEN_DEVICE_FUNC 82 | EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const ReturnByValue& other) 83 | { 84 | other.derived().evalTo(derived()); 85 | return derived(); 86 | } 87 | 88 | } // end namespace Eigen 89 | 90 | #endif // EIGEN_ASSIGN_H 91 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/DiagonalProduct.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008 Gael Guennebaud 5 | // Copyright (C) 2007-2009 Benoit Jacob 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_DIAGONALPRODUCT_H 12 | #define EIGEN_DIAGONALPRODUCT_H 13 | 14 | namespace Eigen { 15 | 16 | /** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal. 17 | */ 18 | template 19 | template 20 | inline const Product 21 | MatrixBase::operator*(const DiagonalBase &a_diagonal) const 22 | { 23 | return Product(derived(),a_diagonal.derived()); 24 | } 25 | 26 | } // end namespace Eigen 27 | 28 | #endif // EIGEN_DIAGONALPRODUCT_H 29 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/Inverse.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2014 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_INVERSE_H 11 | #define EIGEN_INVERSE_H 12 | 13 | namespace Eigen { 14 | 15 | template class InverseImpl; 16 | 17 | namespace internal { 18 | 19 | template 20 | struct traits > 21 | : traits 22 | { 23 | typedef typename XprType::PlainObject PlainObject; 24 | typedef traits BaseTraits; 25 | enum { 26 | Flags = BaseTraits::Flags & RowMajorBit 27 | }; 28 | }; 29 | 30 | } // end namespace internal 31 | 32 | /** \class Inverse 33 | * 34 | * \brief Expression of the inverse of another expression 35 | * 36 | * \tparam XprType the type of the expression we are taking the inverse 37 | * 38 | * This class represents an abstract expression of A.inverse() 39 | * and most of the time this is the only way it is used. 40 | * 41 | */ 42 | template 43 | class Inverse : public InverseImpl::StorageKind> 44 | { 45 | public: 46 | typedef typename XprType::StorageIndex StorageIndex; 47 | typedef typename XprType::PlainObject PlainObject; 48 | typedef typename XprType::Scalar Scalar; 49 | typedef typename internal::ref_selector::type XprTypeNested; 50 | typedef typename internal::remove_all::type XprTypeNestedCleaned; 51 | typedef typename internal::ref_selector::type Nested; 52 | typedef typename internal::remove_all::type NestedExpression; 53 | 54 | explicit EIGEN_DEVICE_FUNC Inverse(const XprType &xpr) 55 | : m_xpr(xpr) 56 | {} 57 | 58 | EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); } 59 | EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); } 60 | 61 | EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; } 62 | 63 | protected: 64 | XprTypeNested m_xpr; 65 | }; 66 | 67 | // Generic API dispatcher 68 | template 69 | class InverseImpl 70 | : public internal::generic_xpr_base >::type 71 | { 72 | public: 73 | typedef typename internal::generic_xpr_base >::type Base; 74 | typedef typename XprType::Scalar Scalar; 75 | private: 76 | 77 | Scalar coeff(Index row, Index col) const; 78 | Scalar coeff(Index i) const; 79 | }; 80 | 81 | namespace internal { 82 | 83 | /** \internal 84 | * \brief Default evaluator for Inverse expression. 85 | * 86 | * This default evaluator for Inverse expression simply evaluate the inverse into a temporary 87 | * by a call to internal::call_assignment_no_alias. 88 | * Therefore, inverse implementers only have to specialize Assignment, ...> for 89 | * there own nested expression. 90 | * 91 | * \sa class Inverse 92 | */ 93 | template 94 | struct unary_evaluator > 95 | : public evaluator::PlainObject> 96 | { 97 | typedef Inverse InverseType; 98 | typedef typename InverseType::PlainObject PlainObject; 99 | typedef evaluator Base; 100 | 101 | enum { Flags = Base::Flags | EvalBeforeNestingBit }; 102 | 103 | unary_evaluator(const InverseType& inv_xpr) 104 | : m_result(inv_xpr.rows(), inv_xpr.cols()) 105 | { 106 | ::new (static_cast(this)) Base(m_result); 107 | internal::call_assignment_no_alias(m_result, inv_xpr); 108 | } 109 | 110 | protected: 111 | PlainObject m_result; 112 | }; 113 | 114 | } // end namespace internal 115 | 116 | } // end namespace Eigen 117 | 118 | #endif // EIGEN_INVERSE_H 119 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/MathFunctionsImpl.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com) 5 | // Copyright (C) 2016 Gael Guennebaud 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_MATHFUNCTIONSIMPL_H 12 | #define EIGEN_MATHFUNCTIONSIMPL_H 13 | 14 | namespace Eigen { 15 | 16 | namespace internal { 17 | 18 | /** \internal \returns the hyperbolic tan of \a a (coeff-wise) 19 | Doesn't do anything fancy, just a 13/6-degree rational interpolant which 20 | is accurate up to a couple of ulp in the range [-9, 9], outside of which 21 | the tanh(x) = +/-1. 22 | 23 | This implementation works on both scalars and packets. 24 | */ 25 | template 26 | T generic_fast_tanh_float(const T& a_x) 27 | { 28 | // Clamp the inputs to the range [-9, 9] since anything outside 29 | // this range is +/-1.0f in single-precision. 30 | const T plus_9 = pset1(9.f); 31 | const T minus_9 = pset1(-9.f); 32 | // NOTE GCC prior to 6.3 might improperly optimize this max/min 33 | // step such that if a_x is nan, x will be either 9 or -9, 34 | // and tanh will return 1 or -1 instead of nan. 35 | // This is supposed to be fixed in gcc6.3, 36 | // see: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72867 37 | const T x = pmax(minus_9,pmin(plus_9,a_x)); 38 | // The monomial coefficients of the numerator polynomial (odd). 39 | const T alpha_1 = pset1(4.89352455891786e-03f); 40 | const T alpha_3 = pset1(6.37261928875436e-04f); 41 | const T alpha_5 = pset1(1.48572235717979e-05f); 42 | const T alpha_7 = pset1(5.12229709037114e-08f); 43 | const T alpha_9 = pset1(-8.60467152213735e-11f); 44 | const T alpha_11 = pset1(2.00018790482477e-13f); 45 | const T alpha_13 = pset1(-2.76076847742355e-16f); 46 | 47 | // The monomial coefficients of the denominator polynomial (even). 48 | const T beta_0 = pset1(4.89352518554385e-03f); 49 | const T beta_2 = pset1(2.26843463243900e-03f); 50 | const T beta_4 = pset1(1.18534705686654e-04f); 51 | const T beta_6 = pset1(1.19825839466702e-06f); 52 | 53 | // Since the polynomials are odd/even, we need x^2. 54 | const T x2 = pmul(x, x); 55 | 56 | // Evaluate the numerator polynomial p. 57 | T p = pmadd(x2, alpha_13, alpha_11); 58 | p = pmadd(x2, p, alpha_9); 59 | p = pmadd(x2, p, alpha_7); 60 | p = pmadd(x2, p, alpha_5); 61 | p = pmadd(x2, p, alpha_3); 62 | p = pmadd(x2, p, alpha_1); 63 | p = pmul(x, p); 64 | 65 | // Evaluate the denominator polynomial p. 66 | T q = pmadd(x2, beta_6, beta_4); 67 | q = pmadd(x2, q, beta_2); 68 | q = pmadd(x2, q, beta_0); 69 | 70 | // Divide the numerator by the denominator. 71 | return pdiv(p, q); 72 | } 73 | 74 | } // end namespace internal 75 | 76 | } // end namespace Eigen 77 | 78 | #endif // EIGEN_MATHFUNCTIONSIMPL_H 79 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/NestByValue.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008 Gael Guennebaud 5 | // Copyright (C) 2006-2008 Benoit Jacob 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_NESTBYVALUE_H 12 | #define EIGEN_NESTBYVALUE_H 13 | 14 | namespace Eigen { 15 | 16 | namespace internal { 17 | template 18 | struct traits > : public traits 19 | {}; 20 | } 21 | 22 | /** \class NestByValue 23 | * \ingroup Core_Module 24 | * 25 | * \brief Expression which must be nested by value 26 | * 27 | * \tparam ExpressionType the type of the object of which we are requiring nesting-by-value 28 | * 29 | * This class is the return type of MatrixBase::nestByValue() 30 | * and most of the time this is the only way it is used. 31 | * 32 | * \sa MatrixBase::nestByValue() 33 | */ 34 | template class NestByValue 35 | : public internal::dense_xpr_base< NestByValue >::type 36 | { 37 | public: 38 | 39 | typedef typename internal::dense_xpr_base::type Base; 40 | EIGEN_DENSE_PUBLIC_INTERFACE(NestByValue) 41 | 42 | EIGEN_DEVICE_FUNC explicit inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {} 43 | 44 | EIGEN_DEVICE_FUNC inline Index rows() const { return m_expression.rows(); } 45 | EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); } 46 | EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); } 47 | EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); } 48 | 49 | EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const 50 | { 51 | return m_expression.coeff(row, col); 52 | } 53 | 54 | EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col) 55 | { 56 | return m_expression.const_cast_derived().coeffRef(row, col); 57 | } 58 | 59 | EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const 60 | { 61 | return m_expression.coeff(index); 62 | } 63 | 64 | EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index) 65 | { 66 | return m_expression.const_cast_derived().coeffRef(index); 67 | } 68 | 69 | template 70 | inline const PacketScalar packet(Index row, Index col) const 71 | { 72 | return m_expression.template packet(row, col); 73 | } 74 | 75 | template 76 | inline void writePacket(Index row, Index col, const PacketScalar& x) 77 | { 78 | m_expression.const_cast_derived().template writePacket(row, col, x); 79 | } 80 | 81 | template 82 | inline const PacketScalar packet(Index index) const 83 | { 84 | return m_expression.template packet(index); 85 | } 86 | 87 | template 88 | inline void writePacket(Index index, const PacketScalar& x) 89 | { 90 | m_expression.const_cast_derived().template writePacket(index, x); 91 | } 92 | 93 | EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; } 94 | 95 | protected: 96 | const ExpressionType m_expression; 97 | }; 98 | 99 | /** \returns an expression of the temporary version of *this. 100 | */ 101 | template 102 | inline const NestByValue 103 | DenseBase::nestByValue() const 104 | { 105 | return NestByValue(derived()); 106 | } 107 | 108 | } // end namespace Eigen 109 | 110 | #endif // EIGEN_NESTBYVALUE_H 111 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/NoAlias.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_NOALIAS_H 11 | #define EIGEN_NOALIAS_H 12 | 13 | namespace Eigen { 14 | 15 | /** \class NoAlias 16 | * \ingroup Core_Module 17 | * 18 | * \brief Pseudo expression providing an operator = assuming no aliasing 19 | * 20 | * \tparam ExpressionType the type of the object on which to do the lazy assignment 21 | * 22 | * This class represents an expression with special assignment operators 23 | * assuming no aliasing between the target expression and the source expression. 24 | * More precisely it alloas to bypass the EvalBeforeAssignBit flag of the source expression. 25 | * It is the return type of MatrixBase::noalias() 26 | * and most of the time this is the only way it is used. 27 | * 28 | * \sa MatrixBase::noalias() 29 | */ 30 | template class StorageBase> 31 | class NoAlias 32 | { 33 | public: 34 | typedef typename ExpressionType::Scalar Scalar; 35 | 36 | explicit NoAlias(ExpressionType& expression) : m_expression(expression) {} 37 | 38 | template 39 | EIGEN_DEVICE_FUNC 40 | EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase& other) 41 | { 42 | call_assignment_no_alias(m_expression, other.derived(), internal::assign_op()); 43 | return m_expression; 44 | } 45 | 46 | template 47 | EIGEN_DEVICE_FUNC 48 | EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase& other) 49 | { 50 | call_assignment_no_alias(m_expression, other.derived(), internal::add_assign_op()); 51 | return m_expression; 52 | } 53 | 54 | template 55 | EIGEN_DEVICE_FUNC 56 | EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase& other) 57 | { 58 | call_assignment_no_alias(m_expression, other.derived(), internal::sub_assign_op()); 59 | return m_expression; 60 | } 61 | 62 | EIGEN_DEVICE_FUNC 63 | ExpressionType& expression() const 64 | { 65 | return m_expression; 66 | } 67 | 68 | protected: 69 | ExpressionType& m_expression; 70 | }; 71 | 72 | /** \returns a pseudo expression of \c *this with an operator= assuming 73 | * no aliasing between \c *this and the source expression. 74 | * 75 | * More precisely, noalias() allows to bypass the EvalBeforeAssignBit flag. 76 | * Currently, even though several expressions may alias, only product 77 | * expressions have this flag. Therefore, noalias() is only usefull when 78 | * the source expression contains a matrix product. 79 | * 80 | * Here are some examples where noalias is usefull: 81 | * \code 82 | * D.noalias() = A * B; 83 | * D.noalias() += A.transpose() * B; 84 | * D.noalias() -= 2 * A * B.adjoint(); 85 | * \endcode 86 | * 87 | * On the other hand the following example will lead to a \b wrong result: 88 | * \code 89 | * A.noalias() = A * B; 90 | * \endcode 91 | * because the result matrix A is also an operand of the matrix product. Therefore, 92 | * there is no alternative than evaluating A * B in a temporary, that is the default 93 | * behavior when you write: 94 | * \code 95 | * A = A * B; 96 | * \endcode 97 | * 98 | * \sa class NoAlias 99 | */ 100 | template 101 | NoAlias MatrixBase::noalias() 102 | { 103 | return NoAlias(derived()); 104 | } 105 | 106 | } // end namespace Eigen 107 | 108 | #endif // EIGEN_NOALIAS_H 109 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/SelfCwiseBinaryOp.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009-2010 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_SELFCWISEBINARYOP_H 11 | #define EIGEN_SELFCWISEBINARYOP_H 12 | 13 | namespace Eigen { 14 | 15 | // TODO generalize the scalar type of 'other' 16 | 17 | template 18 | EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::operator*=(const Scalar& other) 19 | { 20 | typedef typename Derived::PlainObject PlainObject; 21 | internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::mul_assign_op()); 22 | return derived(); 23 | } 24 | 25 | template 26 | EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase::operator+=(const Scalar& other) 27 | { 28 | typedef typename Derived::PlainObject PlainObject; 29 | internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::add_assign_op()); 30 | return derived(); 31 | } 32 | 33 | template 34 | EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase::operator-=(const Scalar& other) 35 | { 36 | typedef typename Derived::PlainObject PlainObject; 37 | internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::sub_assign_op()); 38 | return derived(); 39 | } 40 | 41 | template 42 | EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::operator/=(const Scalar& other) 43 | { 44 | typedef typename Derived::PlainObject PlainObject; 45 | internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::div_assign_op()); 46 | return derived(); 47 | } 48 | 49 | } // end namespace Eigen 50 | 51 | #endif // EIGEN_SELFCWISEBINARYOP_H 52 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/Swap.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2006-2008 Benoit Jacob 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_SWAP_H 11 | #define EIGEN_SWAP_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | 17 | // Overload default assignPacket behavior for swapping them 18 | template 19 | class generic_dense_assignment_kernel, Specialized> 20 | : public generic_dense_assignment_kernel, BuiltIn> 21 | { 22 | protected: 23 | typedef generic_dense_assignment_kernel, BuiltIn> Base; 24 | using Base::m_dst; 25 | using Base::m_src; 26 | using Base::m_functor; 27 | 28 | public: 29 | typedef typename Base::Scalar Scalar; 30 | typedef typename Base::DstXprType DstXprType; 31 | typedef swap_assign_op Functor; 32 | 33 | EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr) 34 | : Base(dst, src, func, dstExpr) 35 | {} 36 | 37 | template 38 | void assignPacket(Index row, Index col) 39 | { 40 | PacketType tmp = m_src.template packet(row,col); 41 | const_cast(m_src).template writePacket(row,col, m_dst.template packet(row,col)); 42 | m_dst.template writePacket(row,col,tmp); 43 | } 44 | 45 | template 46 | void assignPacket(Index index) 47 | { 48 | PacketType tmp = m_src.template packet(index); 49 | const_cast(m_src).template writePacket(index, m_dst.template packet(index)); 50 | m_dst.template writePacket(index,tmp); 51 | } 52 | 53 | // TODO find a simple way not to have to copy/paste this function from generic_dense_assignment_kernel, by simple I mean no CRTP (Gael) 54 | template 55 | void assignPacketByOuterInner(Index outer, Index inner) 56 | { 57 | Index row = Base::rowIndexByOuterInner(outer, inner); 58 | Index col = Base::colIndexByOuterInner(outer, inner); 59 | assignPacket(row, col); 60 | } 61 | }; 62 | 63 | } // namespace internal 64 | 65 | } // end namespace Eigen 66 | 67 | #endif // EIGEN_SWAP_H 68 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/VectorBlock.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008-2010 Gael Guennebaud 5 | // Copyright (C) 2006-2008 Benoit Jacob 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_VECTORBLOCK_H 12 | #define EIGEN_VECTORBLOCK_H 13 | 14 | namespace Eigen { 15 | 16 | namespace internal { 17 | template 18 | struct traits > 19 | : public traits::Flags & RowMajorBit ? 1 : Size, 21 | traits::Flags & RowMajorBit ? Size : 1> > 22 | { 23 | }; 24 | } 25 | 26 | /** \class VectorBlock 27 | * \ingroup Core_Module 28 | * 29 | * \brief Expression of a fixed-size or dynamic-size sub-vector 30 | * 31 | * \tparam VectorType the type of the object in which we are taking a sub-vector 32 | * \tparam Size size of the sub-vector we are taking at compile time (optional) 33 | * 34 | * This class represents an expression of either a fixed-size or dynamic-size sub-vector. 