Comment on Why is 2 * (i * i) faster than 2 * i * i in Java?Comments−qwerty4561277yIMHO some kind of logic preprocesor should take care of this before the actual compilation.−isbvhodnvemrwvn7yHow? Java is compiled to bytecode, you don't know the architecture of the system the code is going to run on. It's one of the reasons javac only implements the simplest optimizations possible (constants folding and the like)−pjmlp7yCompiling to bytecode is just one of the possibilities.Since the early days of Java, OEM vendors targeting embedded targets do support AOT compilation, with possible PGO feedback.Some vendors like IBM, also provide similar capabilities on their regular Java toolchains.And Maxime finally graduated as Graal/Substrate, which is also another way of compiling Java.But all in all, everyone is transitioning to the benefits of bytecode as intermediate executable format.Even some cool LLVM optimizations, like ThinLTO, are only possible thanks to using bytecode.
Comments
IMHO some kind of logic preprocesor should take care of this before the actual compilation.
How? Java is compiled to bytecode, you don't know the architecture of the system the code is going to run on. It's one of the reasons javac only implements the simplest optimizations possible (constants folding and the like)
Compiling to bytecode is just one of the possibilities.
Since the early days of Java, OEM vendors targeting embedded targets do support AOT compilation, with possible PGO feedback.
Some vendors like IBM, also provide similar capabilities on their regular Java toolchains.
And Maxime finally graduated as Graal/Substrate, which is also another way of compiling Java.
But all in all, everyone is transitioning to the benefits of bytecode as intermediate executable format.
Even some cool LLVM optimizations, like ThinLTO, are only possible thanks to using bytecode.