This repository provides code for exploring LUT-based tensor core acceleration, including hardware synthesis and GPU-based simulation.
A synthesizable LUT-based tensor core array design.
To run synthesis:
- Set your own Design Compiler and TSMC library paths in
./scripts/dc.tcl. - Follow the instructions in the
svdirectory.
Includes a customized configuration targeting NVIDIA A100 GPUs.
For tuning details and simulation workflow, see accel-sim/README.md.
To be released in our upcoming work.