35 | * It is the return type of DenseBase::segment(Index,Index) and DenseBase::segment(Index) and 36 | * most of the time this is the only way it is used. 37 | * 38 | * However, if you want to directly maniputate sub-vector expressions, 39 | * for instance if you want to write a function returning such an expression, you 40 | * will need to use this class. 41 | * 42 | * Here is an example illustrating the dynamic case: 43 | * \include class_VectorBlock.cpp 44 | * Output: \verbinclude class_VectorBlock.out 45 | * 46 | * \note Even though this expression has dynamic size, in the case where \a VectorType 47 | * has fixed size, this expression inherits a fixed maximal size which means that evaluating 48 | * it does not cause a dynamic memory allocation. 49 | * 50 | * Here is an example illustrating the fixed-size case: 51 | * \include class_FixedVectorBlock.cpp 52 | * Output: \verbinclude class_FixedVectorBlock.out 53 | * 54 | * \sa class Block, DenseBase::segment(Index,Index,Index,Index), DenseBase::segment(Index,Index) 55 | */ 56 | template class VectorBlock 57 | : public Block::Flags & RowMajorBit ? 1 : Size, 59 | internal::traits::Flags & RowMajorBit ? Size : 1> 60 | { 61 | typedef Block::Flags & RowMajorBit ? 1 : Size, 63 | internal::traits::Flags & RowMajorBit ? Size : 1> Base; 64 | enum { 65 | IsColVector = !(internal::traits::Flags & RowMajorBit) 66 | }; 67 | public: 68 | EIGEN_DENSE_PUBLIC_INTERFACE(VectorBlock) 69 | 70 | using Base::operator=; 71 | 72 | /** Dynamic-size constructor 73 | */ 74 | EIGEN_DEVICE_FUNC 75 | inline VectorBlock(VectorType& vector, Index start, Index size) 76 | : Base(vector, 77 | IsColVector ? start : 0, IsColVector ? 0 : start, 78 | IsColVector ? size : 1, IsColVector ? 1 : size) 79 | { 80 | EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorBlock); 81 | } 82 | 83 | /** Fixed-size constructor 84 | */ 85 | EIGEN_DEVICE_FUNC 86 | inline VectorBlock(VectorType& vector, Index start) 87 | : Base(vector, IsColVector ? start : 0, IsColVector ? 0 : start) 88 | { 89 | EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorBlock); 90 | } 91 | }; 92 | 93 | 94 | } // end namespace Eigen 95 | 96 | #endif // EIGEN_VECTORBLOCK_H 97 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/arch/AVX/TypeCasting.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2015 Benoit Steiner 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_TYPE_CASTING_AVX_H 11 | #define EIGEN_TYPE_CASTING_AVX_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | 17 | // For now we use SSE to handle integers, so we can't use AVX instructions to cast 18 | // from int to float 19 | template <> 20 | struct type_casting_traits { 21 | enum { 22 | VectorizedCast = 0, 23 | SrcCoeffRatio = 1, 24 | TgtCoeffRatio = 1 25 | }; 26 | }; 27 | 28 | template <> 29 | struct type_casting_traits { 30 | enum { 31 | VectorizedCast = 0, 32 | SrcCoeffRatio = 1, 33 | TgtCoeffRatio = 1 34 | }; 35 | }; 36 | 37 | 38 | 39 | template<> EIGEN_STRONG_INLINE Packet8i pcast(const Packet8f& a) { 40 | return _mm256_cvtps_epi32(a); 41 | } 42 | 43 | template<> EIGEN_STRONG_INLINE Packet8f pcast(const Packet8i& a) { 44 | return _mm256_cvtepi32_ps(a); 45 | } 46 | 47 | } // end namespace internal 48 | 49 | } // end namespace Eigen 50 | 51 | #endif // EIGEN_TYPE_CASTING_AVX_H 52 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/arch/CUDA/MathFunctions.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2014 Benoit Steiner 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_MATH_FUNCTIONS_CUDA_H 11 | #define EIGEN_MATH_FUNCTIONS_CUDA_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | 17 | // Make sure this is only available when targeting a GPU: we don't want to 18 | // introduce conflicts between these packet_traits definitions and the ones 19 | // we'll use on the host side (SSE, AVX, ...) 20 | #if defined(__CUDACC__) && defined(EIGEN_USE_GPU) 21 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 22 | float4 plog(const float4& a) 23 | { 24 | return make_float4(logf(a.x), logf(a.y), logf(a.z), logf(a.w)); 25 | } 26 | 27 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 28 | double2 plog(const double2& a) 29 | { 30 | using ::log; 31 | return make_double2(log(a.x), log(a.y)); 32 | } 33 | 34 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 35 | float4 plog1p(const float4& a) 36 | { 37 | return make_float4(log1pf(a.x), log1pf(a.y), log1pf(a.z), log1pf(a.w)); 38 | } 39 | 40 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 41 | double2 plog1p(const double2& a) 42 | { 43 | return make_double2(log1p(a.x), log1p(a.y)); 44 | } 45 | 46 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 47 | float4 pexp(const float4& a) 48 | { 49 | return make_float4(expf(a.x), expf(a.y), expf(a.z), expf(a.w)); 50 | } 51 | 52 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 53 | double2 pexp(const double2& a) 54 | { 55 | using ::exp; 56 | return make_double2(exp(a.x), exp(a.y)); 57 | } 58 | 59 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 60 | float4 psqrt(const float4& a) 61 | { 62 | return make_float4(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z), sqrtf(a.w)); 63 | } 64 | 65 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 66 | double2 psqrt(const double2& a) 67 | { 68 | using ::sqrt; 69 | return make_double2(sqrt(a.x), sqrt(a.y)); 70 | } 71 | 72 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 73 | float4 prsqrt(const float4& a) 74 | { 75 | return make_float4(rsqrtf(a.x), rsqrtf(a.y), rsqrtf(a.z), rsqrtf(a.w)); 76 | } 77 | 78 | template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE 79 | double2 prsqrt(const double2& a) 80 | { 81 | return make_double2(rsqrt(a.x), rsqrt(a.y)); 82 | } 83 | 84 | 85 | #endif 86 | 87 | } // end namespace internal 88 | 89 | } // end namespace Eigen 90 | 91 | #endif // EIGEN_MATH_FUNCTIONS_CUDA_H 92 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/arch/Default/Settings.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008-2010 Gael Guennebaud 5 | // Copyright (C) 2006-2008 Benoit Jacob 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | 12 | /* All the parameters defined in this file can be specialized in the 13 | * architecture specific files, and/or by the user. 14 | * More to come... */ 15 | 16 | #ifndef EIGEN_DEFAULT_SETTINGS_H 17 | #define EIGEN_DEFAULT_SETTINGS_H 18 | 19 | /** Defines the maximal loop size to enable meta unrolling of loops. 20 | * Note that the value here is expressed in Eigen's own notion of "number of FLOPS", 21 | * it does not correspond to the number of iterations or the number of instructions 22 | */ 23 | #ifndef EIGEN_UNROLLING_LIMIT 24 | #define EIGEN_UNROLLING_LIMIT 100 25 | #endif 26 | 27 | /** Defines the threshold between a "small" and a "large" matrix. 28 | * This threshold is mainly used to select the proper product implementation. 29 | */ 30 | #ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 31 | #define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 8 32 | #endif 33 | 34 | /** Defines the maximal width of the blocks used in the triangular product and solver 35 | * for vectors (level 2 blas xTRMV and xTRSV). The default is 8. 36 | */ 37 | #ifndef EIGEN_TUNE_TRIANGULAR_PANEL_WIDTH 38 | #define EIGEN_TUNE_TRIANGULAR_PANEL_WIDTH 8 39 | #endif 40 | 41 | 42 | /** Defines the default number of registers available for that architecture. 43 | * Currently it must be 8 or 16. Other values will fail. 44 | */ 45 | #ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 46 | #define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 8 47 | #endif 48 | 49 | #endif // EIGEN_DEFAULT_SETTINGS_H 50 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/arch/NEON/MathFunctions.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // This Source Code Form is subject to the terms of the Mozilla 5 | // Public License v. 2.0. If a copy of the MPL was not distributed 6 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 7 | 8 | /* The sin, cos, exp, and log functions of this file come from 9 | * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/ 10 | */ 11 | 12 | #ifndef EIGEN_MATH_FUNCTIONS_NEON_H 13 | #define EIGEN_MATH_FUNCTIONS_NEON_H 14 | 15 | namespace Eigen { 16 | 17 | namespace internal { 18 | 19 | template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED 20 | Packet4f pexp(const Packet4f& _x) 21 | { 22 | Packet4f x = _x; 23 | Packet4f tmp, fx; 24 | 25 | _EIGEN_DECLARE_CONST_Packet4f(1 , 1.0f); 26 | _EIGEN_DECLARE_CONST_Packet4f(half, 0.5f); 27 | _EIGEN_DECLARE_CONST_Packet4i(0x7f, 0x7f); 28 | _EIGEN_DECLARE_CONST_Packet4f(exp_hi, 88.3762626647950f); 29 | _EIGEN_DECLARE_CONST_Packet4f(exp_lo, -88.3762626647949f); 30 | _EIGEN_DECLARE_CONST_Packet4f(cephes_LOG2EF, 1.44269504088896341f); 31 | _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C1, 0.693359375f); 32 | _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C2, -2.12194440e-4f); 33 | _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p0, 1.9875691500E-4f); 34 | _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p1, 1.3981999507E-3f); 35 | _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p2, 8.3334519073E-3f); 36 | _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p3, 4.1665795894E-2f); 37 | _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p4, 1.6666665459E-1f); 38 | _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p5, 5.0000001201E-1f); 39 | 40 | x = vminq_f32(x, p4f_exp_hi); 41 | x = vmaxq_f32(x, p4f_exp_lo); 42 | 43 | /* express exp(x) as exp(g + n*log(2)) */ 44 | fx = vmlaq_f32(p4f_half, x, p4f_cephes_LOG2EF); 45 | 46 | /* perform a floorf */ 47 | tmp = vcvtq_f32_s32(vcvtq_s32_f32(fx)); 48 | 49 | /* if greater, substract 1 */ 50 | Packet4ui mask = vcgtq_f32(tmp, fx); 51 | mask = vandq_u32(mask, vreinterpretq_u32_f32(p4f_1)); 52 | 53 | fx = vsubq_f32(tmp, vreinterpretq_f32_u32(mask)); 54 | 55 | tmp = vmulq_f32(fx, p4f_cephes_exp_C1); 56 | Packet4f z = vmulq_f32(fx, p4f_cephes_exp_C2); 57 | x = vsubq_f32(x, tmp); 58 | x = vsubq_f32(x, z); 59 | 60 | Packet4f y = vmulq_f32(p4f_cephes_exp_p0, x); 61 | z = vmulq_f32(x, x); 62 | y = vaddq_f32(y, p4f_cephes_exp_p1); 63 | y = vmulq_f32(y, x); 64 | y = vaddq_f32(y, p4f_cephes_exp_p2); 65 | y = vmulq_f32(y, x); 66 | y = vaddq_f32(y, p4f_cephes_exp_p3); 67 | y = vmulq_f32(y, x); 68 | y = vaddq_f32(y, p4f_cephes_exp_p4); 69 | y = vmulq_f32(y, x); 70 | y = vaddq_f32(y, p4f_cephes_exp_p5); 71 | 72 | y = vmulq_f32(y, z); 73 | y = vaddq_f32(y, x); 74 | y = vaddq_f32(y, p4f_1); 75 | 76 | /* build 2^n */ 77 | int32x4_t mm; 78 | mm = vcvtq_s32_f32(fx); 79 | mm = vaddq_s32(mm, p4i_0x7f); 80 | mm = vshlq_n_s32(mm, 23); 81 | Packet4f pow2n = vreinterpretq_f32_s32(mm); 82 | 83 | y = vmulq_f32(y, pow2n); 84 | return y; 85 | } 86 | 87 | } // end namespace internal 88 | 89 | } // end namespace Eigen 90 | 91 | #endif // EIGEN_MATH_FUNCTIONS_NEON_H 92 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/arch/SSE/TypeCasting.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2015 Benoit Steiner 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_TYPE_CASTING_SSE_H 11 | #define EIGEN_TYPE_CASTING_SSE_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | 17 | template <> 18 | struct type_casting_traits { 19 | enum { 20 | VectorizedCast = 1, 21 | SrcCoeffRatio = 1, 22 | TgtCoeffRatio = 1 23 | }; 24 | }; 25 | 26 | template<> EIGEN_STRONG_INLINE Packet4i pcast(const Packet4f& a) { 27 | return _mm_cvttps_epi32(a); 28 | } 29 | 30 | 31 | template <> 32 | struct type_casting_traits { 33 | enum { 34 | VectorizedCast = 1, 35 | SrcCoeffRatio = 1, 36 | TgtCoeffRatio = 1 37 | }; 38 | }; 39 | 40 | template<> EIGEN_STRONG_INLINE Packet4f pcast(const Packet4i& a) { 41 | return _mm_cvtepi32_ps(a); 42 | } 43 | 44 | 45 | template <> 46 | struct type_casting_traits { 47 | enum { 48 | VectorizedCast = 1, 49 | SrcCoeffRatio = 2, 50 | TgtCoeffRatio = 1 51 | }; 52 | }; 53 | 54 | template<> EIGEN_STRONG_INLINE Packet4f pcast(const Packet2d& a, const Packet2d& b) { 55 | return _mm_shuffle_ps(_mm_cvtpd_ps(a), _mm_cvtpd_ps(b), (1 << 2) | (1 << 6)); 56 | } 57 | 58 | template <> 59 | struct type_casting_traits { 60 | enum { 61 | VectorizedCast = 1, 62 | SrcCoeffRatio = 1, 63 | TgtCoeffRatio = 2 64 | }; 65 | }; 66 | 67 | template<> EIGEN_STRONG_INLINE Packet2d pcast(const Packet4f& a) { 68 | // Simply discard the second half of the input 69 | return _mm_cvtps_pd(a); 70 | } 71 | 72 | 73 | } // end namespace internal 74 | 75 | } // end namespace Eigen 76 | 77 | #endif // EIGEN_TYPE_CASTING_SSE_H 78 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/functors/TernaryFunctors.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2016 Eugene Brevdo 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_TERNARY_FUNCTORS_H 11 | #define EIGEN_TERNARY_FUNCTORS_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | 17 | //---------- associative ternary functors ---------- 18 | 19 | 20 | 21 | } // end namespace internal 22 | 23 | } // end namespace Eigen 24 | 25 | #endif // EIGEN_TERNARY_FUNCTORS_H 26 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/util/DisableStupidWarnings.h: -------------------------------------------------------------------------------- 1 | #ifndef EIGEN_WARNINGS_DISABLED 2 | #define EIGEN_WARNINGS_DISABLED 3 | 4 | #ifdef _MSC_VER 5 | // 4100 - unreferenced formal parameter (occurred e.g. in aligned_allocator::destroy(pointer p)) 6 | // 4101 - unreferenced local variable 7 | // 4127 - conditional expression is constant 8 | // 4181 - qualifier applied to reference type ignored 9 | // 4211 - nonstandard extension used : redefined extern to static 10 | // 4244 - 'argument' : conversion from 'type1' to 'type2', possible loss of data 11 | // 4273 - QtAlignedMalloc, inconsistent DLL linkage 12 | // 4324 - structure was padded due to declspec(align()) 13 | // 4503 - decorated name length exceeded, name was truncated 14 | // 4512 - assignment operator could not be generated 15 | // 4522 - 'class' : multiple assignment operators specified 16 | // 4700 - uninitialized local variable 'xyz' used 17 | // 4714 - function marked as __forceinline not inlined 18 | // 4717 - 'function' : recursive on all control paths, function will cause runtime stack overflow 19 | // 4800 - 'type' : forcing value to bool 'true' or 'false' (performance warning) 20 | #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS 21 | #pragma warning( push ) 22 | #endif 23 | #pragma warning( disable : 4100 4101 4127 4181 4211 4244 4273 4324 4503 4512 4522 4700 4714 4717 4800) 24 | 25 | #elif defined __INTEL_COMPILER 26 | // 2196 - routine is both "inline" and "noinline" ("noinline" assumed) 27 | // ICC 12 generates this warning even without any inline keyword, when defining class methods 'inline' i.e. inside of class body 28 | // typedef that may be a reference type. 29 | // 279 - controlling expression is constant 30 | // ICC 12 generates this warning on assert(constant_expression_depending_on_template_params) and frankly this is a legitimate use case. 31 | // 1684 - conversion from pointer to same-sized integral type (potential portability problem) 32 | // 2259 - non-pointer conversion from "Eigen::Index={ptrdiff_t={long}}" to "int" may lose significant bits 33 | #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS 34 | #pragma warning push 35 | #endif 36 | #pragma warning disable 2196 279 1684 2259 37 | 38 | #elif defined __clang__ 39 | // -Wconstant-logical-operand - warning: use of logical && with constant operand; switch to bitwise & or remove constant 40 | // this is really a stupid warning as it warns on compile-time expressions involving enums 41 | #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS 42 | #pragma clang diagnostic push 43 | #endif 44 | #pragma clang diagnostic ignored "-Wconstant-logical-operand" 45 | 46 | #elif defined __GNUC__ && __GNUC__>=6 47 | 48 | #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS 49 | #pragma GCC diagnostic push 50 | #endif 51 | #pragma GCC diagnostic ignored "-Wignored-attributes" 52 | 53 | #endif 54 | 55 | #if defined __NVCC__ 56 | // Disable the "statement is unreachable" message 57 | #pragma diag_suppress code_is_unreachable 58 | // Disable the "dynamic initialization in unreachable code" message 59 | #pragma diag_suppress initialization_not_reachable 60 | // Disable the "invalid error number" message that we get with older versions of nvcc 61 | #pragma diag_suppress 1222 62 | // Disable the "calling a __host__ function from a __host__ __device__ function is not allowed" messages (yes, there are many of them and they seem to change with every version of the compiler) 63 | #pragma diag_suppress 2527 64 | #pragma diag_suppress 2529 65 | #pragma diag_suppress 2651 66 | #pragma diag_suppress 2653 67 | #pragma diag_suppress 2668 68 | #pragma diag_suppress 2669 69 | #pragma diag_suppress 2670 70 | #pragma diag_suppress 2671 71 | #pragma diag_suppress 2735 72 | #pragma diag_suppress 2737 73 | #endif 74 | 75 | #endif // not EIGEN_WARNINGS_DISABLED 76 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/util/NonMPL2.h: -------------------------------------------------------------------------------- 1 | #ifdef EIGEN_MPL2_ONLY 2 | #error Including non-MPL2 code in EIGEN_MPL2_ONLY mode 3 | #endif 4 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Core/util/ReenableStupidWarnings.h: -------------------------------------------------------------------------------- 1 | #ifdef EIGEN_WARNINGS_DISABLED 2 | #undef EIGEN_WARNINGS_DISABLED 3 | 4 | #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS 5 | #ifdef _MSC_VER 6 | #pragma warning( pop ) 7 | #elif defined __INTEL_COMPILER 8 | #pragma warning pop 9 | #elif defined __clang__ 10 | #pragma clang diagnostic pop 11 | #elif defined __GNUC__ && __GNUC__>=6 12 | #pragma GCC diagnostic pop 13 | #endif 14 | 15 | #if defined __NVCC__ 16 | // Don't reenable the diagnostic messages, as it turns out these messages need 17 | // to be disabled at the point of the template instantiation (i.e the user code) 18 | // otherwise they'll be triggered by nvcc. 19 | // #pragma diag_default code_is_unreachable 20 | // #pragma diag_default initialization_not_reachable 21 | // #pragma diag_default 2651 22 | // #pragma diag_default 2653 23 | #endif 24 | 25 | #endif 26 | 27 | #endif // EIGEN_WARNINGS_DISABLED 28 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Eigenvalues/RealSchur_LAPACKE.h: -------------------------------------------------------------------------------- 1 | /* 2 | Copyright (c) 2011, Intel Corporation. All rights reserved. 