search

buddy-compiler / buddy-benchmark

Benchmark Framework for Buddy Projects
Apache License 2.0
44 stars 33 forks source link

[OpOptimization]add IREE BatchNormalization #103

Closed zwshan closed 8 months ago

zwshan commented 9 months ago

the same as iree matmul

SForeKeeper commented 8 months ago

Why there're two copies of each file?

zwshan commented 8 months ago

Why there're two copies of each file?

I'm sorry, there are two identical copies because I forgot to delete some duplicate files while organizing the folders. In my current OSPP experiment, I evaluated three deep learning compilers: IREE, Dynamo, and TVM. After optimization by TVM, each operator showed improved performance. IREE resulted in mostly negative optimization for individual operators, and Dynamo had some bugs and limited performance improvement for individual operators. So, in summary, for individual operators, TVM outperforms the other two deep learning compilers, but the outcome for end-to-end acceleration of the entire model is still unknown.