[ENHANCEMENT] Do you plan to add other operator compiler?

microsoft / nnfusion

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

MIT License

968 stars 164 forks source link

[ENHANCEMENT] Do you plan to add other operator compiler? #204

Open FDInSky opened 3 years ago

FDInSky commented 3 years ago

🚀 Feature

I am reading the code, and find one operator need different kernel like cpu, gpu etc. May you can implement them use one unified format like Halide. Thanks

nnfbot commented 3 years ago

Thanks for the report @FDInSky! I will look into it ASAP! (I'm a bot).

xysmlx commented 3 years ago

Hi, @FDInSky thank you for your attention. NNFusion has supported auto kernel tuning by integrating Antares since the v0.2 release. NNFusion could implement one Antares IR for an operator (e.g., the translate_v2 function in the Dot generic operator define) and get kernel code for different hardware from the kernel tuner.