3 | 4 | Redistribution and use in source and binary forms, with or without modification, 5 | are permitted provided that the following conditions are met: 6 | 7 | * Redistributions of source code must retain the above copyright notice, this 8 | list of conditions and the following disclaimer. 9 | * Redistributions in binary form must reproduce the above copyright notice, 10 | this list of conditions and the following disclaimer in the documentation 11 | and/or other materials provided with the distribution. 12 | * Neither the name of Intel Corporation nor the names of its contributors may 13 | be used to endorse or promote products derived from this software without 14 | specific prior written permission. 15 | 16 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND 17 | ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 18 | WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 19 | DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR 20 | ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 | (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 22 | LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON 23 | ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 | (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 25 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 | 27 | ******************************************************************************** 28 | * Content : Eigen bindings to LAPACKe 29 | * Real Schur needed to real unsymmetrical eigenvalues/eigenvectors. 30 | ******************************************************************************** 31 | */ 32 | 33 | #ifndef EIGEN_REAL_SCHUR_LAPACKE_H 34 | #define EIGEN_REAL_SCHUR_LAPACKE_H 35 | 36 | namespace Eigen { 37 | 38 | /** \internal Specialization for the data types supported by LAPACKe */ 39 | 40 | #define EIGEN_LAPACKE_SCHUR_REAL(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX, LAPACKE_PREFIX_U, EIGCOLROW, LAPACKE_COLROW) \ 41 | template<> template inline \ 42 | RealSchur >& \ 43 | RealSchur >::compute(const EigenBase& matrix, bool computeU) \ 44 | { \ 45 | eigen_assert(matrix.cols() == matrix.rows()); \ 46 | \ 47 | lapack_int n = internal::convert_index(matrix.cols()), sdim, info; \ 48 | lapack_int matrix_order = LAPACKE_COLROW; \ 49 | char jobvs, sort='N'; \ 50 | LAPACK_##LAPACKE_PREFIX_U##_SELECT2 select = 0; \ 51 | jobvs = (computeU) ? 'V' : 'N'; \ 52 | m_matU.resize(n, n); \ 53 | lapack_int ldvs = internal::convert_index(m_matU.outerStride()); \ 54 | m_matT = matrix; \ 55 | lapack_int lda = internal::convert_index(m_matT.outerStride()); \ 56 | Matrix wr, wi; \ 57 | wr.resize(n, 1); wi.resize(n, 1); \ 58 | info = LAPACKE_##LAPACKE_PREFIX##gees( matrix_order, jobvs, sort, select, n, (LAPACKE_TYPE*)m_matT.data(), lda, &sdim, (LAPACKE_TYPE*)wr.data(), (LAPACKE_TYPE*)wi.data(), (LAPACKE_TYPE*)m_matU.data(), ldvs ); \ 59 | if(info == 0) \ 60 | m_info = Success; \ 61 | else \ 62 | m_info = NoConvergence; \ 63 | \ 64 | m_isInitialized = true; \ 65 | m_matUisUptodate = computeU; \ 66 | return *this; \ 67 | \ 68 | } 69 | 70 | EIGEN_LAPACKE_SCHUR_REAL(double, double, d, D, ColMajor, LAPACK_COL_MAJOR) 71 | EIGEN_LAPACKE_SCHUR_REAL(float, float, s, S, ColMajor, LAPACK_COL_MAJOR) 72 | EIGEN_LAPACKE_SCHUR_REAL(double, double, d, D, RowMajor, LAPACK_ROW_MAJOR) 73 | EIGEN_LAPACKE_SCHUR_REAL(float, float, s, S, RowMajor, LAPACK_ROW_MAJOR) 74 | 75 | } // end namespace Eigen 76 | 77 | #endif // EIGEN_REAL_SCHUR_LAPACKE_H 78 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/Geometry/EulerAngles.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_EULERANGLES_H 11 | #define EIGEN_EULERANGLES_H 12 | 13 | namespace Eigen { 14 | 15 | /** \geometry_module \ingroup Geometry_Module 16 | * 17 | * 18 | * \returns the Euler-angles of the rotation matrix \c *this using the convention defined by the triplet (\a a0,\a a1,\a a2) 19 | * 20 | * Each of the three parameters \a a0,\a a1,\a a2 represents the respective rotation axis as an integer in {0,1,2}. 21 | * For instance, in: 22 | * \code Vector3f ea = mat.eulerAngles(2, 0, 2); \endcode 23 | * "2" represents the z axis and "0" the x axis, etc. The returned angles are such that 24 | * we have the following equality: 25 | * \code 26 | * mat == AngleAxisf(ea[0], Vector3f::UnitZ()) 27 | * * AngleAxisf(ea[1], Vector3f::UnitX()) 28 | * * AngleAxisf(ea[2], Vector3f::UnitZ()); \endcode 29 | * This corresponds to the right-multiply conventions (with right hand side frames). 30 | * 31 | * The returned angles are in the ranges [0:pi]x[-pi:pi]x[-pi:pi]. 32 | * 33 | * \sa class AngleAxis 34 | */ 35 | template 36 | EIGEN_DEVICE_FUNC inline Matrix::Scalar,3,1> 37 | MatrixBase::eulerAngles(Index a0, Index a1, Index a2) const 38 | { 39 | EIGEN_USING_STD_MATH(atan2) 40 | EIGEN_USING_STD_MATH(sin) 41 | EIGEN_USING_STD_MATH(cos) 42 | /* Implemented from Graphics Gems IV */ 43 | EIGEN_STATIC_ASSERT_MATRIX_SPECIFIC_SIZE(Derived,3,3) 44 | 45 | Matrix res; 46 | typedef Matrix Vector2; 47 | 48 | const Index odd = ((a0+1)%3 == a1) ? 0 : 1; 49 | const Index i = a0; 50 | const Index j = (a0 + 1 + odd)%3; 51 | const Index k = (a0 + 2 - odd)%3; 52 | 53 | if (a0==a2) 54 | { 55 | res[0] = atan2(coeff(j,i), coeff(k,i)); 56 | if((odd && res[0]Scalar(0))) 57 | { 58 | if(res[0] > Scalar(0)) { 59 | res[0] -= Scalar(EIGEN_PI); 60 | } 61 | else { 62 | res[0] += Scalar(EIGEN_PI); 63 | } 64 | Scalar s2 = Vector2(coeff(j,i), coeff(k,i)).norm(); 65 | res[1] = -atan2(s2, coeff(i,i)); 66 | } 67 | else 68 | { 69 | Scalar s2 = Vector2(coeff(j,i), coeff(k,i)).norm(); 70 | res[1] = atan2(s2, coeff(i,i)); 71 | } 72 | 73 | // With a=(0,1,0), we have i=0; j=1; k=2, and after computing the first two angles, 74 | // we can compute their respective rotation, and apply its inverse to M. Since the result must 75 | // be a rotation around x, we have: 76 | // 77 | // c2 s1.s2 c1.s2 1 0 0 78 | // 0 c1 -s1 * M = 0 c3 s3 79 | // -s2 s1.c2 c1.c2 0 -s3 c3 80 | // 81 | // Thus: m11.c1 - m21.s1 = c3 & m12.c1 - m22.s1 = s3 82 | 83 | Scalar s1 = sin(res[0]); 84 | Scalar c1 = cos(res[0]); 85 | res[2] = atan2(c1*coeff(j,k)-s1*coeff(k,k), c1*coeff(j,j) - s1 * coeff(k,j)); 86 | } 87 | else 88 | { 89 | res[0] = atan2(coeff(j,k), coeff(k,k)); 90 | Scalar c2 = Vector2(coeff(i,i), coeff(i,j)).norm(); 91 | if((odd && res[0]Scalar(0))) { 92 | if(res[0] > Scalar(0)) { 93 | res[0] -= Scalar(EIGEN_PI); 94 | } 95 | else { 96 | res[0] += Scalar(EIGEN_PI); 97 | } 98 | res[1] = atan2(-coeff(i,k), -c2); 99 | } 100 | else 101 | res[1] = atan2(-coeff(i,k), c2); 102 | Scalar s1 = sin(res[0]); 103 | Scalar c1 = cos(res[0]); 104 | res[2] = atan2(s1*coeff(k,i)-c1*coeff(j,i), c1*coeff(j,j) - s1 * coeff(k,j)); 105 | } 106 | if (!odd) 107 | res = -res; 108 | 109 | return res; 110 | } 111 | 112 | } // end namespace Eigen 113 | 114 | #endif // EIGEN_EULERANGLES_H 115 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/LU/Determinant.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008 Benoit Jacob 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_DETERMINANT_H 11 | #define EIGEN_DETERMINANT_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | 17 | template 18 | inline const typename Derived::Scalar bruteforce_det3_helper 19 | (const MatrixBase& matrix, int a, int b, int c) 20 | { 21 | return matrix.coeff(0,a) 22 | * (matrix.coeff(1,b) * matrix.coeff(2,c) - matrix.coeff(1,c) * matrix.coeff(2,b)); 23 | } 24 | 25 | template 26 | const typename Derived::Scalar bruteforce_det4_helper 27 | (const MatrixBase& matrix, int j, int k, int m, int n) 28 | { 29 | return (matrix.coeff(j,0) * matrix.coeff(k,1) - matrix.coeff(k,0) * matrix.coeff(j,1)) 30 | * (matrix.coeff(m,2) * matrix.coeff(n,3) - matrix.coeff(n,2) * matrix.coeff(m,3)); 31 | } 32 | 33 | template struct determinant_impl 36 | { 37 | static inline typename traits::Scalar run(const Derived& m) 38 | { 39 | if(Derived::ColsAtCompileTime==Dynamic && m.rows()==0) 40 | return typename traits::Scalar(1); 41 | return m.partialPivLu().determinant(); 42 | } 43 | }; 44 | 45 | template struct determinant_impl 46 | { 47 | static inline typename traits::Scalar run(const Derived& m) 48 | { 49 | return m.coeff(0,0); 50 | } 51 | }; 52 | 53 | template struct determinant_impl 54 | { 55 | static inline typename traits::Scalar run(const Derived& m) 56 | { 57 | return m.coeff(0,0) * m.coeff(1,1) - m.coeff(1,0) * m.coeff(0,1); 58 | } 59 | }; 60 | 61 | template struct determinant_impl 62 | { 63 | static inline typename traits::Scalar run(const Derived& m) 64 | { 65 | return bruteforce_det3_helper(m,0,1,2) 66 | - bruteforce_det3_helper(m,1,0,2) 67 | + bruteforce_det3_helper(m,2,0,1); 68 | } 69 | }; 70 | 71 | template struct determinant_impl 72 | { 73 | static typename traits::Scalar run(const Derived& m) 74 | { 75 | // trick by Martin Costabel to compute 4x4 det with only 30 muls 76 | return bruteforce_det4_helper(m,0,1,2,3) 77 | - bruteforce_det4_helper(m,0,2,1,3) 78 | + bruteforce_det4_helper(m,0,3,1,2) 79 | + bruteforce_det4_helper(m,1,2,0,3) 80 | - bruteforce_det4_helper(m,1,3,0,2) 81 | + bruteforce_det4_helper(m,2,3,0,1); 82 | } 83 | }; 84 | 85 | } // end namespace internal 86 | 87 | /** \lu_module 88 | * 89 | * \returns the determinant of this matrix 90 | */ 91 | template 92 | inline typename internal::traits::Scalar MatrixBase::determinant() const 93 | { 94 | eigen_assert(rows() == cols()); 95 | typedef typename internal::nested_eval::type Nested; 96 | return internal::determinant_impl::type>::run(derived()); 97 | } 98 | 99 | } // end namespace Eigen 100 | 101 | #endif // EIGEN_DETERMINANT_H 102 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/LU/PartialPivLU_LAPACKE.h: -------------------------------------------------------------------------------- 1 | /* 2 | Copyright (c) 2011, Intel Corporation. All rights reserved. 3 | 4 | Redistribution and use in source and binary forms, with or without modification, 5 | are permitted provided that the following conditions are met: 6 | 7 | * Redistributions of source code must retain the above copyright notice, this 8 | list of conditions and the following disclaimer. 9 | * Redistributions in binary form must reproduce the above copyright notice, 10 | this list of conditions and the following disclaimer in the documentation 11 | and/or other materials provided with the distribution. 12 | * Neither the name of Intel Corporation nor the names of its contributors may 13 | be used to endorse or promote products derived from this software without 14 | specific prior written permission. 15 | 16 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND 17 | ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 18 | WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 19 | DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR 20 | ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 | (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 22 | LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON 23 | ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 | (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 25 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 | 27 | ******************************************************************************** 28 | * Content : Eigen bindings to LAPACKe 29 | * LU decomposition with partial pivoting based on LAPACKE_?getrf function. 30 | ******************************************************************************** 31 | */ 32 | 33 | #ifndef EIGEN_PARTIALLU_LAPACK_H 34 | #define EIGEN_PARTIALLU_LAPACK_H 35 | 36 | namespace Eigen { 37 | 38 | namespace internal { 39 | 40 | /** \internal Specialization for the data types supported by LAPACKe */ 41 | 42 | #define EIGEN_LAPACKE_LU_PARTPIV(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX) \ 43 | template \ 44 | struct partial_lu_impl \ 45 | { \ 46 | /* \internal performs the LU decomposition in-place of the matrix represented */ \ 47 | static lapack_int blocked_lu(Index rows, Index cols, EIGTYPE* lu_data, Index luStride, lapack_int* row_transpositions, lapack_int& nb_transpositions, lapack_int maxBlockSize=256) \ 48 | { \ 49 | EIGEN_UNUSED_VARIABLE(maxBlockSize);\ 50 | lapack_int matrix_order, first_zero_pivot; \ 51 | lapack_int m, n, lda, *ipiv, info; \ 52 | EIGTYPE* a; \ 53 | /* Set up parameters for ?getrf */ \ 54 | matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \ 55 | lda = convert_index(luStride); \ 56 | a = lu_data; \ 57 | ipiv = row_transpositions; \ 58 | m = convert_index(rows); \ 59 | n = convert_index(cols); \ 60 | nb_transpositions = 0; \ 61 | \ 62 | info = LAPACKE_##LAPACKE_PREFIX##getrf( matrix_order, m, n, (LAPACKE_TYPE*)a, lda, ipiv ); \ 63 | \ 64 | for(int i=0;i= 0); \ 67 | /* something should be done with nb_transpositions */ \ 68 | \ 69 | first_zero_pivot = info; \ 70 | return first_zero_pivot; \ 71 | } \ 72 | }; 73 | 74 | EIGEN_LAPACKE_LU_PARTPIV(double, double, d) 75 | EIGEN_LAPACKE_LU_PARTPIV(float, float, s) 76 | EIGEN_LAPACKE_LU_PARTPIV(dcomplex, lapack_complex_double, z) 77 | EIGEN_LAPACKE_LU_PARTPIV(scomplex, lapack_complex_float, c) 78 | 79 | } // end namespace internal 80 | 81 | } // end namespace Eigen 82 | 83 | #endif // EIGEN_PARTIALLU_LAPACK_H 84 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/QR/HouseholderQR_LAPACKE.h: -------------------------------------------------------------------------------- 1 | /* 2 | Copyright (c) 2011, Intel Corporation. All rights reserved. 3 | 4 | Redistribution and use in source and binary forms, with or without modification, 5 | are permitted provided that the following conditions are met: 6 | 7 | * Redistributions of source code must retain the above copyright notice, this 8 | list of conditions and the following disclaimer. 9 | * Redistributions in binary form must reproduce the above copyright notice, 10 | this list of conditions and the following disclaimer in the documentation 11 | and/or other materials provided with the distribution. 12 | * Neither the name of Intel Corporation nor the names of its contributors may 13 | be used to endorse or promote products derived from this software without 14 | specific prior written permission. 15 | 16 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND 17 | ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 18 | WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 19 | DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR 20 | ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 | (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 22 | LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON 23 | ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 | (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 25 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 | 27 | ******************************************************************************** 28 | * Content : Eigen bindings to LAPACKe 29 | * Householder QR decomposition of a matrix w/o pivoting based on 30 | * LAPACKE_?geqrf function. 31 | ******************************************************************************** 32 | */ 33 | 34 | #ifndef EIGEN_QR_LAPACKE_H 35 | #define EIGEN_QR_LAPACKE_H 36 | 37 | namespace Eigen { 38 | 39 | namespace internal { 40 | 41 | /** \internal Specialization for the data types supported by LAPACKe */ 42 | 43 | #define EIGEN_LAPACKE_QR_NOPIV(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX) \ 44 | template \ 45 | struct householder_qr_inplace_blocked \ 46 | { \ 47 | static void run(MatrixQR& mat, HCoeffs& hCoeffs, Index = 32, \ 48 | typename MatrixQR::Scalar* = 0) \ 49 | { \ 50 | lapack_int m = (lapack_int) mat.rows(); \ 51 | lapack_int n = (lapack_int) mat.cols(); \ 52 | lapack_int lda = (lapack_int) mat.outerStride(); \ 53 | lapack_int matrix_order = (MatrixQR::IsRowMajor) ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \ 54 | LAPACKE_##LAPACKE_PREFIX##geqrf( matrix_order, m, n, (LAPACKE_TYPE*)mat.data(), lda, (LAPACKE_TYPE*)hCoeffs.data()); \ 55 | hCoeffs.adjointInPlace(); \ 56 | } \ 57 | }; 58 | 59 | EIGEN_LAPACKE_QR_NOPIV(double, double, d) 60 | EIGEN_LAPACKE_QR_NOPIV(float, float, s) 61 | EIGEN_LAPACKE_QR_NOPIV(dcomplex, lapack_complex_double, z) 62 | EIGEN_LAPACKE_QR_NOPIV(scomplex, lapack_complex_float, c) 63 | 64 | } // end namespace internal 65 | 66 | } // end namespace Eigen 67 | 68 | #endif // EIGEN_QR_LAPACKE_H 69 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/SparseCore/MappedSparseMatrix.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008-2014 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_MAPPED_SPARSEMATRIX_H 11 | #define EIGEN_MAPPED_SPARSEMATRIX_H 12 | 13 | namespace Eigen { 14 | 15 | /** \deprecated Use Map > 16 | * \class MappedSparseMatrix 17 | * 18 | * \brief Sparse matrix 19 | * 20 | * \param _Scalar the scalar type, i.e. the type of the coefficients 21 | * 22 | * See http://www.netlib.org/linalg/html_templates/node91.html for details on the storage scheme. 