Roadmap source: Official Github .5 has many changes, here are a few: Performance improvements include: improvements to regular expressions, string and char methods, HTTP 1.1/2 performance improvements, layered compilation improvements, using xmm registers to clear stack prolog to speed up performance...etc., more details in the PR project on Github.
Compilation is divided into immediate and hierarchical compilation. In . Core 3.0 or later, the compilation defaults to hierarchical compilation. In this released version, two major improvements have been made to hierarchical compilation. In layered compilation, the Call-counting china dataset mechanism plays an important role. Once a method is called multiple times, the Runtime will ask the JIT to try to recompile in a better way. However, from the results of performance analysis in the past, we can observe that the speed of Call-counting slows down the overall performance.
This new release of .5 finally improves this part of the problem and also enhances layered compilation. efficiency. Another improvement point is about the performance of methods containing loops in hierarchical compilation. For example, a method that is rarely called is placed in a loop that recurs one million times - this is of course a more extreme example. The most common example is Main.