skimming the paper > We implemented our approach on top of the GraalVM, a high-performance virtual machine for Java, and evaluated it with a set of Java and Ja

skimming the paper > We implemented our approach on top of the GraalVM, a high-performance virtual machine for Java, and evaluated it with a set of Java and JavaScript benchmarks in terms of peak performance, compilation time and code size increase. We show that our approach can improve performance by up to 150% while generating a median code size and compile-time increase of not more than 25%. Our results indicate that fast-path unrolling of non-counted loops can be used in practice to increase the performance of Java applications. 150% of non-counted loop speed or 150% of total application speed? i suspect the first. i wonder how prevalent those kind of loops are ... ah: > We evaluated our implementation on top of the GraalVM by running and analyzing a set of industry-standard benchmarks. for the Java DaCapo suite it looks like it didn't do much except making the variability of the jython runtime more unpredictable. throughput for javascript jetstream looks good.