hanhwi / mdm-functional-simulator

MDM-functional-simulator

This directory includes code for running MDM with a functional simulator to collect trace.

The directory GPGPUsim-generating-trace is developed from GPGPUsim to collect instruction/memory trace for GPU applications. You need first run the Simulator in the function simulation mode. The trace is like "interval_info_i.txt" and "memory_access_i.txt". I equals to the kernel id.

Then copy the trace to the model directory. The scripts to run the performance model is included in DRAM_run. We implement a script for sweeping different DRAM bandwidth in this directory. In particular, generate_L2_access.py is used to analyze memory trace and generate miss information. interval_warp_E_DRAM_sensitivity.py is the main MDM model script including interval analysis and MSHR/NOC/DRAM model.

Most important, you need to configure the parameter (#warps/SM, #SMs, NOC/DRAM bandwidth, core_frequency, latency, etc.)in generate_L2_access.py and interval_warp_E_DRAM_sensitivity.py to run your own hardware platforms. L1_cache and L2_cache are also needed to configure.

You can also use your own cache-simulator to run the memory trace.

The codes may still need more optimization for cleaning and efficiency.

Welcome to join this project! Also welcome to further extend the MDM model!