This repository contains the source code for reproducing the experiments in the paper "Cache-locality Based Adaptive Warp Scheduling for Neural Network Acceleration on GPGPUs" in IEEE 35th International System-on-Chip Conference.
This work based on GPGPU-Sim v3.2.0
You can find GPGPU-Sim v3.2.0 in https://hub.docker.com/r/tgrogers/gpgpu-sim_regress
- Ubuntu 20.04
- GPGPU-Sim v3.2.0
- CUDA 11.0
- gcc 5.5
The benchmark used in this work.
- ISPASS, https://github.com/gpgpu-sim/ispass2009-benchmarks.git
- Parboil, http://impact.crhc.illinois.edu/parboil/parboil.aspx
- PolyBench, http://web.cs.ucla.edu/~pouchet/software/polybench/
- Rodinia-3.1, https://www.cs.virginia.edu/rodinia/doku.php
- Tango, https://gitlab.com/Tango-DNNbench/Tango/-/tree/master/
Use src
to replace the src
directory in gpgpu-sim_distribution you download.
Use configs/tested-cfgs/SM2_GTX480/gpgpusim.config
when you run GPGPU-Sim. We add mode -gpgpu_scheduler caws
.
modified file:
- shader.cc. shader.h
- gpu-sim.cc, gpu-sim.h
- gpu-cache.cc, gpu-cache.h
- configs/tested-cfgs/SM2_GTX480/gpgpusim.config