This is the code for SEAL.
To run the experiments, you need:
- Build TVM (we tested on the version published on Feb.21, 2022, we use Python 3.8.10).
- Build DietCode (in the main branch).
- Run
nohup python3 solution0.py --cuda 0 > solution0_nohupout0.log 2>&1 &
in TVM environment. - Switch to DietCode environment,
go to the folder
DietCode/tests/codegen
, copy the generated schedules to this foler, and runCUDA_VISIBLE_DEVICES=0 python3 run_local_padding_server.py
to measure latency.