Fix AntaresCpuKernelEmitter and add ir_based_fusion in GENERIC_CPU backend

microsoft / nnfusion

A flexible and efficient deep neural network (DNN) compiler that generates high-performance executable from a DNN model description.

MIT License

959 stars 163 forks source link

Closed xysmlx closed 2 years ago

xysmlx commented 2 years ago

Update AntaresCpuKernelEmitter for Antares v0.2.x
Add IRBasedFusion for GENERIC_CPU backend
Performance could match Antares when the compilation environments (e.g., compiler, compile flags) of the generated model code and Antares kernel tuning are the same.
Not solved: Antares returns multiple kernels for an op