23 | * 24 | */ 25 | namespace internal { 26 | template 27 | struct traits > : traits > 28 | {}; 29 | } // end namespace internal 30 | 31 | template 32 | class MappedSparseMatrix 33 | : public Map > 34 | { 35 | typedef Map > Base; 36 | 37 | public: 38 | 39 | typedef typename Base::StorageIndex StorageIndex; 40 | typedef typename Base::Scalar Scalar; 41 | 42 | inline MappedSparseMatrix(Index rows, Index cols, Index nnz, StorageIndex* outerIndexPtr, StorageIndex* innerIndexPtr, Scalar* valuePtr, StorageIndex* innerNonZeroPtr = 0) 43 | : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZeroPtr) 44 | {} 45 | 46 | /** Empty destructor */ 47 | inline ~MappedSparseMatrix() {} 48 | }; 49 | 50 | namespace internal { 51 | 52 | template 53 | struct evaluator > 54 | : evaluator > > 55 | { 56 | typedef MappedSparseMatrix<_Scalar,_Options,_StorageIndex> XprType; 57 | typedef evaluator > Base; 58 | 59 | evaluator() : Base() {} 60 | explicit evaluator(const XprType &mat) : Base(mat) {} 61 | }; 62 | 63 | } 64 | 65 | } // end namespace Eigen 66 | 67 | #endif // EIGEN_MAPPED_SPARSEMATRIX_H 68 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/SparseCore/SparseDot.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_SPARSE_DOT_H 11 | #define EIGEN_SPARSE_DOT_H 12 | 13 | namespace Eigen { 14 | 15 | template 16 | template 17 | typename internal::traits::Scalar 18 | SparseMatrixBase::dot(const MatrixBase& other) const 19 | { 20 | EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) 21 | EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived) 22 | EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived) 23 | EIGEN_STATIC_ASSERT((internal::is_same::value), 24 | YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) 25 | 26 | eigen_assert(size() == other.size()); 27 | eigen_assert(other.size()>0 && "you are using a non initialized vector"); 28 | 29 | internal::evaluator thisEval(derived()); 30 | typename internal::evaluator::InnerIterator i(thisEval, 0); 31 | Scalar res(0); 32 | while (i) 33 | { 34 | res += numext::conj(i.value()) * other.coeff(i.index()); 35 | ++i; 36 | } 37 | return res; 38 | } 39 | 40 | template 41 | template 42 | typename internal::traits::Scalar 43 | SparseMatrixBase::dot(const SparseMatrixBase& other) const 44 | { 45 | EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) 46 | EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived) 47 | EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived) 48 | EIGEN_STATIC_ASSERT((internal::is_same::value), 49 | YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) 50 | 51 | eigen_assert(size() == other.size()); 52 | 53 | internal::evaluator thisEval(derived()); 54 | typename internal::evaluator::InnerIterator i(thisEval, 0); 55 | 56 | internal::evaluator otherEval(other.derived()); 57 | typename internal::evaluator::InnerIterator j(otherEval, 0); 58 | 59 | Scalar res(0); 60 | while (i && j) 61 | { 62 | if (i.index()==j.index()) 63 | { 64 | res += numext::conj(i.value()) * j.value(); 65 | ++i; ++j; 66 | } 67 | else if (i.index() 76 | inline typename NumTraits::Scalar>::Real 77 | SparseMatrixBase::squaredNorm() const 78 | { 79 | return numext::real((*this).cwiseAbs2().sum()); 80 | } 81 | 82 | template 83 | inline typename NumTraits::Scalar>::Real 84 | SparseMatrixBase::norm() const 85 | { 86 | using std::sqrt; 87 | return sqrt(squaredNorm()); 88 | } 89 | 90 | template 91 | inline typename NumTraits::Scalar>::Real 92 | SparseMatrixBase::blueNorm() const 93 | { 94 | return internal::blueNorm_impl(*this); 95 | } 96 | } // end namespace Eigen 97 | 98 | #endif // EIGEN_SPARSE_DOT_H 99 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/SparseCore/SparseFuzzy.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008-2014 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_SPARSE_FUZZY_H 11 | #define EIGEN_SPARSE_FUZZY_H 12 | 13 | namespace Eigen { 14 | 15 | template 16 | template 17 | bool SparseMatrixBase::isApprox(const SparseMatrixBase& other, const RealScalar &prec) const 18 | { 19 | const typename internal::nested_eval::type actualA(derived()); 20 | typename internal::conditional::type, 22 | const PlainObject>::type actualB(other.derived()); 23 | 24 | return (actualA - actualB).squaredNorm() <= prec * prec * numext::mini(actualA.squaredNorm(), actualB.squaredNorm()); 25 | } 26 | 27 | } // end namespace Eigen 28 | 29 | #endif // EIGEN_SPARSE_FUZZY_H 30 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/SparseCore/SparseRedux.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008-2014 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_SPARSEREDUX_H 11 | #define EIGEN_SPARSEREDUX_H 12 | 13 | namespace Eigen { 14 | 15 | template 16 | typename internal::traits::Scalar 17 | SparseMatrixBase::sum() const 18 | { 19 | eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix"); 20 | Scalar res(0); 21 | internal::evaluator thisEval(derived()); 22 | for (Index j=0; j::InnerIterator iter(thisEval,j); iter; ++iter) 24 | res += iter.value(); 25 | return res; 26 | } 27 | 28 | template 29 | typename internal::traits >::Scalar 30 | SparseMatrix<_Scalar,_Options,_Index>::sum() const 31 | { 32 | eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix"); 33 | if(this->isCompressed()) 34 | return Matrix::Map(m_data.valuePtr(), m_data.size()).sum(); 35 | else 36 | return Base::sum(); 37 | } 38 | 39 | template 40 | typename internal::traits >::Scalar 41 | SparseVector<_Scalar,_Options,_Index>::sum() const 42 | { 43 | eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix"); 44 | return Matrix::Map(m_data.valuePtr(), m_data.size()).sum(); 45 | } 46 | 47 | } // end namespace Eigen 48 | 49 | #endif // EIGEN_SPARSEREDUX_H 50 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/SparseCore/SparseTranspose.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008-2015 Gael Guennebaud 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_SPARSETRANSPOSE_H 11 | #define EIGEN_SPARSETRANSPOSE_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | template 17 | class SparseTransposeImpl 18 | : public SparseMatrixBase > 19 | {}; 20 | 21 | template 22 | class SparseTransposeImpl 23 | : public SparseCompressedBase > 24 | { 25 | typedef SparseCompressedBase > Base; 26 | public: 27 | using Base::derived; 28 | typedef typename Base::Scalar Scalar; 29 | typedef typename Base::StorageIndex StorageIndex; 30 | 31 | inline Index nonZeros() const { return derived().nestedExpression().nonZeros(); } 32 | 33 | inline const Scalar* valuePtr() const { return derived().nestedExpression().valuePtr(); } 34 | inline const StorageIndex* innerIndexPtr() const { return derived().nestedExpression().innerIndexPtr(); } 35 | inline const StorageIndex* outerIndexPtr() const { return derived().nestedExpression().outerIndexPtr(); } 36 | inline const StorageIndex* innerNonZeroPtr() const { return derived().nestedExpression().innerNonZeroPtr(); } 37 | 38 | inline Scalar* valuePtr() { return derived().nestedExpression().valuePtr(); } 39 | inline StorageIndex* innerIndexPtr() { return derived().nestedExpression().innerIndexPtr(); } 40 | inline StorageIndex* outerIndexPtr() { return derived().nestedExpression().outerIndexPtr(); } 41 | inline StorageIndex* innerNonZeroPtr() { return derived().nestedExpression().innerNonZeroPtr(); } 42 | }; 43 | } 44 | 45 | template class TransposeImpl 46 | : public internal::SparseTransposeImpl 47 | { 48 | protected: 49 | typedef internal::SparseTransposeImpl Base; 50 | }; 51 | 52 | namespace internal { 53 | 54 | template 55 | struct unary_evaluator, IteratorBased> 56 | : public evaluator_base > 57 | { 58 | typedef typename evaluator::InnerIterator EvalIterator; 59 | public: 60 | typedef Transpose XprType; 61 | 62 | inline Index nonZerosEstimate() const { 63 | return m_argImpl.nonZerosEstimate(); 64 | } 65 | 66 | class InnerIterator : public EvalIterator 67 | { 68 | public: 69 | EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& unaryOp, Index outer) 70 | : EvalIterator(unaryOp.m_argImpl,outer) 71 | {} 72 | 73 | Index row() const { return EvalIterator::col(); } 74 | Index col() const { return EvalIterator::row(); } 75 | }; 76 | 77 | enum { 78 | CoeffReadCost = evaluator::CoeffReadCost, 79 | Flags = XprType::Flags 80 | }; 81 | 82 | explicit unary_evaluator(const XprType& op) :m_argImpl(op.nestedExpression()) {} 83 | 84 | protected: 85 | evaluator m_argImpl; 86 | }; 87 | 88 | } // end namespace internal 89 | 90 | } // end namespace Eigen 91 | 92 | #endif // EIGEN_SPARSETRANSPOSE_H 93 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/SparseLU/SparseLU_Utils.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2012 Désiré Nuentsa-Wakam 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | 11 | #ifndef EIGEN_SPARSELU_UTILS_H 12 | #define EIGEN_SPARSELU_UTILS_H 13 | 14 | namespace Eigen { 15 | namespace internal { 16 | 17 | /** 18 | * \brief Count Nonzero elements in the factors 19 | */ 20 | template 21 | void SparseLUImpl::countnz(const Index n, Index& nnzL, Index& nnzU, GlobalLU_t& glu) 22 | { 23 | nnzL = 0; 24 | nnzU = (glu.xusub)(n); 25 | Index nsuper = (glu.supno)(n); 26 | Index jlen; 27 | Index i, j, fsupc; 28 | if (n <= 0 ) return; 29 | // For each supernode 30 | for (i = 0; i <= nsuper; i++) 31 | { 32 | fsupc = glu.xsup(i); 33 | jlen = glu.xlsub(fsupc+1) - glu.xlsub(fsupc); 34 | 35 | for (j = fsupc; j < glu.xsup(i+1); j++) 36 | { 37 | nnzL += jlen; 38 | nnzU += j - fsupc + 1; 39 | jlen--; 40 | } 41 | } 42 | } 43 | 44 | /** 45 | * \brief Fix up the data storage lsub for L-subscripts. 46 | * 47 | * It removes the subscripts sets for structural pruning, 48 | * and applies permutation to the remaining subscripts 49 | * 50 | */ 51 | template 52 | void SparseLUImpl::fixupL(const Index n, const IndexVector& perm_r, GlobalLU_t& glu) 53 | { 54 | Index fsupc, i, j, k, jstart; 55 | 56 | StorageIndex nextl = 0; 57 | Index nsuper = (glu.supno)(n); 58 | 59 | // For each supernode 60 | for (i = 0; i <= nsuper; i++) 61 | { 62 | fsupc = glu.xsup(i); 63 | jstart = glu.xlsub(fsupc); 64 | glu.xlsub(fsupc) = nextl; 65 | for (j = jstart; j < glu.xlsub(fsupc + 1); j++) 66 | { 67 | glu.lsub(nextl) = perm_r(glu.lsub(j)); // Now indexed into P*A 68 | nextl++; 69 | } 70 | for (k = fsupc+1; k < glu.xsup(i+1); k++) 71 | glu.xlsub(k) = nextl; // other columns in supernode i 72 | } 73 | 74 | glu.xlsub(n) = nextl; 75 | } 76 | 77 | } // end namespace internal 78 | 79 | } // end namespace Eigen 80 | #endif // EIGEN_SPARSELU_UTILS_H 81 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/SparseLU/SparseLU_relax_snode.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2012 Désiré Nuentsa-Wakam 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | /* This file is a modified version of heap_relax_snode.c file in SuperLU 11 | * -- SuperLU routine (version 3.0) -- 12 | * Univ. of California Berkeley, Xerox Palo Alto Research Center, 13 | * and Lawrence Berkeley National Lab. 14 | * October 15, 2003 15 | * 16 | * Copyright (c) 1994 by Xerox Corporation. All rights reserved. 17 | * 18 | * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY 19 | * EXPRESSED OR IMPLIED. ANY USE IS AT YOUR OWN RISK. 20 | * 21 | * Permission is hereby granted to use or copy this program for any 22 | * purpose, provided the above notices are retained on all copies. 23 | * Permission to modify the code and to distribute modified code is 24 | * granted, provided the above notices are retained, and a notice that 25 | * the code was modified is included with the above copyright notice. 26 | */ 27 | 28 | #ifndef SPARSELU_RELAX_SNODE_H 29 | #define SPARSELU_RELAX_SNODE_H 30 | 31 | namespace Eigen { 32 | 33 | namespace internal { 34 | 35 | /** 36 | * \brief Identify the initial relaxed supernodes 37 | * 38 | * This routine is applied to a column elimination tree. 39 | * It assumes that the matrix has been reordered according to the postorder of the etree 40 | * \param n the number of columns 41 | * \param et elimination tree 42 | * \param relax_columns Maximum number of columns allowed in a relaxed snode 43 | * \param descendants Number of descendants of each node in the etree 44 | * \param relax_end last column in a supernode 45 | */ 46 | template 47 | void SparseLUImpl::relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end) 48 | { 49 | 50 | // compute the number of descendants of each node in the etree 51 | Index parent; 52 | relax_end.setConstant(emptyIdxLU); 53 | descendants.setZero(); 54 | for (Index j = 0; j < n; j++) 55 | { 56 | parent = et(j); 57 | if (parent != n) // not the dummy root 58 | descendants(parent) += descendants(j) + 1; 59 | } 60 | // Identify the relaxed supernodes by postorder traversal of the etree 61 | Index snode_start; // beginning of a snode 62 | for (Index j = 0; j < n; ) 63 | { 64 | parent = et(j); 65 | snode_start = j; 66 | while ( parent != n && descendants(parent) < relax_columns ) 67 | { 68 | j = parent; 69 | parent = et(j); 70 | } 71 | // Found a supernode in postordered etree, j is the last column 72 | relax_end(snode_start) = StorageIndex(j); // Record last column 73 | j++; 74 | // Search for a new leaf 75 | while (descendants(j) != 0 && j < n) j++; 76 | } // End postorder traversal of the etree 77 | 78 | } 79 | 80 | } // end namespace internal 81 | 82 | } // end namespace Eigen 83 | #endif 84 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/StlSupport/details.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009 Gael Guennebaud 5 | // Copyright (C) 2009 Hauke Heibel 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_STL_DETAILS_H 12 | #define EIGEN_STL_DETAILS_H 13 | 14 | #ifndef EIGEN_ALIGNED_ALLOCATOR 15 | #define EIGEN_ALIGNED_ALLOCATOR Eigen::aligned_allocator 16 | #endif 17 | 18 | namespace Eigen { 19 | 20 | // This one is needed to prevent reimplementing the whole std::vector. 21 | template 22 | class aligned_allocator_indirection : public EIGEN_ALIGNED_ALLOCATOR 23 | { 24 | public: 25 | typedef std::size_t size_type; 26 | typedef std::ptrdiff_t difference_type; 27 | typedef T* pointer; 28 | typedef const T* const_pointer; 29 | typedef T& reference; 30 | typedef const T& const_reference; 31 | typedef T value_type; 32 | 33 | template 34 | struct rebind 35 | { 36 | typedef aligned_allocator_indirection other; 37 | }; 38 | 39 | aligned_allocator_indirection() {} 40 | aligned_allocator_indirection(const aligned_allocator_indirection& ) : EIGEN_ALIGNED_ALLOCATOR() {} 41 | aligned_allocator_indirection(const EIGEN_ALIGNED_ALLOCATOR& ) {} 42 | template 43 | aligned_allocator_indirection(const aligned_allocator_indirection& ) {} 44 | template 45 | aligned_allocator_indirection(const EIGEN_ALIGNED_ALLOCATOR& ) {} 46 | ~aligned_allocator_indirection() {} 47 | }; 48 | 49 | #if EIGEN_COMP_MSVC 50 | 51 | // sometimes, MSVC detects, at compile time, that the argument x 52 | // in std::vector::resize(size_t s,T x) won't be aligned and generate an error 53 | // even if this function is never called. Whence this little wrapper. 54 | #define EIGEN_WORKAROUND_MSVC_STL_SUPPORT(T) \ 55 | typename Eigen::internal::conditional< \ 56 | Eigen::internal::is_arithmetic::value, \ 57 | T, \ 58 | Eigen::internal::workaround_msvc_stl_support \ 59 | >::type 60 | 61 | namespace internal { 62 | template struct workaround_msvc_stl_support : public T 63 | { 64 | inline workaround_msvc_stl_support() : T() {} 65 | inline workaround_msvc_stl_support(const T& other) : T(other) {} 66 | inline operator T& () { return *static_cast(this); } 67 | inline operator const T& () const { return *static_cast(this); } 68 | template 69 | inline T& operator=(const OtherT& other) 70 | { T::operator=(other); return *this; } 71 | inline workaround_msvc_stl_support& operator=(const workaround_msvc_stl_support& other) 72 | { T::operator=(other); return *this; } 73 | }; 74 | } 75 | 76 | #else 77 | 78 | #define EIGEN_WORKAROUND_MSVC_STL_SUPPORT(T) T 79 | 80 | #endif 81 | 82 | } 83 | 84 | #endif // EIGEN_STL_DETAILS_H 85 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/misc/Image.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009 Benoit Jacob 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_MISC_IMAGE_H 11 | #define EIGEN_MISC_IMAGE_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | 17 | /** \class image_retval_base 18 | * 19 | */ 20 | template 21 | struct traits > 22 | { 23 | typedef typename DecompositionType::MatrixType MatrixType; 24 | typedef Matrix< 25 | typename MatrixType::Scalar, 26 | MatrixType::RowsAtCompileTime, // the image is a subspace of the destination space, whose 27 | // dimension is the number of rows of the original matrix 28 | Dynamic, // we don't know at compile time the dimension of the image (the rank) 29 | MatrixType::Options, 30 | MatrixType::MaxRowsAtCompileTime, // the image matrix will consist of columns from the original matrix, 31 | MatrixType::MaxColsAtCompileTime // so it has the same number of rows and at most as many columns. 32 | > ReturnType; 33 | }; 34 | 35 | template struct image_retval_base 36 | : public ReturnByValue > 37 | { 38 | typedef _DecompositionType DecompositionType; 39 | typedef typename DecompositionType::MatrixType MatrixType; 40 | typedef ReturnByValue Base; 41 | 42 | image_retval_base(const DecompositionType& dec, const MatrixType& originalMatrix) 43 | : m_dec(dec), m_rank(dec.rank()), 44 | m_cols(m_rank == 0 ? 