rend (v.) - (archaic) wrench something apart violently.
HPLRender tears apart HPL output files in an incredibly messy way to produce usable results. Also included are tools used for analysis of code.
analyze.py requires 2 arguments:
- --statistic PROPERTY, which sets the statistic that will be analyzed per bin
- an input file
...where PROPERTY is one of the following: encodedtime, n, nb, p, q, time, or gflops.
There is a third, optional argument you will most likely want to set named --bin. This argument is meant to describe how your results will be categorized. Like --statistic, you should provide it with PROPERTY. However, if --bin is not specified, it defaults to a null binning function which throws everything in the same bin.
There are also three more optional arguments:
- --output FILE, which allows you to specify an output file (outputs to stdout by default)
- --title TITLE, which allows you to set a custom title in the analysis results
- --verbose, which will add the binned results to the output
For instance, to analyze a run to see the time it took to solve each problem,
you would use ./analyze.py --statistic time --verbose --output timeTaken.txt HPL.out.
graph.py requires 3 arguments:
- --bin PROPERTY, which describes how your results will be categorized (this is your x-axis)
- --statistic PROPERTY, which sets the statistic that will be analyzed per bin (this is your y-axis)
- an input file
...where PROPERTY is one of the following: encodedtime, n, nb, p, q, time, or gflops. Unlike analyze.py, you cannot use the null binning function for --bin here, because every point would get plotted at x = 0, which wouldn't make any sense.
There's also the optional argument --title, which will add a run title in parentheses after the regular graph title.
For instance, to visualize how many gigaFLOPS the system pulled per problem for
several block sizes, you would use ./graph.py --bin nb --statistic gflops HPL.out.