[Instructions on how to do github pull of code and/or singularity container]
[Instructions on how to run run-perf-testsuit.sh]
Raw data process
- Once output from parflow simulation has been verified as correct, then do not need to save that.
- Need to save all raw performance data.
- [Need a script that takes raw performance data and creates a master csv file]
Summary csv file format
- [Ask Michael to propose this]
- Not specific to a particular plot, so contains all of the raw performance data.
Plots we forsee needing
- execution time versus time stamp on the commit for all the cores of a node on a particular machine with one series per OpenMP, CUDA, and MPI. One of these graphs per (tcl file, forcing data, machine)
- [all plots used in Michael's thesis. Michael please list here.]
performanceData/ [people can do pull requests to share their data] [each subdirectory is going to have very specific structure, one per pull request?]
runscripts/ run-perf-testsuit.sh, run some default configuration for all of the benchmarks, will have a usage message that describes how to run the script and how all of the default configurations and parameters were selected
benchmarks/ tcl files for each benchmark and a parameter handling script if parameterizable
sinusoidal.tcl sinusoidal-run.sh