1 : m_rank), 45 | m_originalMatrix(originalMatrix) 46 | {} 47 | 48 | inline Index rows() const { return m_dec.rows(); } 49 | inline Index cols() const { return m_cols; } 50 | inline Index rank() const { return m_rank; } 51 | inline const DecompositionType& dec() const { return m_dec; } 52 | inline const MatrixType& originalMatrix() const { return m_originalMatrix; } 53 | 54 | template inline void evalTo(Dest& dst) const 55 | { 56 | static_cast*>(this)->evalTo(dst); 57 | } 58 | 59 | protected: 60 | const DecompositionType& m_dec; 61 | Index m_rank, m_cols; 62 | const MatrixType& m_originalMatrix; 63 | }; 64 | 65 | } // end namespace internal 66 | 67 | #define EIGEN_MAKE_IMAGE_HELPERS(DecompositionType) \ 68 | typedef typename DecompositionType::MatrixType MatrixType; \ 69 | typedef typename MatrixType::Scalar Scalar; \ 70 | typedef typename MatrixType::RealScalar RealScalar; \ 71 | typedef Eigen::internal::image_retval_base Base; \ 72 | using Base::dec; \ 73 | using Base::originalMatrix; \ 74 | using Base::rank; \ 75 | using Base::rows; \ 76 | using Base::cols; \ 77 | image_retval(const DecompositionType& dec, const MatrixType& originalMatrix) \ 78 | : Base(dec, originalMatrix) {} 79 | 80 | } // end namespace Eigen 81 | 82 | #endif // EIGEN_MISC_IMAGE_H 83 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/misc/Kernel.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009 Benoit Jacob 5 | // 6 | // This Source Code Form is subject to the terms of the Mozilla 7 | // Public License v. 2.0. If a copy of the MPL was not distributed 8 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 | 10 | #ifndef EIGEN_MISC_KERNEL_H 11 | #define EIGEN_MISC_KERNEL_H 12 | 13 | namespace Eigen { 14 | 15 | namespace internal { 16 | 17 | /** \class kernel_retval_base 18 | * 19 | */ 20 | template 21 | struct traits > 22 | { 23 | typedef typename DecompositionType::MatrixType MatrixType; 24 | typedef Matrix< 25 | typename MatrixType::Scalar, 26 | MatrixType::ColsAtCompileTime, // the number of rows in the "kernel matrix" 27 | // is the number of cols of the original matrix 28 | // so that the product "matrix * kernel = zero" makes sense 29 | Dynamic, // we don't know at compile-time the dimension of the kernel 30 | MatrixType::Options, 31 | MatrixType::MaxColsAtCompileTime, // see explanation for 2nd template parameter 32 | MatrixType::MaxColsAtCompileTime // the kernel is a subspace of the domain space, 33 | // whose dimension is the number of columns of the original matrix 34 | > ReturnType; 35 | }; 36 | 37 | template struct kernel_retval_base 38 | : public ReturnByValue > 39 | { 40 | typedef _DecompositionType DecompositionType; 41 | typedef ReturnByValue Base; 42 | 43 | explicit kernel_retval_base(const DecompositionType& dec) 44 | : m_dec(dec), 45 | m_rank(dec.rank()), 46 | m_cols(m_rank==dec.cols() ? 1 : dec.cols() - m_rank) 47 | {} 48 | 49 | inline Index rows() const { return m_dec.cols(); } 50 | inline Index cols() const { return m_cols; } 51 | inline Index rank() const { return m_rank; } 52 | inline const DecompositionType& dec() const { return m_dec; } 53 | 54 | template inline void evalTo(Dest& dst) const 55 | { 56 | static_cast*>(this)->evalTo(dst); 57 | } 58 | 59 | protected: 60 | const DecompositionType& m_dec; 61 | Index m_rank, m_cols; 62 | }; 63 | 64 | } // end namespace internal 65 | 66 | #define EIGEN_MAKE_KERNEL_HELPERS(DecompositionType) \ 67 | typedef typename DecompositionType::MatrixType MatrixType; \ 68 | typedef typename MatrixType::Scalar Scalar; \ 69 | typedef typename MatrixType::RealScalar RealScalar; \ 70 | typedef Eigen::internal::kernel_retval_base Base; \ 71 | using Base::dec; \ 72 | using Base::rank; \ 73 | using Base::rows; \ 74 | using Base::cols; \ 75 | kernel_retval(const DecompositionType& dec) : Base(dec) {} 76 | 77 | } // end namespace Eigen 78 | 79 | #endif // EIGEN_MISC_KERNEL_H 80 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/misc/RealSvd2x2.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2009-2010 Benoit Jacob 5 | // Copyright (C) 2013-2016 Gael Guennebaud 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | #ifndef EIGEN_REALSVD2X2_H 12 | #define EIGEN_REALSVD2X2_H 13 | 14 | namespace Eigen { 15 | 16 | namespace internal { 17 | 18 | template 19 | void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q, 20 | JacobiRotation *j_left, 21 | JacobiRotation *j_right) 22 | { 23 | using std::sqrt; 24 | using std::abs; 25 | Matrix m; 26 | m << numext::real(matrix.coeff(p,p)), numext::real(matrix.coeff(p,q)), 27 | numext::real(matrix.coeff(q,p)), numext::real(matrix.coeff(q,q)); 28 | JacobiRotation rot1; 29 | RealScalar t = m.coeff(0,0) + m.coeff(1,1); 30 | RealScalar d = m.coeff(1,0) - m.coeff(0,1); 31 | 32 | if(abs(d) < (std::numeric_limits::min)()) 33 | { 34 | rot1.s() = RealScalar(0); 35 | rot1.c() = RealScalar(1); 36 | } 37 | else 38 | { 39 | // If d!=0, then t/d cannot overflow because the magnitude of the 40 | // entries forming d are not too small compared to the ones forming t. 41 | RealScalar u = t / d; 42 | RealScalar tmp = sqrt(RealScalar(1) + numext::abs2(u)); 43 | rot1.s() = RealScalar(1) / tmp; 44 | rot1.c() = u / tmp; 45 | } 46 | m.applyOnTheLeft(0,1,rot1); 47 | j_right->makeJacobi(m,0,1); 48 | *j_left = rot1 * j_right->transpose(); 49 | } 50 | 51 | } // end namespace internal 52 | 53 | } // end namespace Eigen 54 | 55 | #endif // EIGEN_REALSVD2X2_H 56 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/misc/lapacke_mangling.h: -------------------------------------------------------------------------------- 1 | #ifndef LAPACK_HEADER_INCLUDED 2 | #define LAPACK_HEADER_INCLUDED 3 | 4 | #ifndef LAPACK_GLOBAL 5 | #if defined(LAPACK_GLOBAL_PATTERN_LC) || defined(ADD_) 6 | #define LAPACK_GLOBAL(lcname,UCNAME) lcname##_ 7 | #elif defined(LAPACK_GLOBAL_PATTERN_UC) || defined(UPPER) 8 | #define LAPACK_GLOBAL(lcname,UCNAME) UCNAME 9 | #elif defined(LAPACK_GLOBAL_PATTERN_MC) || defined(NOCHANGE) 10 | #define LAPACK_GLOBAL(lcname,UCNAME) lcname 11 | #else 12 | #define LAPACK_GLOBAL(lcname,UCNAME) lcname##_ 13 | #endif 14 | #endif 15 | 16 | #endif 17 | 18 | -------------------------------------------------------------------------------- /src/application/tools/Eigen/src/plugins/MatrixCwiseUnaryOps.h: -------------------------------------------------------------------------------- 1 | // This file is part of Eigen, a lightweight C++ template library 2 | // for linear algebra. 3 | // 4 | // Copyright (C) 2008-2009 Gael Guennebaud 5 | // Copyright (C) 2006-2008 Benoit Jacob 6 | // 7 | // This Source Code Form is subject to the terms of the Mozilla 8 | // Public License v. 2.0. If a copy of the MPL was not distributed 9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 | 11 | // This file is included into the body of the base classes supporting matrix specific coefficient-wise functions. 12 | // This include MatrixBase and SparseMatrixBase. 13 | 14 | 15 | typedef CwiseUnaryOp, const Derived> CwiseAbsReturnType; 16 | typedef CwiseUnaryOp, const Derived> CwiseAbs2ReturnType; 17 | typedef CwiseUnaryOp, const Derived> CwiseSqrtReturnType; 18 | typedef CwiseUnaryOp, const Derived> CwiseSignReturnType; 19 | typedef CwiseUnaryOp, const Derived> CwiseInverseReturnType; 20 | 21 | /// \returns an expression of the coefficient-wise absolute value of \c *this 22 | /// 23 | /// Example: \include MatrixBase_cwiseAbs.cpp 24 | /// Output: \verbinclude MatrixBase_cwiseAbs.out 25 | /// 26 | EIGEN_DOC_UNARY_ADDONS(cwiseAbs,absolute value) 27 | /// 28 | /// \sa cwiseAbs2() 29 | /// 30 | EIGEN_DEVICE_FUNC 31 | EIGEN_STRONG_INLINE const CwiseAbsReturnType 32 | cwiseAbs() const { return CwiseAbsReturnType(derived()); } 33 | 34 | /// \returns an expression of the coefficient-wise squared absolute value of \c *this 35 | /// 36 | /// Example: \include MatrixBase_cwiseAbs2.cpp 37 | /// Output: \verbinclude MatrixBase_cwiseAbs2.out 38 | /// 39 | EIGEN_DOC_UNARY_ADDONS(cwiseAbs2,squared absolute value) 40 | /// 41 | /// \sa cwiseAbs() 42 | /// 43 | EIGEN_DEVICE_FUNC 44 | EIGEN_STRONG_INLINE const CwiseAbs2ReturnType 45 | cwiseAbs2() const { return CwiseAbs2ReturnType(derived()); } 46 | 47 | /// \returns an expression of the coefficient-wise square root of *this. 48 | /// 49 | /// Example: \include MatrixBase_cwiseSqrt.cpp 50 | /// Output: \verbinclude MatrixBase_cwiseSqrt.out 51 | /// 52 | EIGEN_DOC_UNARY_ADDONS(cwiseSqrt,square-root) 53 | /// 54 | /// \sa cwisePow(), cwiseSquare() 55 | /// 56 | EIGEN_DEVICE_FUNC 57 | inline const CwiseSqrtReturnType 58 | cwiseSqrt() const { return CwiseSqrtReturnType(derived()); } 59 | 60 | /// \returns an expression of the coefficient-wise signum of *this. 61 | /// 62 | /// Example: \include MatrixBase_cwiseSign.cpp 63 | /// Output: \verbinclude MatrixBase_cwiseSign.out 64 | /// 65 | EIGEN_DOC_UNARY_ADDONS(cwiseSign,sign function) 66 | /// 67 | EIGEN_DEVICE_FUNC 68 | inline const CwiseSignReturnType 69 | cwiseSign() const { return CwiseSignReturnType(derived()); } 70 | 71 | 72 | /// \returns an expression of the coefficient-wise inverse of *this. 73 | /// 74 | /// Example: \include MatrixBase_cwiseInverse.cpp 75 | /// Output: \verbinclude MatrixBase_cwiseInverse.out 76 | /// 77 | EIGEN_DOC_UNARY_ADDONS(cwiseInverse,inverse) 78 | /// 79 | /// \sa cwiseProduct() 80 | /// 81 | EIGEN_DEVICE_FUNC 82 | inline const CwiseInverseReturnType 83 | cwiseInverse() const { return CwiseInverseReturnType(derived()); } 84 | 85 | 86 | -------------------------------------------------------------------------------- /src/application/tools/auto_download.cpp: -------------------------------------------------------------------------------- 1 | 2 | #include 3 | #include 4 | 5 | using namespace std; 6 | 7 | bool requires(const char* name) { 8 | 9 | auto onnx_file = iLogger::format("%s.onnx", name); 10 | if (not iLogger::exists(onnx_file)) { 11 | INFO("Auto download %s", onnx_file.c_str()); 12 | system(iLogger::format("wget http://zifuture.com:1556/fs/25.shared/%s", onnx_file.c_str()).c_str()); 13 | } 14 | 15 | bool exists = iLogger::exists(onnx_file); 16 | if (not exists) { 17 | INFOE("Download %s failed", onnx_file.c_str()); 18 | } 19 | return exists; 20 | } -------------------------------------------------------------------------------- /src/application/tools/deepsort.hpp: -------------------------------------------------------------------------------- 1 | 2 | #ifndef DEEPSORT_HPP 3 | #define DEEPSORT_HPP 4 | 5 | #include 6 | #include 7 | #include 8 | 9 | namespace DeepSORT { 10 | 11 | struct Box{ 12 | float left, top, right, bottom; 13 | cv::Mat feature; 14 | 15 | Box() = default; 16 | Box(float left, float top, float right, float bottom):left(left), top(top), right(right), bottom(bottom){} 17 | const float width() const{return right - left;} 18 | const float height() const{return bottom - top;} 19 | const cv::Point2f center() const{return cv::Point2f((left+right)/2, (top+bottom)/2);} 20 | }; 21 | 22 | template 23 | inline Box convert_to_box(const _T& b){ 24 | return Box(b.left, b.top, b.right, b.bottom); 25 | } 26 | 27 | template 28 | inline cv::Rect convert_box_to_rect(const _T& b){ 29 | return cv::Rect(b.left, b.top, b.right-b.left, b.bottom-b.top); 30 | } 31 | 32 | enum class State : int{ 33 | Tentative = 1, 34 | Confirmed = 2, 35 | Deleted = 3 36 | }; 37 | 38 | struct TrackerConfig{ 39 | 40 | int max_age = 150; 41 | int nhit = 3; 42 | float distance_threshold = 100; 43 | int nbuckets = 0; 44 | bool has_feature = false; 45 | 46 | // kalman 47 | // /** 初始状态 **/ 48 | float initiate_state[8]; 49 | 50 | // /** 每一侦的运动量协方差，下一侦 = 当前帧 + 运动量 **/ 51 | float per_frame_motion[8]; 52 | 53 | // /** 测量噪声，把输入映射到测量空间中后的噪声 **/ 54 | float noise[4]; 55 | 56 | void set_initiate_state(const std::vector& values); 57 | void set_per_frame_motion(const std::vector& values); 58 | void set_noise(const std::vector& values); 59 | 60 | TrackerConfig(); 61 | }; 62 | 63 | typedef std::vector BBoxes; 64 | 65 | class TrackObject{ 66 | public: 67 | virtual int id() const = 0; 68 | virtual State state() const = 0; 69 | virtual Box predict_box() const = 0; 70 | virtual Box last_position() const = 0; 71 | virtual bool is_confirmed() const = 0; 72 | virtual int time_since_update() const = 0; 73 | virtual std::vector trace_line() const = 0; 74 | virtual int trace_size() const = 0; 75 | virtual Box& location(int time_since_update=0) = 0; 76 | virtual const cv::Mat& feature_bucket() const = 0; 77 | }; 78 | 79 | class Tracker{ 80 | public: 81 | virtual std::vector get_objects() = 0; 82 | virtual void update(const BBoxes& boxes) = 0; 83 | }; 84 | 85 | std::shared_ptr create_tracker( 86 | const TrackerConfig& config = TrackerConfig() 87 | ); 88 | 89 | } 90 | 91 | #endif // DEEPSORT_HPP -------------------------------------------------------------------------------- /src/application/tools/zmq_remote_show.cpp: -------------------------------------------------------------------------------- 1 | 2 | #include "zmq_remote_show.hpp" 3 | #include "zmq_u.hpp" 4 | #include 5 | 6 | using namespace std; 7 | 8 | class ZMQRemoteShowImpl : public ZMQRemoteShow{ 9 | public: 10 | bool listen(const char* url){ 11 | try{ 12 | context_.reset(new zmq::context_t()); 13 | socket_.reset(new zmq::socket_t(*context_.get(), zmq::socket_type::rep)); 14 | socket_->bind(url); 15 | return true; 16 | }catch(zmq::error_t err){ 17 | INFOE("ZMQ exception: %s", err.what()); 18 | socket_.reset(); 19 | context_.reset(); 20 | } 21 | return false; 22 | } 23 | 24 | virtual void post(const void* data, int size) override{ 25 | 26 | if(size < 1 || data == nullptr){ 27 | INFOE("Null data to post"); 28 | return; 29 | } 30 | 31 | zmq::message_t msg; 32 | socket_->recv(msg); 33 | socket_->send(zmq::message_t(data, size)); 34 | } 35 | 36 | virtual void post(const cv::Mat& image) override{ 37 | 38 | vector data; 39 | cv::imencode(".jpg", image, data); 40 | post(data.data(), data.size()); 41 | } 42 | 43 | private: 44 | shared_ptr context_; 45 | shared_ptr socket_; 46 | }; 47 | 48 | std::shared_ptr create_zmq_remote_show(const char* listen){ 49 | 50 | shared_ptr instance(new ZMQRemoteShowImpl()); 51 | if(!instance->listen(listen)){ 52 | instance.reset(); 53 | } 54 | return instance; 55 | } 56 | -------------------------------------------------------------------------------- /src/application/tools/zmq_remote_show.hpp: -------------------------------------------------------------------------------- 1 | 2 | 3 | #ifndef ZMQ_REMOTE_SHOW_HPP 4 | #define ZMQ_REMOTE_SHOW_HPP 5 | 6 | #include 7 | #include 8 | 9 | class ZMQRemoteShow{ 10 | public: 11 | virtual void post(const void* data, int size) = 0; 12 | virtual void post(const cv::Mat& image) = 0; 13 | }; 14 | 15 | std::shared_ptr create_zmq_remote_show(const char* listen="tcp://0.0.0.0:15556"); 16 | 17 | #endif // ZMQ_REMOTE_SHOW_HPP -------------------------------------------------------------------------------- /src/main.cpp: -------------------------------------------------------------------------------- 1 | 2 | #include 3 | #include 4 | #include 5 | #include 6 | 7 | int yolo_obb_convert(); 8 | int yolo_obb_infer(); 9 | 10 | int main(int argc, char **argv) 11 | { 12 | if (argc > 1) 13 | { 14 | if (strcmp(argv[1], "infer") == 0) 15 | { 16 | yolo_obb_infer(); 17 | } 18 | else if (strcmp(argv[1], "convert") == 0) 19 | { 20 | yolo_obb_convert(); 21 | } 22 | else 23 | { 24 | INFOE("Invalid argument. Use 'infer' or 'convert'.\n"); 25 | } 26 | } 27 | else 28 | { 29 | INFO("No arguments provided. Defult to use infer.\n"); 30 | yolo_obb_infer(); 31 | } 32 | return 0; 33 | } 34 | -------------------------------------------------------------------------------- /src/tensorRT/builder/trt_builder.hpp: -------------------------------------------------------------------------------- 1 | 2 | 3 | #ifndef TRT_BUILDER_HPP 4 | #define TRT_BUILDER_HPP 5 | 6 | #include 7 | #include 8 | #include 9 | #include 10 | 11 | namespace TRT { 12 | 13 | typedef std::function& files, std::shared_ptr& tensor)> Int8Process; 14 | typedef std::function(const std::string& name, const std::vector& shape)> LayerHookFuncReshape; 15 | 16 | enum class ModelSourceType : int{ 17 | OnnX, 18 | OnnXData 19 | }; 20 | 21 | class ModelSource { 22 | public: 23 | ModelSource() = default; 24 | ModelSource(const std::string& onnxmodel); 25 | ModelSource(const char* onnxmodel); 26 | ModelSourceType type() const; 27 | std::string onnxmodel() const; 28 | std::string descript() const; 29 | const void* onnx_data() const; 30 | size_t onnx_data_size() const; 31 | 32 | static ModelSource onnx(const std::string& file){ 33 | ModelSource output; 34 | output.onnxmodel_ = file; 35 | output.type_ = ModelSourceType::OnnX; 36 | return output; 37 | } 38 | 39 | static ModelSource onnx_data(const void* ptr, size_t size){ 40 | ModelSource output; 41 | output.onnx_data_ = ptr; 42 | output.onnx_data_size_ = size; 43 | output.type_ = ModelSourceType::OnnXData; 44 | return output; 45 | } 46 | 47 | private: 48 | std::string onnxmodel_; 49 | const void* onnx_data_ = nullptr; 50 | size_t onnx_data_size_ = 0; 51 | ModelSourceType type_; 52 | }; 53 | 54 | enum class CompileOutputType : int{ 55 | File, 56 | Memory 57 | }; 58 | 59 | class CompileOutput{ 60 | public: 61 | CompileOutput(CompileOutputType type = CompileOutputType::Memory); 62 | CompileOutput(const std::string& file); 63 | CompileOutput(const char* file); 64 | void set_data(const std::vector& data); 65 | void set_data(std::vector&& data); 66 | 67 | const std::vector& data() const{return data_;}; 68 | CompileOutputType type() const{return type_;} 69 | std::string file() const{return file_;} 70 | 71 | private: 72 | CompileOutputType type_ = CompileOutputType::Memory; 73 | std::vector data_; 74 | std::string file_; 75 | }; 76 | 77 | class InputDims { 78 | public: 79 | InputDims() = default; 80 | 81 | // 当为-1时，保留导入时的网络结构尺寸 82 | InputDims(const std::initializer_list& dims); 83 | InputDims(const std::vector& dims); 84 | 85 | const std::vector& dims() const; 86 | 87 | private: 88 | std::vector dims_; 89 | }; 90 | 91 | enum class Mode : int { 92 | FP32, 93 | FP16, 94 | INT8 95 | }; 96 | 97 | enum class Calibrator : int { 98 | Entropy, 99 | MinMax 100 | }; 101 | 102 | const char* mode_string(Mode type); 103 | 104 | void set_layer_hook_reshape(const LayerHookFuncReshape& func); 105 | 106 | /** 