Document how to compare benchmarks from two different source repositories

[gasche]

To avoid having to make world.opt each time I switch between the pristine branch and the optimization-enabled branch, I often use two distinct copies of the source directories, one in each branch (with two shells open, one in each corresponding switch).

The operf-micro webpage seems to assume that the two test runs to compare have been run from the same directory, and thus share the same .operf directory. It does not indicate how to compare test across .operf directories, or how to specify at init/build time another place to place or find .operf than the current directory.

Would it be possible to document the workflow where the two compiler versions to compare are located under distinct repository paths?

[chambart]

It is the case. There are two .operf directories. When you run some benchmarks, the compilation happens in a local one, and the results are stored in $(HOME)/.operf/micro/. When you run operf-micro compare it compares all the results that are stored in your home directory. This command does not need to run where you ran the benchmarks.

It effectively does not seem obvious from the documentation, I will try to improve that.

[gasche]

Thanks for the clarification. Maybe the compiler-local .operf directory could be named .operf-build to avoid the confusion?