我有一个物理模拟，该模拟依赖于通过Eigen对角较大的厄米特矩阵。我最近打开了MKL，它在Mac OSX上可以很好地加速对角化。但是我无法在具有相同代码的Ubuntu上使用它（即，我没有速度提高，也没有多线程）。
我尝试了稀疏和密集矩阵表示，以及ComplexEigenSolver和SelfAdjointEigenSolver。
我在Ubuntu上使用Eigen 3.3.3和MKL 2018.u1，并遵循了此页面上的指示：https://eigen.tuxfamily.org/dox/TopicUsingIntelMKL.html和其他Eigen文档。
编译和链接行为：
g++ -I/home/.../Development/physics/googletest/googletest/include -I/home/.../Development/googletest/googletest/include -I/home/.../lapack-3.7.1/LAPACKE/include -I/opt/intel/compilers_and_libraries/mac/mkl/include -I/usr/local/include/eigen3 -I/home/.../boost-inst/include -I/home/.../eigen3 -I/usr/local/include/boost -I/usr/local/include -O3 -Wall -Wno-sign-compare -std=gnu++11 -fopenmp -I/include -c -o obj/MajoranaKitaevHamiltonian.o src/MajoranaKitaevHamiltonian.cc
g++ obj/main.o obj/defs.o obj/kitaev.o obj/BasisState.o obj/DisorderParameter.o obj/KitaevHamiltonian.o obj/KitaevHamiltonianBlock.o obj/Spectrum.o obj/MajoranaDisorderParameter.o obj/NaiveMajoranaKitaevHamiltonian.o obj/MajoranaKitaevHamiltonian.o obj/Correlators.o obj/Timer.o obj/FockSpaceUtils.o obj/RandomMatrix.o obj/TSVFile.o -L/home/.../Development/physics/googletest/googletest/include -L/share/software/user/restricted/imkl/2018/mkl/lib/intel64 -L/share/software/user/restricted/imkl/2018.u1/lib/intel64/ -L/opt/intel/compilers_and_libraries/mac/mkl/lib -L/share/software/user/open/boost/1.64.0/lib -L/usr/local/lib -lboost_program_options -lboost_iostreams -lboost_system -lboost_random -lmkl_core -lmkl_intel_thread -lmkl_core -lmkl_intel_ilp64 -liomp5 -fopenmp -o kitaev
（冗余路径同时支持Mac和Ubuntu平台）