当处于INT8模式时，int8process必须制定 107 | int8ImageDirectory和int8EntropyCalibratorFile指定一个即可 108 | 如果初次生成，指定了int8EntropyCalibratorFile，calibrator会保存到int8EntropyCalibratorFile指定的文件 109 | 如果已经生成过，指定了int8EntropyCalibratorFile，calibrator会从int8EntropyCalibratorFile指定的文件加载，而不是 110 | 从int8ImageDirectory读取图片再重新生成 111 | 当处于FP32或者FP16时，int8process、int8ImageDirectory、int8EntropyCalibratorFile都不需要指定 112 | 对于嵌入式设备，请把maxWorkspaceSize设置小一点，比如128MB = 1ul << 27 113 | **/ 114 | bool compile( 115 | Mode mode, 116 | unsigned int maxBatchSize, 117 | const ModelSource& source, 118 | const CompileOutput& saveto, 119 | const std::vector inputsDimsSetup = {}, 120 | Int8Process int8process = nullptr, 121 | const std::string& int8ImageDirectory = "", 122 | const std::string& int8EntropyCalibratorFile = "", 123 | Calibrator calibrator = Calibrator::Entropy, 124 | const size_t maxWorkspaceSize = 1ul << 30 // 1ul << 30 = 1GB 125 | ); 126 | }; 127 | 128 | #endif //TRT_BUILDER_HPP -------------------------------------------------------------------------------- /src/tensorRT/common/cuda_tools.cpp: -------------------------------------------------------------------------------- 1 | 2 | /* 3 | * 系统关于CUDA的功能函数 4 | */ 5 | 6 | 7 | #include "cuda_tools.hpp" 8 | 9 | namespace CUDATools{ 10 | bool check_driver(CUresult e, const char* call, int line, const char *file) { 11 | if (e != CUDA_SUCCESS) { 12 | 13 | const char* message = nullptr; 14 | const char* name = nullptr; 15 | cuGetErrorString(e, &message); 16 | cuGetErrorName(e, &name); 17 | INFOE("CUDA Driver error %s # %s, code = %s [ %d ] in file %s:%d", call, message, name, e, file, line); 18 | return false; 19 | } 20 | return true; 21 | } 22 | 23 | bool check_runtime(cudaError_t e, const char* call, int line, const char *file){ 24 | if (e != cudaSuccess) { 25 | INFOE("CUDA Runtime error %s # %s, code = %s [ %d ] in file %s:%d", call, cudaGetErrorString(e), cudaGetErrorName(e), e, file, line); 26 | return false; 27 | } 28 | return true; 29 | } 30 | 31 | bool check_device_id(int device_id){ 32 | int device_count = -1; 33 | checkCudaRuntime(cudaGetDeviceCount(&device_count)); 34 | if(device_id < 0 || device_id >= device_count){ 35 | INFOE("Invalid device id: %d, count = %d", device_id, device_count); 36 | return false; 37 | } 38 | return true; 39 | } 40 | 41 | int current_device_id(){ 42 | int device_id = 0; 43 | checkCudaRuntime(cudaGetDevice(&device_id)); 44 | return device_id; 45 | } 46 | 47 | dim3 grid_dims(int numJobs) { 48 | int numBlockThreads = numJobs < GPU_BLOCK_THREADS ? numJobs : GPU_BLOCK_THREADS; 49 | return dim3(((numJobs + numBlockThreads - 1) / (float)numBlockThreads)); 50 | } 51 | 52 | dim3 block_dims(int numJobs) { 53 | return numJobs < GPU_BLOCK_THREADS ? numJobs : GPU_BLOCK_THREADS; 54 | } 55 | 56 | std::string device_capability(int device_id){ 57 | cudaDeviceProp prop; 58 | checkCudaRuntime(cudaGetDeviceProperties(&prop, device_id)); 59 | return iLogger::format("%d.%d", prop.major, prop.minor); 60 | } 61 | 62 | std::string device_name(int device_id){ 63 | cudaDeviceProp prop; 64 | checkCudaRuntime(cudaGetDeviceProperties(&prop, device_id)); 65 | return prop.name; 66 | } 67 | 68 | std::string device_description(){ 69 | 70 | cudaDeviceProp prop; 71 | size_t free_mem, total_mem; 72 | int device_id = 0; 73 | 74 | checkCudaRuntime(cudaGetDevice(&device_id)); 75 | checkCudaRuntime(cudaGetDeviceProperties(&prop, device_id)); 76 | checkCudaRuntime(cudaMemGetInfo(&free_mem, &total_mem)); 77 | 78 | return iLogger::format( 79 | "[ID %d]<%s>[arch %d.%d][GMEM %.2f GB/%.2f GB]", 80 | device_id, prop.name, prop.major, prop.minor, 81 | free_mem / 1024.0f / 1024.0f / 1024.0f, 82 | total_mem / 1024.0f / 1024.0f / 1024.0f 83 | ); 84 | } 85 | 86 | AutoDevice::AutoDevice(int device_id){ 87 | 88 | cudaGetDevice(&old_); 89 | checkCudaRuntime(cudaSetDevice(device_id)); 90 | } 91 | 92 | AutoDevice::~AutoDevice(){ 93 | checkCudaRuntime(cudaSetDevice(old_)); 94 | } 95 | } -------------------------------------------------------------------------------- /src/tensorRT/common/cuda_tools.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CUDA_TOOLS_HPP 2 | #define CUDA_TOOLS_HPP 3 | 4 | /* 5 | * 系统关于CUDA的功能函数 6 | */ 7 | 8 | #include 9 | #include 10 | #include "ilogger.hpp" 11 | 12 | #define GPU_BLOCK_THREADS 512 13 | 14 | 15 | #define KernelPositionBlock \ 16 | int position = (blockDim.x * blockIdx.x + threadIdx.x); \ 17 | if (position >= (edge)) return; 18 | 19 | 20 | #define checkCudaDriver(call) CUDATools::check_driver(call, #call, __LINE__, __FILE__) 21 | #define checkCudaRuntime(call) CUDATools::check_runtime(call, #call, __LINE__, __FILE__) 22 | 23 | #define checkCudaKernel(...) \ 24 | __VA_ARGS__; \ 25 | do{cudaError_t cudaStatus = cudaPeekAtLastError(); \ 26 | if (cudaStatus != cudaSuccess){ \ 27 | INFOE("launch failed: %s", cudaGetErrorString(cudaStatus)); \ 28 | }} while(0); 29 | 30 | 31 | #define Assert(op) \ 32 | do{ \ 33 | bool cond = !(!(op)); \ 34 | if(!cond){ \ 35 | INFOF("Assert failed, " #op); \ 36 | } \ 37 | }while(false) 38 | 39 | 40 | struct CUctx_st; 41 | struct CUstream_st; 42 | 43 | typedef CUstream_st* ICUStream; 44 | typedef CUctx_st* ICUContext; 45 | typedef void* ICUDeviceptr; 46 | typedef int DeviceID; 47 | 48 | namespace CUDATools{ 49 | bool check_driver(CUresult e, const char* call, int iLine, const char *szFile); 50 | bool check_runtime(cudaError_t e, const char* call, int iLine, const char *szFile); 51 | bool check_device_id(int device_id); 52 | int current_device_id(); 53 | 54 | dim3 grid_dims(int numJobs); 55 | dim3 block_dims(int numJobs); 56 | 57 | // return 8.6 etc. 58 | std::string device_capability(int device_id); 59 | std::string device_name(int device_id); 60 | std::string device_description(); 61 | 62 | class AutoDevice{ 63 | public: 64 | AutoDevice(int device_id = 0); 65 | virtual ~AutoDevice(); 66 | 67 | private: 68 | int old_ = -1; 69 | }; 70 | } 71 | 72 | 73 | #endif // CUDA_TOOLS_HPP -------------------------------------------------------------------------------- /src/tensorRT/common/preprocess_kernel.cuh: -------------------------------------------------------------------------------- 1 | #ifndef PREPROCESS_KERNEL_CUH 2 | #define PREPROCESS_KERNEL_CUH 3 | 4 | #include "cuda_tools.hpp" 5 | 6 | namespace CUDAKernel{ 7 | 8 | enum class NormType : int{ 9 | None = 0, 10 | MeanStd = 1, 11 | AlphaBeta = 2 12 | }; 13 | 14 | enum class ChannelType : int{ 15 | None = 0, 16 | Invert = 1 17 | }; 18 | 19 | struct Norm{ 20 | float mean[3]; 21 | float std[3]; 22 | float alpha, beta; 23 | NormType type = NormType::None; 24 | ChannelType channel_type = ChannelType::None; 25 | 26 | // out = (x * alpha - mean) / std 27 | static Norm mean_std(const float mean[3], const float std[3], float alpha = 1/255.0f, ChannelType channel_type=ChannelType::None); 28 | 29 | // out = x * alpha + beta 30 | static Norm alpha_beta(float alpha, float beta = 0, ChannelType channel_type=ChannelType::None); 31 | 32 | // None 33 | static Norm None(); 34 | }; 35 | 36 | void resize_bilinear_and_normalize( 37 | uint8_t* src, int src_line_size, int src_width, int src_height, float* dst, int dst_width, int dst_height, 38 | const Norm& norm, 39 | cudaStream_t stream); 40 | 41 | void crop_resize_bilinear_and_normalize( 42 | uint8_t* src, int src_line_size, int src_width, int src_height, float* dst, int dst_width, int dst_height, 43 | const Norm& norm, 44 | cudaStream_t stream); 45 | 46 | void warp_affine_bilinear_and_normalize_plane( 47 | uint8_t* src, int src_line_size, int src_width, int src_height, 48 | float* dst , int dst_width, int dst_height, 49 | float* matrix_2_3, uint8_t const_value, const Norm& norm, 50 | cudaStream_t stream); 51 | 52 | void warp_affine_bilinear_and_normalize_focus( 53 | uint8_t* src, int src_line_size, int src_width, int src_height, 54 | float* dst , int dst_width, int dst_height, 55 | float* matrix_2_3, uint8_t const_value, const Norm& norm, 56 | cudaStream_t stream); 57 | 58 | // 可以用来图像校正、图像旋转等等 (测试比cpu快10倍以上) 59 | // 使用示范: 60 | // float* matrix_3_3 = nullptr; 61 | // size_t matrix_bytes = 3 * 3 * sizeof(f32); 62 | // checkCudaRuntime(cudaMalloc(&matrix_3_3, matrix_bytes)); 63 | // checkCudaRuntime(cudaMemset(matrix_3_3, 0, matrix_bytes)); 64 | // 65 | // #左上、右上、右下、左下 原图像四个点的坐标 66 | // cv::Point2f src_points[] = { 67 | // vctvctPoints[nImageIdx][0], 68 | // vctvctPoints[nImageIdx][1], 69 | // vctvctPoints[nImageIdx][2], 70 | // vctvctPoints[nImageIdx][3]}; 71 | // 72 | // #左上、右上、左下、右下（Z 字形排列） 目标图像四个点的坐标 73 | // cv::Point2f dst_points[] = { 74 | // cv::Point2f(0, 0), 75 | // cv::Point2f(nw-1, 0), 76 | // cv::Point2f(0, nh-1), 77 | // cv::Point2f(nw-1, nh-1) }; 78 | // 利用opencv 得到变换矩阵 dst -> src 的 矩阵 79 | // cv::Mat Perspect_Matrix = cv::getPerspectiveTransform(dst_points, src_points); 80 | // Perspect_Matrix.convertTo(Perspect_Matrix, CV_32FC1); 81 | // 拷贝到 gpu 82 | // checkCudaRuntime(cudaMemcpy(matrix_3_3, Perspect_Matrix.data, matrix_bytes, cudaMemcpyHostToDevice)); 83 | void warp_perspective( 84 | uint8_t* src, int src_line_size, int src_width, int src_height, float* dst, int dst_width, int dst_height, 85 | float* matrix_3_3, uint8_t const_value, const Norm& norm, cudaStream_t stream 86 | ); 87 | 88 | void norm_feature( 89 | float* feature_array, int num_feature, int feature_length, 90 | cudaStream_t stream 91 | ); 92 | 93 | void convert_nv12_to_bgr_invoke( 94 | const uint8_t* y, const uint8_t* uv, int width, int height, 95 | int linesize, uint8_t* dst, 96 | cudaStream_t stream); 97 | }; 98 | 99 | #endif // PREPROCESS_KERNEL_CUH -------------------------------------------------------------------------------- /src/tensorRT/import_lib.cpp: -------------------------------------------------------------------------------- 1 |  2 | #if defined(_WIN32) 3 | # define U_OS_WINDOWS 4 | #else 5 | # define U_OS_LINUX 6 | #endif 7 | 8 | #ifdef U_OS_WINDOWS 9 | #if defined(_DEBUG) 10 | # pragma comment(lib, "opencv_world346d.lib") 11 | #else 12 | # pragma comment(lib, "opencv_world346.lib") 13 | #endif 14 | 15 | //导入cuda 16 | #pragma comment(lib, "cuda.lib") 17 | #pragma comment(lib, "cudart.lib") 18 | #pragma comment(lib, "cublas.lib") 19 | #pragma comment(lib, "cudnn.lib") 20 | 21 | //导入tensorRT 22 | #pragma comment(lib, "nvinfer.lib") 23 | #pragma comment(lib, "nvinfer_plugin.lib") 24 | //#pragma comment(lib, "nvparsers.lib") 25 | 26 | #if defined(_DEBUG) 27 | #pragma comment(lib, "libprotobufd.lib") 28 | #else 29 | #pragma comment(lib, "libprotobuf.lib") 30 | #endif 31 | 32 | #ifdef HAS_PYTHON 33 | #pragma comment(lib, "python37.lib") 34 | #endif 35 | 36 | #endif // U_OS_WINDOWS -------------------------------------------------------------------------------- /src/tensorRT/infer/trt_infer.hpp: -------------------------------------------------------------------------------- 1 | 2 | 3 | #ifndef TRT_INFER_HPP 4 | #define TRT_INFER_HPP 5 | 6 | #include 7 | #include 8 | #include 9 | #include 10 | #include 11 | 12 | namespace TRT { 13 | 14 | class Infer { 15 | public: 16 | virtual void forward(bool sync = true) = 0; 17 | virtual int get_max_batch_size() = 0; 18 | virtual void set_stream(CUStream stream) = 0; 19 | virtual CUStream get_stream() = 0; 20 | virtual void synchronize() = 0; 21 | virtual size_t get_device_memory_size() = 0; 22 | virtual std::shared_ptr get_workspace() = 0; 23 | virtual std::shared_ptr input (int index = 0) = 0; 24 | virtual std::shared_ptr output(int index = 0) = 0; 25 | virtual std::shared_ptr tensor(const std::string& name) = 0; 26 | virtual std::string get_input_name (int index = 0) = 0; 27 | virtual std::string get_output_name(int index = 0) = 0; 28 | virtual bool is_output_name(const std::string& name) = 0; 29 | virtual bool is_input_name (const std::string& name) = 0; 30 | virtual int num_output() = 0; 31 | virtual int num_input() = 0; 32 | virtual void print() = 0; 33 | virtual int device() = 0; 34 | virtual void set_input (int index, std::shared_ptr tensor) = 0; 35 | virtual void set_output(int index, std::shared_ptr tensor) = 0; 36 | virtual std::shared_ptr> serial_engine() = 0; 37 | }; 38 | 39 | struct DeviceMemorySummary { 40 | size_t total; 41 | size_t available; 42 | }; 43 | 44 | DeviceMemorySummary get_current_device_summary(); 45 | int get_device_count(); 46 | int get_device(); 47 | 48 | void set_device(int device_id); 49 | std::shared_ptr load_infer_from_memory(const void* pdata, size_t size); 50 | std::shared_ptr load_infer(const std::string& file); 51 | bool init_nv_plugins(); 52 | 53 | }; //TRTInfer 54 | 55 | 56 | #endif //TRT_INFER_HPP -------------------------------------------------------------------------------- /src/tensorRT/onnx/onnx_pb.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) ONNX Project Contributors. 2 | // Licensed under the MIT license. 3 | 4 | #ifndef ONNX_ONNX_PB_H 5 | #define ONNX_ONNX_PB_H 6 | 7 | // Defines ONNX_EXPORT and ONNX_IMPORT. On Windows, this corresponds to 8 | // different declarations (dllexport and dllimport). On Linux/Mac, it just 9 | // resolves to the same "default visibility" setting. 10 | #if defined(_MSC_VER) 11 | #if defined(ONNX_BUILD_SHARED_LIBS) || defined(ONNX_BUILD_MAIN_LIB) 12 | #define ONNX_EXPORT __declspec(dllexport) 13 | #define ONNX_IMPORT __declspec(dllimport) 14 | #else 15 | #define ONNX_EXPORT 16 | #define ONNX_IMPORT 17 | #endif 18 | #else 19 | #if defined(__GNUC__) 20 | #define ONNX_EXPORT __attribute__((__visibility__("default"))) 21 | #else 22 | #define ONNX_EXPORT 23 | #endif 24 | #define ONNX_IMPORT ONNX_EXPORT 25 | #endif 26 | 27 | // ONNX_API is a macro that, depends on whether you are building the 28 | // main ONNX library or not, resolves to either ONNX_EXPORT or 29 | // ONNX_IMPORT. 30 | // 31 | // This is used in e.g. ONNX's protobuf files: when building the main library, 32 | // it is defined as ONNX_EXPORT to fix a Windows global-variable-in-dll 33 | // issue, and for anyone dependent on ONNX it will be defined as 34 | // ONNX_IMPORT. ONNX_BUILD_MAIN_LIB can also be set when being built 35 | // statically if ONNX is being linked into a shared library that wants 36 | // to export the ONNX APIs and classes. 37 | // 38 | // More details on Windows dllimport / dllexport can be found at 39 | // https://msdn.microsoft.com/en-us/library/3y1sfaz2.aspx 40 | // 41 | // This solution is similar to 42 | // https://github.com/pytorch/pytorch/blob/master/caffe2/core/common.h 43 | #define ONNX_API 44 | #include "onnx-ml.pb.h" 45 | 46 | #endif // ! ONNX_ONNX_PB_H 47 | -------------------------------------------------------------------------------- /src/tensorRT/onnx/readme.md: -------------------------------------------------------------------------------- 1 | # ONNX 2 | - 这几个文件来自于对ONNX的编译后提取的结果，由protoc生成的cpp 3 | - https://github.com/onnx/onnx -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/LoopHelpers.cpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #include "LoopHelpers.hpp" 6 | #include "onnx2trt_utils.hpp" 7 | 8 | namespace onnx2trt 9 | { 10 | 11 | nvinfer1::ITensor* addLoopCounter(IImporterContext* ctx, nvinfer1::ILoop* loop, int32_t initial) 12 | { 13 | nvinfer1::ITensor* initialTensor = addConstantScalar(ctx, initial, ::onnx::TensorProto::INT32, nvinfer1::Dims{1, 1})->getOutput(0); 14 | nvinfer1::ITensor* one = addConstantScalar(ctx, 1, ::onnx::TensorProto::INT32, nvinfer1::Dims{1, 1})->getOutput(0); 15 | 16 | auto counter = loop->addRecurrence(*initialTensor); 17 | nvinfer1::ITensor* addOne = ctx->network()->addElementWise(*counter->getOutput(0), *one, nvinfer1::ElementWiseOperation::kSUM)->getOutput(0); 18 | counter->setInput(1, *addOne); 19 | return counter->getOutput(0); 20 | } 21 | 22 | } // namespace onnx2trt 23 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/LoopHelpers.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include 8 | 9 | #include "ImporterContext.hpp" 10 | 11 | namespace onnx2trt 12 | { 13 | 14 | nvinfer1::ITensor* addLoopCounter(IImporterContext* ctx, nvinfer1::ILoop* loop, int32_t initial = 0); 15 | 16 | } // namespace onnx2trt 17 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/ModelImporter.