[PATCH] D140271: [NFCI][llvm-exegesis] Benchmark: parallelize codegen (5x ... 8x less wallclock)

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

Additionally, i would like to note a well-known fact: exegesis makes *very* liberal use of randomness.
Personally, i've found that a single measurement run (a single llvm-exegesis invocation) is not quite useful,
and practically always do several runs, and concatenate the yamls. That e.g. filters out some of
the "oh but why is this instruction's schedule suddenly not match measurements?" So there goes some of that
performance headroom. But sure, everyone has 10 hours to wait while `sleep(100); a+=1; sleep(100)` finishes...


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D140271/new/

https://reviews.llvm.org/D140271