You will first need to download a large set of blender files (.blend). Some websites to find them include http://www.weybec.com/portfolio-item/crazyglue/ and https://orange.blender.org/download/. You will probably want to find additional ones online.
Then, once they are all in a single folder (can be in subfolders within that folder), call:
./prepare <nameOfBlendFilesFolder> <SPECblenderPath>
which will determine which ones are valid, put them into the created folder "valid", and set up all needed info files.
For ones that are invalid, it will copy their log into the folder invalid so you can see what went wrong. These files will be invalid because of either missing functionality from the original blender, or these files will depend on other files that are not available.
This can also be called multiple times on different folders, and is able to handle files with the same name.
The files sortedmemory.txt and sortedtime.txt will be created, and they each contain information on the amount of time or maximum memory used by the program.
To create a workload, there are a few different options you have:
You can call:
./select_random.py <numGroups> <numberPerGroups>
to randomly create
You can call:
./select_size.py <sortedOn> <sizeOrdering> <numberKept>
where
You can also hand select blend files based on their memory or time to, for example, test if there is any commonalities between blend files that use the same amount of memory.
To customize from which frame the workload will start, end, or how often it will generate a frame between start and end, edit the file run_options.txt. Of course, if you change these times then the times will be different from the sorted times, but it appears to just increase by a multiple of how many are created. To change the start, change the number after -s, to change the end, change the number after -e, and to change how often it will generate an image, change the number after -j.