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include "ImporterContext.hpp" 8 | #include "NvInferPlugin.h" 9 | #include "NvOnnxParser.h" 10 | #include "builtin_op_importers.hpp" 11 | #include "utils.hpp" 12 | 13 | namespace onnx2trt 14 | { 15 | 16 | Status parseGraph(IImporterContext* ctx, const ::onnx::GraphProto& graph, bool deserializingINetwork = false, int* currentNode = nullptr); 17 | 18 | class ModelImporter : public nvonnxparser::IParser 19 | { 20 | protected: 21 | string_map _op_importers; 22 | virtual Status importModel(::onnx::ModelProto const& model); 23 | 24 | private: 25 | ImporterContext _importer_ctx; 26 | std::list<::onnx::ModelProto> _onnx_models; // Needed for ownership of weights 27 | int _current_node; 28 | std::vector _errors; 29 | std::vector _input_dims; 30 | 31 | public: 32 | ModelImporter(nvinfer1::INetworkDefinition* network, nvinfer1::ILogger* logger, const std::vector& input_dims) 33 | : _op_importers(getBuiltinOpImporterMap()) 34 | , _importer_ctx(network, logger) 35 | , _input_dims(input_dims) 36 | { 37 | } 38 | bool parseWithWeightDescriptors(void const* serialized_onnx_model, size_t serialized_onnx_model_size) override; 39 | bool parse(void const* serialized_onnx_model, size_t serialized_onnx_model_size, const char* model_path = nullptr) override; 40 | bool supportsModel(void const* serialized_onnx_model, size_t serialized_onnx_model_size, 41 | SubGraphCollection_t& sub_graph_collection, const char* model_path = nullptr) override; 42 | 43 | bool supportsOperator(const char* op_name) const override; 44 | void destroy() override 45 | { 46 | delete this; 47 | } 48 | // virtual void registerOpImporter(std::string op, 49 | // NodeImporter const &node_importer) override { 50 | // // Note: This allows existing importers to be replaced 51 | // _op_importers[op] = node_importer; 52 | //} 53 | // virtual Status const &setInput(const char *name, 54 | // nvinfer1::ITensor *input) override; 55 | // virtual Status const& setOutput(const char* name, nvinfer1::ITensor** output) override; 56 | int getNbErrors() const override 57 | { 58 | return _errors.size(); 59 | } 60 | nvonnxparser::IParserError const* getError(int index) const override 61 | { 62 | assert(0 <= index && index < (int) _errors.size()); 63 | return &_errors[index]; 64 | } 65 | void clearErrors() override 66 | { 67 | _errors.clear(); 68 | } 69 | 70 | //...LG: Move the implementation to .cpp 71 | bool parseFromFile(const char* onnxModelFile, int verbosity) override; 72 | bool parseFromData(const void* onnx_data, size_t size, int verbosity) override; 73 | }; 74 | 75 | } // namespace onnx2trt 76 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/NvOnnxParser.cpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #include "NvOnnxParser.h" 6 | #include "ModelImporter.hpp" 7 | 8 | extern "C" void* createNvOnnxParser_INTERNAL(void* network_, void* logger_, int version, const std::vector& input_dims) 9 | { 10 | auto network = static_cast(network_); 11 | auto logger = static_cast(logger_); 12 | return new onnx2trt::ModelImporter(network, logger, input_dims); 13 | } 14 | 15 | extern "C" int getNvOnnxParserVersion() 16 | { 17 | return NV_ONNX_PARSER_VERSION; 18 | } -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/OnnxAttrs.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include 8 | #include 9 | #include 10 | #include 11 | 12 | #include "ImporterContext.hpp" 13 | 14 | class OnnxAttrs 15 | { 16 | template 17 | using string_map = std::unordered_map; 18 | typedef string_map<::onnx::AttributeProto const*> AttrMap; 19 | AttrMap _attrs; 20 | onnx2trt::IImporterContext* mCtx; 21 | 22 | public: 23 | explicit OnnxAttrs(::onnx::NodeProto const& onnx_node, onnx2trt::IImporterContext* ctx) 24 | : mCtx{ctx} 25 | { 26 | for (auto const& attr : onnx_node.attribute()) 27 | { 28 | _attrs.insert({attr.name(), &attr}); 29 | } 30 | } 31 | 32 | bool count(const std::string& key) const 33 | { 34 | return _attrs.count(key); 35 | } 36 | 37 | ::onnx::AttributeProto const* at(std::string key) const 38 | { 39 | if (!_attrs.count(key)) 40 | { 41 | throw std::out_of_range("Attribute not found: " + key); 42 | } 43 | return _attrs.at(key); 44 | } 45 | 46 | ::onnx::AttributeProto::AttributeType type(const std::string& key) const 47 | { 48 | return this->at(key)->type(); 49 | } 50 | 51 | 52 | template 53 | T get(const std::string& key) const; 54 | 55 | template 56 | T get(const std::string& key, T const& default_value) const 57 | { 58 | return _attrs.count(key) ? this->get(key) : default_value; 59 | } 60 | }; 61 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/RNNHelpers.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include 8 | #include 9 | #include 10 | 11 | #include "TensorOrWeights.hpp" 12 | #include "ImporterContext.hpp" 13 | 14 | namespace onnx2trt 15 | { 16 | 17 | nvinfer1::ITensor* addRNNInput(IImporterContext* ctx, const ::onnx::NodeProto& node, nvinfer1::ILoop* loop, std::vector& inputs, const std::string& direction); 18 | 19 | // Zeros out invalid timesteps in toMask. maxLen must be provided if reverse is true 20 | nvinfer1::ITensor* clearMissingSequenceElements(IImporterContext* ctx, const ::onnx::NodeProto& node, nvinfer1::ILoop* loop, nvinfer1::ITensor* seqLens, nvinfer1::ITensor* toMask, nvinfer1::ITensor* maxLen, bool reverse = false, nvinfer1::ITensor* counter = nullptr); 21 | 22 | // Returns a bool tensor which is true during valid timesteps 23 | nvinfer1::ITensor* getRaggedMask(IImporterContext* ctx, const ::onnx::NodeProto& node, nvinfer1::ILoop* loop, nvinfer1::ITensor* seqLens, nvinfer1::ITensor* maxLen = nullptr, bool reverse = false, nvinfer1::ITensor* counter = nullptr); 24 | 25 | // Selects between prevH and Ht to forward previous hidden state through invalid timesteps 26 | nvinfer1::ITensor* maskRNNHidden(IImporterContext* ctx, const ::onnx::NodeProto& node, nvinfer1::ILoop* loop, nvinfer1::ITensor* seqLens, nvinfer1::ITensor* prevH, nvinfer1::ITensor* Ht, nvinfer1::ITensor* maxLen = nullptr, bool reverse = false, nvinfer1::ITensor* counter = nullptr); 27 | 28 | // Splits a bidirectional hidden state into forward and reverse passes, masks each using maskRNNHidden, then concatenates 29 | nvinfer1::ITensor* maskBidirRNNHidden(IImporterContext* ctx, const ::onnx::NodeProto& node, nvinfer1::ILoop* loop, nvinfer1::ITensor* seqLens, nvinfer1::ITensor* maxLen, nvinfer1::ITensor* Ht1, nvinfer1::ITensor* Ht, nvinfer1::ITensor* singlePassShape); 30 | 31 | } // namespace onnx2trt 32 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/ShapedWeights.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include 8 | #include 9 | 10 | namespace onnx2trt 11 | { 12 | 13 | class ShapedWeights 14 | { 15 | public: 16 | using DataType = int32_t; 17 | 18 | static ShapedWeights empty(DataType type); 19 | 20 | ShapedWeights(); 21 | 22 | explicit ShapedWeights(DataType type, void* values, nvinfer1::Dims shape_); 23 | 24 | size_t count() const; 25 | 26 | size_t size_bytes() const; 27 | 28 | const char* getName() const; 29 | 30 | void setName(const char* name); 31 | 32 | explicit operator bool() const; 33 | 34 | operator nvinfer1::Weights() const; 35 | 36 | template 37 | T& at(size_t index) 38 | { 39 | assert(index >= 0 && (index * sizeof(T)) < size_bytes()); 40 | return static_cast(values)[index]; 41 | } 42 | 43 | template 44 | const T& at(size_t index) const 45 | { 46 | assert(index >= 0 && (index * sizeof(T)) < size_bytes()); 47 | return static_cast(values)[index]; 48 | } 49 | 50 | public: 51 | DataType type; 52 | void* values; 53 | nvinfer1::Dims shape; 54 | const char* name{}; 55 | }; 56 | 57 | class IImporterContext; 58 | bool transposeWeights(ShapedWeights const& weights, nvinfer1::Permutation const& perm, ShapedWeights* result, IImporterContext* ctx); 59 | 60 | } // namespace onnx2trt 61 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/TensorOrWeights.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include "ShapedWeights.hpp" 8 | 9 | #include 10 | #include 11 | 12 | namespace onnx2trt 13 | { 14 | 15 | class TensorOrWeights 16 | { 17 | union 18 | { 19 | nvinfer1::ITensor* _tensor; 20 | ShapedWeights _weights; 21 | }; 22 | enum 23 | { 24 | NODE_TENSOR, 25 | NODE_WEIGHTS 26 | } _variant; 27 | 28 | public: 29 | TensorOrWeights() 30 | : _tensor(nullptr) 31 | , _variant(NODE_TENSOR) 32 | { 33 | } 34 | TensorOrWeights(nvinfer1::ITensor* tensor) 35 | : _tensor(tensor) 36 | , _variant(NODE_TENSOR) 37 | { 38 | } 39 | TensorOrWeights(ShapedWeights const& weights) 40 | : _weights(weights) 41 | , _variant(NODE_WEIGHTS) 42 | { 43 | } 44 | bool is_tensor() const 45 | { 46 | return _variant == NODE_TENSOR; 47 | } 48 | bool is_weights() const 49 | { 50 | return _variant == NODE_WEIGHTS; 51 | } 52 | bool isNullTensor() const 53 | { 54 | return is_tensor() && _tensor == nullptr; 55 | } 56 | nvinfer1::ITensor& tensor() 57 | { 58 | assert(!isNullTensor()); 59 | return *_tensor; 60 | } 61 | nvinfer1::ITensor const& tensor() const 62 | { 63 | assert(!isNullTensor()); 64 | return *_tensor; 65 | } 66 | ShapedWeights& weights() 67 | { 68 | assert(is_weights()); 69 | return _weights; 70 | } 71 | ShapedWeights const& weights() const 72 | { 73 | assert(is_weights()); 74 | return _weights; 75 | } 76 | nvinfer1::Dims shape() const 77 | { 78 | return is_tensor() ? _tensor->getDimensions() : _weights.shape; 79 | } 80 | explicit operator bool() const 81 | { 82 | return is_tensor() ? _tensor != nullptr : static_cast(_weights); 83 | } 84 | bool isInt32() const 85 | { 86 | return is_tensor() ? _tensor->getType() == nvinfer1::DataType::kINT32 : _weights.type == ::onnx::TensorProto_DataType_INT32; 87 | } 88 | bool isBool() const 89 | { 90 | return is_tensor() ? _tensor->getType() == nvinfer1::DataType::kBOOL : _weights.type == ::onnx::TensorProto_DataType_BOOL; 91 | } 92 | std::string getName() const 93 | { 94 | return is_tensor() ? _tensor->getName() : _weights.getName(); 95 | } 96 | std::string getType() const 97 | { 98 | if (is_tensor()) 99 | { 100 | switch(_tensor->getType()) 101 | { 102 | case nvinfer1::DataType::kFLOAT:return "FLOAT"; 103 | case nvinfer1::DataType::kHALF: return "HALF"; 104 | case nvinfer1::DataType::kINT8: return "INT8"; 105 | case nvinfer1::DataType::kINT32: return "INT32"; 106 | case nvinfer1::DataType::kBOOL: return "BOOL"; 107 | default: return "UNKNOWN TYPE"; 108 | } 109 | } 110 | else 111 | { 112 | switch(_weights.type) 113 | { 114 | case ::onnx::TensorProto::DOUBLE: return "DOUBLE -> FLOAT"; 115 | case ::onnx::TensorProto::FLOAT: return "FLOAT"; 116 | case ::onnx::TensorProto::INT8: return "INT8"; 117 | case ::onnx::TensorProto::FLOAT16: return "HALF"; 118 | case ::onnx::TensorProto::BOOL: return "BOOL"; 119 | case ::onnx::TensorProto::INT32: return "INT32"; 120 | case ::onnx::TensorProto::INT64: return "INT64 -> INT32"; 121 | default: return "UNKNOWN TYPE"; 122 | } 123 | } 124 | } 125 | }; 126 | 127 | } // namespace onnx2trt 128 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/builtin_op_importers.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include "onnx2trt.hpp" 8 | #include "utils.hpp" 9 | 10 | namespace onnx2trt 11 | { 12 | 13 | string_map& getBuiltinOpImporterMap(); 14 | 15 | } // namespace onnx2trt 16 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/onnx2trt.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include "NvOnnxParser.h" 8 | #include "ShapedWeights.hpp" 9 | #include "Status.hpp" 10 | #include "TensorOrWeights.hpp" 11 | 12 | #include 13 | #include 14 | #include 15 | #include 16 | #include 17 | #include 18 | #include 19 | 20 | namespace onnx2trt 21 | { 22 | 23 | class IImporterContext; 24 | 25 | // TODO: Find ABI-safe alternative approach for this: 26 | // Can't use std::vector 27 | // Can't use ::onnx::NodeProto 28 | // Can't use std::function 29 | typedef ValueOrStatus> NodeImportResult; 30 | typedef std::function& inputs)> 32 | NodeImporter; 33 | 34 | template 35 | using StringMap = std::unordered_map; 36 | 37 | class IImporterContext 38 | { 39 | public: 40 | virtual nvinfer1::INetworkDefinition* network() = 0; 41 | virtual StringMap& tensors() = 0; 42 | virtual StringMap& tensorLocations() = 0; 43 | virtual StringMap& tensorRangeMins() = 0; 44 | virtual StringMap& tensorRangeMaxes() = 0; 45 | virtual StringMap& layerPrecisions() = 0; 46 | virtual std::unordered_set& unsupportedShapeTensors() = 0; 47 | virtual StringMap& loopTensors() = 0; 48 | virtual void setOnnxFileLocation(std::string location) = 0; 49 | virtual std::string getOnnxFileLocation() = 0; 50 | virtual void registerTensor(TensorOrWeights tensor, const std::string& basename) = 0; 51 | virtual void registerLayer(nvinfer1::ILayer* layer, const std::string& basename) = 0; 52 | virtual ShapedWeights createTempWeights(ShapedWeights::DataType type, nvinfer1::Dims shape, uint8_t value = 0) = 0; 53 | virtual int64_t getOpsetVersion(const char* domain = "") const = 0; 54 | virtual nvinfer1::ILogger& logger() = 0; 55 | virtual bool hasError() const = 0; 56 | virtual nvinfer1::IErrorRecorder* getErrorRecorder() const = 0; 57 | 58 | protected: 59 | virtual ~IImporterContext() 60 | { 61 | } 62 | }; 63 | 64 | } // namespace onnx2trt 65 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/onnx2trt_common.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include 8 | #include 9 | 10 | #if NV_TENSORRT_MAJOR < 4 11 | namespace nvinfer1 12 | { 13 | 14 | enum class PluginFormat : uint8_t 15 | { 16 | kNCHW = 0, //!< NCHW 17 | kNC2HW2 = 1, //!< NCHW with 2-element packed channels 18 | kNHWC8 = 2 //!< NHWC with 8-element packed channels (C 19 | //! must be a multiple of 8) 20 | }; 21 | // from NvInfer.h 22 | class IPluginExt : public IPlugin 23 | { 24 | public: 25 | virtual int getTensorRTVersion() const noexcept 26 | { 27 | return NV_TENSORRT_VERSION; 28 | } 29 | virtual bool supportsFormat(DataType type, PluginFormat format) const noexcept = 0; 30 | virtual void configureWithFormat(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, 31 | DataType type, PluginFormat format, int maxBatchSize) noexcept 32 | = 0; 33 | 34 | protected: 35 | void configure( 36 | const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, int maxBatchSize) noexcept final 37 | { 38 | try 39 | { 40 | DataType type = nvinfer1::DataType::kFLOAT; 41 | PluginFormat format = nvinfer1::PluginFormat::kLINEAR; 42 | return this->configureWithFormat(inputDims, nbInputs, outputDims, nbOutputs, type, format, maxBatchSize); 43 | } 44 | catch (const std::exception& e) 45 | { 46 | nvinfer1::getLogger()->log(nvinfer1::ILogger::Severity::kERROR, e.what().c_str()); 47 | } 48 | } 49 | virtual ~IPluginExt() 50 | { 51 | } 52 | }; 53 | 54 | } // namespace nvinfer1 55 | #endif 56 | 57 | namespace onnx2trt 58 | { 59 | 60 | struct IOwnable 61 | { 62 | virtual void destroy() = 0; 63 | 64 | protected: 65 | virtual ~IOwnable() 66 | { 67 | } 68 | }; 69 | 70 | struct OwnableDeleter 71 | { 72 | void operator()(IOwnable* obj) const 73 | { 74 | obj->destroy(); 75 | } 76 | }; 77 | 78 | using UniqueOwnable = std::unique_ptr; 79 | class Plugin; 80 | class PluginV2; 81 | 82 | } // namespace onnx2trt 83 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/onnx2trt_runtime.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include "onnx2trt_common.hpp" 8 | 9 | namespace onnx2trt 10 | { 11 | 12 | typedef Plugin* (*plugin_deserializer)(const void* serialData, size_t serialLength); 13 | 14 | } // namespace onnx2trt 15 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/onnxErrorRecorder.cpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #include "onnxErrorRecorder.hpp" 6 | #include 7 | 8 | namespace onnx2trt 9 | { 10 | 11 | 12 | ONNXParserErrorRecorder* ONNXParserErrorRecorder::create( 13 | nvinfer1::ILogger* logger, nvinfer1::IErrorRecorder* otherRecorder) 14 | { 15 | try 16 | { 17 | auto recorder = new ONNXParserErrorRecorder(logger, otherRecorder); 18 | if (recorder) 19 | { 20 | recorder->incRefCount(); 21 | } 22 | return recorder; 23 | } 24 | catch (const std::exception& e) 25 | { 26 | logError(logger, e.what()); 27 | return nullptr; 28 | } 29 | } 30 | 31 | void ONNXParserErrorRecorder::destroy(ONNXParserErrorRecorder*& recorder) 32 | { 33 | if (recorder) 34 | { 35 | recorder->decRefCount(); 36 | recorder = nullptr; 37 | } 38 | } 39 | 40 | void ONNXParserErrorRecorder::logError(nvinfer1::ILogger* logger, const char* str) 41 | { 42 | if (logger) 43 | { 44 | logger->log(ILogger::Severity::kERROR, str); 45 | } 46 | } 47 | 48 | ONNXParserErrorRecorder::ONNXParserErrorRecorder( 49 | nvinfer1::ILogger* logger, nvinfer1::IErrorRecorder* otherRecorder) 50 | : mUserRecorder(otherRecorder) 51 | , mLogger(logger) 52 | { 53 | if (mUserRecorder) 54 | { 55 | mUserRecorder->incRefCount(); 56 | } 57 | } 58 | 59 | ONNXParserErrorRecorder::~ONNXParserErrorRecorder() noexcept 60 | { 61 | if (mUserRecorder) 62 | { 63 | mUserRecorder->decRefCount(); 64 | } 65 | } 66 | 67 | void ONNXParserErrorRecorder::clear() noexcept 68 | { 69 | try 70 | { 71 | // grab a lock so that there is no addition while clearing. 72 | std::lock_guard guard(mStackLock); 73 | mErrorStack.clear(); 74 | } 75 | catch (const std::exception& e) 76 | { 77 | logError(mLogger, e.what()); 78 | } 79 | }; 80 | 81 | bool ONNXParserErrorRecorder::reportError( 82 | nvinfer1::ErrorCode val, nvinfer1::IErrorRecorder::ErrorDesc desc) noexcept 83 | { 84 | try 85 | { 86 | std::lock_guard guard(mStackLock); 87 | mErrorStack.push_back(errorPair(val, desc)); 88 | if (mUserRecorder) 89 | { 90 | mUserRecorder->reportError(val, desc); 91 | } 92 | else 93 | { 94 | logError(mLogger, desc); 95 | } 96 | } 97 | catch (const std::exception& e) 98 | { 99 | logError(mLogger, e.what()); 100 | } 101 | // All errors are considered fatal. 102 | return true; 103 | } 104 | 105 | nvinfer1::IErrorRecorder::RefCount ONNXParserErrorRecorder::incRefCount() noexcept 106 | { 107 | // Atomically increment or decrement the ref counter. 108 | return ++mRefCount; 109 | } 110 | 111 | nvinfer1::IErrorRecorder::RefCount ONNXParserErrorRecorder::decRefCount() noexcept 112 | { 113 | auto newVal = --mRefCount; 114 | if (newVal == 0) 115 | { 116 | delete this; 117 | } 118 | return newVal; 119 | } 120 | 121 | } // namespace onnx2trt 122 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/onnxErrorRecorder.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include "NvInferRuntimeCommon.h" 8 | #include "onnx2trt_utils.hpp" 9 | #include 10 | #include 11 | #include 12 | #include 13 | #include 14 | 15 | namespace onnx2trt 16 | { 17 | 18 | //! 19 | //! A simple implementation of the IErrorRecorder interface for 20 | //! use by ONNX importer. 21 | //! ONNX-importer Error recorder is based on a vector that pairs the error 22 | //! code and the error string into a single element. It also uses 23 | //! standard mutex and atomics in order to make sure that the code 24 | //! works in a multi-threaded environment. 25 | //! 