This patch adds a benchmark command to the telamon CLI, allowing to
re-run benchmarks (both of the generated code, and the reference code)
for a given kernel and replay file after-the-fact.
This will print the same information as in the benchmark.txt generated
during the search, namely the average runtime of the benchmarked kernel
and the reference. The bound for the fixed candidate generated is also
displayed.
A batch mode is also provided, for running the same kernel with
different replays. In that case, the output (to stdout) is a .csv
containing one line per replay, with the raw measurements.
This patch adds a
benchmark
command to the telamon CLI, allowing to re-run benchmarks (both of the generated code, and the reference code) for a given kernel and replay file after-the-fact.This will print the same information as in the
benchmark.txt
generated during the search, namely the average runtime of the benchmarked kernel and the reference. The bound for the fixed candidate generated is also displayed.A batch mode is also provided, for running the same kernel with different replays. In that case, the output (to stdout) is a .csv containing one line per replay, with the raw measurements.