McPAT Plug-in for Accelergy

An energy estimation plug-in for Accelergy framework. It is a wrapper for the McPAT integrated power, area, and timing framework. It works by substituting in attributes to the properties.xml definition file, running McPAT, and parsing the output. Since queries to McPAT can take some time, results are cached in the file .cache so repeated invocations are not repeated. To clear the cache delete the .cache file. Cache entries older than CACHE_TIMEOUT days are invalidated.

Get started

• Install Accelergy framework
• Download and build McPat 1.3

The McPAT executable directory can either be located in the plugin directory or on the PATH environment variable.

Use the plug-in

• Clone the repo by git clone https://github.com/Accelergy-Project/accelergy-mcpat-plug-in.git
• To set the relative accuracy of your McPat plug-in
  • open mcpat_wrapper.py
  • Edit the first line to set the MCPAT_ACCURACY (default is 80)
• Option 1
  • Run pip3 install . and use the same arguments as installing Accelergy
• Option 2
  • Open Accelergy's config file accelergy_config.yaml and add a new list item that points to the cloned folder
• Place the built McPat
  • Inside the installed McPat plug-in folder if you choose option 1 above/cloned McPat folder if you choose option2 above (or its subfolder)
  • Inside any folder (or its subfolder) that is included in the $PATH
• Run Accelergy (Accelergy's log will show that it identifies the McPat plug-in )

Supported components

• Branch predictor tournament_bp
• Branch target buffer btb
• Cache cache
• Crossbar xbar
• Decoder decoder
• Fetch buffer fetch_buffer
• Functional unit func_unit
• Instruction queue inst_queue
• Load store queue load_store_queue
• Register file cpu_regfile
• Renaming unit renaming_unit
• Reorder buffer reorder_buffer
• Translation lookaside buffer tlb