26 | class ONNXParserErrorRecorder : public nvinfer1::IErrorRecorder 27 | { 28 | using RefCount = nvinfer1::IErrorRecorder::RefCount; 29 | using ErrorDesc = nvinfer1::IErrorRecorder::ErrorDesc; 30 | using ErrorCode = nvinfer1::ErrorCode; 31 | using IErrorRecorder = nvinfer1::IErrorRecorder; 32 | using ILogger = nvinfer1::ILogger; 33 | 34 | using errorPair = std::pair; 35 | using errorStack = std::vector; 36 | 37 | public: 38 | static ONNXParserErrorRecorder* create( 39 | ILogger* logger, IErrorRecorder* otherRecorder = nullptr); 40 | 41 | static void destroy(ONNXParserErrorRecorder*& recorder); 42 | 43 | void clear() noexcept final; 44 | RefCount incRefCount() noexcept final; 45 | RefCount decRefCount() noexcept final; 46 | bool reportError(ErrorCode val, ErrorDesc desc) noexcept final; 47 | 48 | int32_t getNbErrors() const noexcept final 49 | { 50 | return mErrorStack.size(); 51 | } 52 | 53 | ErrorCode getErrorCode(int32_t errorIdx) const noexcept final 54 | { 55 | return invalidIndexCheck(errorIdx) ? ErrorCode::kINVALID_ARGUMENT : (*this)[errorIdx].first; 56 | } 57 | 58 | ErrorDesc getErrorDesc(int32_t errorIdx) const noexcept final 59 | { 60 | return invalidIndexCheck(errorIdx) ? "errorIdx out of range." : (*this)[errorIdx].second.c_str(); 61 | } 62 | 63 | bool hasOverflowed() const noexcept final 64 | { 65 | // This class can never overflow since we have dynamic resize via std::vector usage. 66 | return false; 67 | } 68 | 69 | protected: 70 | ONNXParserErrorRecorder(ILogger* logger, IErrorRecorder* otherRecorder = nullptr); 71 | 72 | virtual ~ONNXParserErrorRecorder() noexcept; 73 | 74 | static void logError(ILogger* logger, const char* str); 75 | 76 | // Simple helper functions. 77 | const errorPair& operator[](size_t index) const noexcept 78 | { 79 | return mErrorStack[index]; 80 | } 81 | 82 | bool invalidIndexCheck(int32_t index) const noexcept 83 | { 84 | // By converting signed to unsigned, we only need a single check since 85 | // negative numbers turn into large positive greater than the size. 86 | size_t sIndex = index; 87 | return sIndex >= mErrorStack.size(); 88 | } 89 | // Mutex to hold when locking mErrorStack. 90 | std::mutex mStackLock; 91 | 92 | // Reference count of the class. Destruction of the class when mRefCount 93 | // is not zero causes undefined behavior. 94 | std::atomic mRefCount{0}; 95 | 96 | // The error stack that holds the errors recorded by TensorRT. 97 | errorStack mErrorStack; 98 | 99 | // Original error recorder (set by user) 100 | IErrorRecorder* mUserRecorder{nullptr}; 101 | 102 | // logger 103 | ILogger* mLogger{nullptr}; 104 | }; // class ONNXParserErrorRecorder 105 | 106 | } // namespace onnx2trt 107 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/readme.md: -------------------------------------------------------------------------------- 1 | # ONNX Parser 2 | - 这几个文件提取自官方的onnx-tensorrt，去掉python方面，其他都在 3 | - 另外增加了Plugin节点的支持 4 | - https://github.com/onnx/onnx-tensorrt -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/toposort.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include 8 | #include 9 | 10 | #include 11 | using std::cout; 12 | using std::cerr; 13 | using std::endl; 14 | 15 | namespace 16 | { 17 | 18 | enum NodeState 19 | { 20 | NODE_UNVISITED, 21 | NODE_ACTIVE, 22 | NODE_VISITED 23 | }; 24 | 25 | template 26 | bool get_post_order(size_t node_idx, Container const& nodes, std::unordered_map const& node_map, 27 | std::vector* node_states, std::vector* order) 28 | { 29 | NodeState& node_state = node_states->at(node_idx); 30 | if (node_state == NODE_ACTIVE) 31 | { 32 | // Cycle detected! 33 | cerr << "ERROR: Graph contains a cycle" << endl; 34 | return false; 35 | } 36 | else if (node_state == NODE_VISITED) 37 | { 38 | return true; 39 | } 40 | else 41 | { 42 | node_state = NODE_ACTIVE; 43 | // TODO: This .Get().input() is highly specific to protobuf, should 44 | // generalise it somehow. 45 | for (auto const& input : nodes.Get(node_idx).input()) 46 | { 47 | if (!node_map.count(input)) 48 | { 49 | // Input node not found in graph! 50 | // cerr << "ERROR: Input node not found in graph: " 51 | // << input << endl; 52 | // return false; 53 | continue; // Skip missing input edges 54 | } 55 | size_t input_node_idx = node_map.at(input); 56 | if (!get_post_order(input_node_idx, nodes, node_map, node_states, order)) 57 | { 58 | return false; 59 | } 60 | } 61 | node_state = NODE_VISITED; 62 | order->push_back(node_idx); 63 | } 64 | return true; 65 | } 66 | 67 | } // anonymous namespace 68 | 69 | template 70 | bool toposort(Container const& nodes, std::vector* order) 71 | { 72 | std::unordered_map node_map; 73 | for (size_t i = 0; i < (size_t) nodes.size(); ++i) 74 | { 75 | // TODO: This .Get().input() is highly specific to protobuf, should 76 | // generalise it somehow. 77 | for (auto const& output : nodes.Get(i).output()) 78 | { 79 | if (!node_map.emplace(output, i).second) 80 | { 81 | // Output name appears more than once in graph! 82 | cerr << "ERROR: Output name is not unique: " << output << endl; 83 | return false; 84 | } 85 | } 86 | } 87 | order->reserve(nodes.size()); 88 | std::vector node_states(nodes.size(), NODE_UNVISITED); 89 | for (size_t i = 0; i < (size_t) nodes.size(); ++i) 90 | { 91 | if (!get_post_order(i, nodes, node_map, &node_states, order)) 92 | { 93 | return false; 94 | } 95 | } 96 | return true; 97 | } 98 | -------------------------------------------------------------------------------- /src/tensorRT/onnx_parser/utils.hpp: -------------------------------------------------------------------------------- 1 | /* 2 | * SPDX-License-Identifier: Apache-2.0 3 | */ 4 | 5 | #pragma once 6 | 7 | #include 8 | 9 | template 10 | using string_map = std::unordered_map; 11 | -------------------------------------------------------------------------------- /src/tensorRT/onnxplugin/plugin_binary_io.cpp: -------------------------------------------------------------------------------- 1 | 2 | #include "plugin_binary_io.hpp" 3 | #include "ilogger.hpp" 4 | #include 5 | 6 | namespace Plugin{ 7 | 8 | using namespace std; 9 | 10 | BinIO::~BinIO(){ 11 | close(); 12 | } 13 | 14 | bool BinIO::opened(){ 15 | if (flag_ == MemoryRead) 16 | return memoryRead_ != nullptr; 17 | else if (flag_ == MemoryWrite) 18 | return true; 19 | return false; 20 | } 21 | 22 | void BinIO::close(){ 23 | if (flag_ == MemoryRead) { 24 | memoryRead_ = nullptr; 25 | memoryCursor_ = 0; 26 | memoryLength_ = -1; 27 | } 28 | else if (flag_ == MemoryWrite) { 29 | memoryWrite_.clear(); 30 | memoryCursor_ = 0; 31 | memoryLength_ = -1; 32 | } 33 | } 34 | 35 | string BinIO::readData(int numBytes){ 36 | string output; 37 | output.resize(numBytes); 38 | 39 | int readlen = read((void*)output.data(), output.size()); 40 | output.resize(readlen); 41 | return output; 42 | } 43 | 44 | int BinIO::read(void* pdata, size_t length){ 45 | 46 | if (flag_ == MemoryRead) { 47 | if (memoryLength_ != -1) { 48 | 49 | if (memoryLength_ < memoryCursor_ + length) { 50 | int remain = memoryLength_ - memoryCursor_; 51 | if (remain > 0) { 52 | memcpy(pdata, memoryRead_ + memoryCursor_, remain); 53 | memoryCursor_ += remain; 54 | return remain; 55 | } 56 | else { 57 | return -1; 58 | } 59 | } 60 | } 61 | memcpy(pdata, memoryRead_ + memoryCursor_, length); 62 | memoryCursor_ += length; 63 | return length; 64 | } 65 | else { 66 | return -1; 67 | } 68 | } 69 | 70 | bool BinIO::eof(){ 71 | if (!opened()) return true; 72 | 73 | if (flag_ == MemoryRead){ 74 | return this->memoryCursor_ >= this->memoryLength_; 75 | } 76 | else if (flag_ == MemoryWrite){ 77 | return false; 78 | } 79 | else { 80 | opstate_ = false; 81 | INFO("Unsupport flag: %d", flag_); 82 | return true; 83 | } 84 | } 85 | 86 | int BinIO::write(const void* pdata, size_t length){ 87 | 88 | if (flag_ == MemoryWrite) { 89 | memoryWrite_.append((char*)pdata, (char*)pdata + length); 90 | return length; 91 | } 92 | else { 93 | return -1; 94 | } 95 | } 96 | 97 | int BinIO::writeData(const string& data){ 98 | return write(data.data(), data.size()); 99 | } 100 | 101 | BinIO& BinIO::operator >> (string& value){ 102 | //read 103 | int length = 0; 104 | (*this) >> length; 105 | value = readData(length); 106 | return *this; 107 | } 108 | 109 | int BinIO::readInt(){ 110 | int value = 0; 111 | (*this) >> value; 112 | return value; 113 | } 114 | 115 | float BinIO::readFloat(){ 116 | float value = 0; 117 | (*this) >> value; 118 | return value; 119 | } 120 | 121 | BinIO& BinIO::operator << (const string& value){ 122 | //write 123 | (*this) << (int)value.size(); 124 | writeData(value); 125 | return *this; 126 | } 127 | 128 | BinIO& BinIO::operator << (const char* value){ 129 | 130 | int length = strlen(value); 131 | (*this) << (int)length; 132 | write(value, length); 133 | return *this; 134 | } 135 | 136 | BinIO& BinIO::operator << (const vector& value){ 137 | (*this) << (int)value.size(); 138 | for (int i = 0; i < value.size(); ++i){ 139 | (*this) << value[i]; 140 | } 141 | return *this; 142 | } 143 | 144 | BinIO& BinIO::operator >> (vector& value){ 145 | int num; 146 | (*this) >> num; 147 | 148 | value.resize(num); 149 | for (int i = 0; i < value.size(); ++i) 150 | (*this) >> value[i]; 151 | return *this; 152 | } 153 | 154 | bool BinIO::openMemoryRead(const void* ptr, int memoryLength) { 155 | close(); 156 | 157 | if (!ptr) return false; 158 | memoryRead_ = (const char*)ptr; 159 | memoryCursor_ = 0; 160 | memoryLength_ = memoryLength; 161 | flag_ = MemoryRead; 162 | return true; 163 | } 164 | 165 | void BinIO::openMemoryWrite() { 166 | close(); 167 | 168 | memoryWrite_.clear(); 169 | memoryCursor_ = 0; 170 | memoryLength_ = -1; 171 | flag_ = MemoryWrite; 172 | } 173 | 174 | }; // namespace Plugin -------------------------------------------------------------------------------- /src/tensorRT/onnxplugin/plugin_binary_io.hpp: -------------------------------------------------------------------------------- 1 | #ifndef PLUGIN_BINARY_IO_HPP 2 | #define PLUGIN_BINARY_IO_HPP 3 | 4 | #include 5 | #include 6 | 7 | namespace Plugin{ 8 | 9 | class BinIO { 10 | public: 11 | enum Head { 12 | MemoryRead = 1, 13 | MemoryWrite = 2 14 | }; 15 | 16 | BinIO() { openMemoryWrite(); } 17 | BinIO(const void* ptr, int memoryLength = -1) { openMemoryRead(ptr, memoryLength); } 18 | virtual ~BinIO(); 19 | bool opened(); 20 | bool openMemoryRead(const void* ptr, int memoryLength = -1); 21 | void openMemoryWrite(); 22 | const std::string& writedMemory() { return memoryWrite_; } 23 | void close(); 24 | int write(const void* pdata, size_t length); 25 | int writeData(const std::string& data); 26 | int read(void* pdata, size_t length); 27 | std::string readData(int numBytes); 28 | int readInt(); 29 | float readFloat(); 30 | bool eof(); 31 | 32 | BinIO& operator >> (std::string& value); 33 | BinIO& operator << (const std::string& value); 34 | BinIO& operator << (const char* value); 35 | BinIO& operator << (const std::vector& value); 36 | BinIO& operator >> (std::vector& value); 37 | 38 | template 39 | BinIO& operator >> (std::vector<_T>& value) { 40 | int length = 0; 41 | (*this) >> length; 42 | 43 | value.resize(length); 44 | read(value.data(), length * sizeof(_T)); 45 | return *this; 46 | } 47 | 48 | template 49 | BinIO& operator << (const std::vector<_T>& value) { 50 | (*this) << (int)value.size(); 51 | write(value.data(), sizeof(_T) * value.size()); 52 | return *this; 53 | } 54 | 55 | template 56 | BinIO& operator >> (_T& value) { 57 | read(&value, sizeof(_T)); 58 | return *this; 59 | } 60 | 61 | template 62 | BinIO& operator << (const _T& value) { 63 | write(&value, sizeof(_T)); 64 | return *this; 65 | } 66 | 67 | bool opstate() const { 68 | return opstate_; 69 | } 70 | 71 | private: 72 | size_t readModeEndSEEK_ = 0; 73 | std::string memoryWrite_; 74 | const char* memoryRead_ = nullptr; 75 | int memoryCursor_ = 0; 76 | int memoryLength_ = -1; 77 | Head flag_ = MemoryWrite; 78 | bool opstate_ = true; 79 | }; 80 | }; // namespace Plugin 81 | 82 | #endif //PLUGIN_BINARY_IO_HPP -------------------------------------------------------------------------------- /src/tensorRT/onnxplugin/plugins/HSigmoid.cu: -------------------------------------------------------------------------------- 1 | 2 | #include 3 | #include 4 | 5 | using namespace ONNXPlugin; 6 | 7 | static __global__ void hsigmoid_kernel_fp32(float* input, float* output, int edge) { 8 | 9 | KernelPositionBlock; 10 | float x = input[position]; 11 | float a = x + 3; 12 | a = a < 0 ? 0 : (a >= 6 ? 6 : a); 13 | output[position] = a / 6; 14 | } 15 | 16 | // static __global__ void hsigmoid_kernel_fp16(__half* input, __half* output, int edge) { 17 | 18 | // KernelPositionBlock; 19 | 20 | // __half _six = 6.0f; 21 | // __half _three = 3.0f; 22 | // __half x = input[position]; 23 | // __half a = x + _three; 24 | // __half _zero = 0.0f; 25 | // a = a < _zero ? _zero : (a >= _six ? _six : a); 26 | // output[position] = a / _six; 27 | // } 28 | 29 | class HSigmoid : public TRTPlugin { 30 | public: 31 | SetupPlugin(HSigmoid); 32 | 33 | virtual void config_finish() override{ 34 | 35 | // INFO("init hsigmoid config: %s", config_->info_.c_str()); 36 | // INFO("weights = %d", config_->weights_.size()); 37 | // for(int i = 0; i < config_->weights_.size(); ++i){ 38 | // auto& w = config_->weights_[i]; 39 | // if(w->type() == TRT::DataType::Float16){ 40 | // INFO("Weight[%d] shape is %s, dtype = %s, value[0] = %f", i, w->shape_string(), data_type_string(w->type()), float(w->at<__half>(0))); 41 | // }else{ 42 | // INFO("Weight[%d] shape is %s, dtype = %s, value[0] = %f", i, w->shape_string(), data_type_string(w->type()), w->at(0)); 43 | // } 44 | // } 45 | } 46 | 47 | virtual std::shared_ptr new_config() override{ 48 | auto cfg = TRTPlugin::new_config(); 49 | 50 | // cfg->support_dtype_set_ = {nvinfer1::DataType::kHALF, nvinfer1::DataType::kFLOAT}; 51 | cfg->support_dtype_set_ = {nvinfer1::DataType::kFLOAT}; 52 | return cfg; 53 | } 54 | 55 | virtual nvinfer1::DimsExprs getOutputDimensions( 56 | int32_t outputIndex, const nvinfer1::DimsExprs* inputs, int32_t nbInputs, nvinfer1::IExprBuilder& exprBuilder) noexcept override{ 57 | 58 | return inputs[0]; 59 | } 60 | 61 | int enqueue(const std::vector& inputs, std::vector& outputs, const std::vector& weights, void* workspace, cudaStream_t stream) override{ 62 | 63 | int count = inputs[0].count(); 64 | auto grid = CUDATools::grid_dims(count); 65 | auto block = CUDATools::block_dims(count); 66 | 67 | if (config_->usage_dtype_ == TRT::DataType::Float) { 68 | hsigmoid_kernel_fp32 <<>> (inputs[0].ptr(), outputs[0].ptr(), count); 69 | } 70 | else if (config_->usage_dtype_ == TRT::DataType::Float16) { 71 | //hsigmoid_kernel_fp16 <<>> (inputs[0].ptr<__half>(), outputs[0].ptr<__half>(), count); 72 | INFOF("not implement function"); 73 | } 74 | else{ 75 | INFOF("not implement function"); 76 | } 77 | return 0; 78 | } 79 | }; 80 | 81 | RegisterPlugin(HSigmoid); -------------------------------------------------------------------------------- /src/tensorRT/onnxplugin/plugins/HSwish.cu: -------------------------------------------------------------------------------- 1 | 2 | #include 3 | #include 4 | 5 | using namespace ONNXPlugin; 6 | 7 | static __global__ void hswish_kernel_fp32(float* input, float* output, int edge) { 8 | 9 | KernelPositionBlock; 10 | float x = input[position]; 11 | float a = x + 3; 12 | a = a < 0 ? 0 : (a >= 6 ? 6 : a); 13 | output[position] = x * a / 6; 14 | } 15 | 16 | // static __global__ void hswish_kernel_fp16(__half* input, __half* output, int edge) { 17 | 18 | // KernelPositionBlock; 19 | 20 | // __half _six = 6.0f; 21 | // __half _three = 3.0f; 22 | // __half x = input[position]; 23 | // __half a = x + _three; 24 | // __half _zero = 0.0f; 25 | // a = a < _zero ? _zero : (a >= _six ? _six : a); 26 | // output[position] = x * a / _six; 27 | // } 28 | 29 | class HSwish : public TRTPlugin { 30 | public: 31 | SetupPlugin(HSwish); 32 | 33 | virtual void config_finish() override{ 34 | 35 | // INFO("init hswish config: %s", config_->info_.c_str()); 36 | // INFO("weights = %d", config_->weights_.size()); 37 | // for(int i = 0; i < config_->weights_.size(); ++i){ 38 | // auto& w = config_->weights_[i]; 39 | // if(w->type() == TRT::DataType::Float16){ 40 | // INFO("Weight[%d] shape is %s, dtype = %s, value[0] = %f", i, w->shape_string(), data_type_string(w->type()), float(w->at<__half>(0))); 41 | // }else{ 42 | // INFO("Weight[%d] shape is %s, dtype = %s, value[0] = %f", i, w->shape_string(), data_type_string(w->type()), w->at(0)); 43 | // } 44 | // } 45 | } 46 | 47 | virtual std::shared_ptr new_config() override{ 48 | auto cfg = TRTPlugin::new_config(); 49 | 50 | //cfg->support_dtype_set_ = {nvinfer1::DataType::kHALF, nvinfer1::DataType::kFLOAT}; 51 | cfg->support_dtype_set_ = {nvinfer1::DataType::kFLOAT}; 52 | return cfg; 53 | } 54 | 55 | virtual nvinfer1::DimsExprs getOutputDimensions( 56 | int32_t outputIndex, const nvinfer1::DimsExprs* inputs, int32_t nbInputs, nvinfer1::IExprBuilder& exprBuilder) noexcept override{ 57 | 58 | return inputs[0]; 59 | } 60 | 61 | int enqueue(const std::vector& inputs, std::vector& outputs, const std::vector& weights, void* workspace, cudaStream_t stream) override{ 62 | 63 | int count = inputs[0].count(); 64 | auto grid = CUDATools::grid_dims(count); 65 | auto block = CUDATools::block_dims(count); 66 | 67 | if (config_->usage_dtype_ == TRT::DataType::Float) { 68 | hswish_kernel_fp32 <<>> (inputs[0].ptr(), outputs[0].ptr(), count); 69 | } 70 | else if (config_->usage_dtype_ == TRT::DataType::Float16) { 71 | // hswish_kernel_fp16 <<>> (inputs[0].ptr<__half>(), outputs[0].ptr<__half>(), count); 72 | INFOF("not implement function"); 73 | } 74 | else{ 75 | INFOF("not implement function"); 76 | } 77 | return 0; 78 | } 79 | }; 80 | 81 | RegisterPlugin(HSwish); -------------------------------------------------------------------------------- /yolov8_obb芯片引脚缺陷检测.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/xjhaz/yolov8_obb_ChipPinDefectDetection/11aa8aa73472a2436809f2f21208127a5836a912/yolov8_obb芯片引脚缺陷检测.pdf --------------------------------------------------------------------------------