Calling Convention microbenchmark
tobias.hartmann at oracle.com
Wed Mar 13 13:11:11 UTC 2019
I wrote a simple benchmark to measure the impact of the calling/return convention optimization in
the optimal case when inlining fails due to a megamorphic interface call:
With default flags the throughput is 36.615 (± 2.668) ops/us on my machine and without the calling
and return convention optimizations it is 24.105 (± 0.899) ops/us. So with this benchmark, the
calling convention optimization improves throughput by 34%.
Of course it's just a microbenchmark so please take this with a grain of salt (one can probably
design a microbenchmark that shows the reverse). If there are no objections, I'll push this.
More information about the valhalla-dev