[PATCH] D39506: Add a script to run various benchmarks and send the result to lnt

[James Henderson via Phabricator via llvm-commits]

jhenderson added a comment.

Ok, the script looks fine to me. My one hesitation is the way the cset stuff works - it's pretty inflexible currently, but it gave me an idea. As I mentioned separately in the email thread, this feels like one of the "wrapper" programs I mentioned. Indeed, perf itself is as well, although I accept that this particular case is different, since the script needs to parse the output to store in the database.

To make it more general purpose, how about the following? We replace the "cset" switch with something like "wrapper". The switch then takes a list of strings that are a wrapper program and its arguments. The current behaviour would then be emulated by something like "--wrapper=cset,shield,-e" or whatever syntax is easiest (and this could be the default to reduce the amount of typing on Linux). More than one --wrapper could be specified and they'd be daisy-chained together, so something like "--wrapper=cset,shield,-e --wrapper=something,-somewhere=blah,-somehow" would lead to python running "sudo cset shield -e something -somewhere=blah -somehow perf -- stat ...". This would allow users to not use cset, if they wanted, and/or use other wrapper programs to configure other aspects of the LLD runtime environment.

How does that sound?


https://reviews.llvm.org/D39506