主要更新

  • ArrayFire现在支持OPENCL 设备  Intel Xeon Phi coprocessors,AMD APUs and AMD GPUs
  • 改善提高文档
  • Windows 例子工程文件
  • 新GPUs架构优化 (NVIDIA Kepler and Titan)

ArrayFire OpenCL

Function Additions

API Changes

  • Changes to enums
  • Deprecated
    • array(A, dims) and array(A, m,n,p,q) for changing dimensions of A

Feature Improvements

Bug Fixes

  • anytrue() and alltrue()
    • Fixed bug when reducing along dimensions with only one element
  • minmax()
    • Fixed bug when reduction dimension is greater than input dimensions
  • seq()
  • plot2(Y, X)
    • X axis now scales properly
  • Fixed issues arising with newer CUDA drivers
  • Fixed memory management issues in multi gpu environment
  • loadimage() and saveimage()
    • Fixed bugs related to windows
  • matmul()
    • Bug fixed when inputs are of different complexity and precision
  • median()
    • Fixed value error for vector inputs of even length
  • grad() for non square images
  • reorder()
  • Binary arithmetic operations on arrays of type af::b8 and scalars work as expected
    • Was casting the scalar to bool earlier
  • var() and cov() no longer cast downcast inputs
  • erode() and dilate()
    • Fixed erros in boundary conditions
    • Fixed potential bug when mask has zeros

Performance Improvements

  • transpose and conjugate transpose
    • overall improvements in performance
  • matmul()
    • Overall performance improvements
    • Additional options to reduce overhead
  • convolve()
    • Two dimensional convolution faster for kernel size <= 11
  • grad() rewritten to increase performance

Other changes