Skip to content

v0.5

Pre-release
Pre-release
Compare
Choose a tag to compare
@zeratax zeratax released this 04 Mar 01:37
· 219 commits to master since this release
fd9f417

Changelog

Features

  • C Executor
  • Benchmarking
    • can benchmark CUDA and even works with rise-lang/executor
    • KernelArgCreator to easily benchmark with different KernelArg inputs
  • more example Kernels
  • JNI
    • more java and scala examples
    • junits tests
    • KernelArg Outputs are instantly reusable as Inputs see #89
  • added sbt project file

Changes

  • JNI
    • exceptions in case of illegal arguments (e.g. NULL)
    • split up KernelArg-class into BooleanArg, ByteArg, ShortArg, IntArg, LongArg, HalfArg, FloatArg, DoubleArg and PaddingArg
      • PaddingArg helps to easily pad matrices to work more easily with e.g. TensorCores
      • HalfArg will convert a float array with a CUDA Kernel
    • Java files were moved to a proper package: src/{java=>main/java/yacx}/