boost::compute这个库靠谱不？

fender0107401

既然能进boost，那应该是靠谱的吧。

windoze

太新，我脚着还是应该再等等。
话说刚进Boost的库一般都不太靠谱，至少要等上一两个版本才行。

fender0107401

回复 2# windoze
好吧，那我先试用一段时间，1.61才进boost的。

等我正式用在生产中，估计还有一段时间，希望能尽早变得靠谱。

zhujiang73

fender0107401 发表于 2017-02-08 21:48
既然能进boost，那应该是靠谱的吧。

      如果有需要也可以用这个： https://sourceforge.net/projects/viennacl/     

Features

    Three computing backends: CUDA, OpenCL, OpenMP
    Iterative Solvers: Conjugate Gradient, Stabilized BiConjugate Gradient, Generalized Minimum Residual
    Preconditioners: ICHOL, ILUT, ILU0, Block-ILU, AMG, (F)SPAI, Jacobi
    BLAS Level 1, Level 2 and Level 3 routines on GPUs and multi-core CPUs
    Fast sparse matrix-vector and sparse matrix-matrix products
    Convenient C++ wrappers for common linear algebra operations
    Fast Fourier transform
    C++ Interface is mostly uBLAS compatible
    Interfaces for uBLAS, Armadillo, Eigen and MTL 4
    Iterative Solvers can directly be used with uBLAS, Armadillo, Eigen and MTL4 objects
    Structured matrices: Circulant, Hankel, Toeplitz, Vandermonde
    OpenCL Kernel optimization environment for optimal performance on the target device
    Header-only library
    MATLAB interface for the iterative solvers (separate download)
    Python interface (PyViennaCL)