OOMIR optimiser stage

[Plixo2]

Why even bother with a OOMIR optimiser, as the JVM and R8 will optimize it anyway?

Even the OpenJDK compiler only does some constant folding and simple optimizations that the JVM will do anyways.

At least i strongly think so. Do you have any real world measurements? Or are there other goals for this stage?

[IntegralPilot]

Hi! The stage mainly exists because if we can cut down the amount of OOMIR (and thus JVM bytecode), it reduces the time taken by lower2 and r8 as they have less bytecode (r8) or OOMIR (lower2) to handle.

An interesting quirk you can observe is that on the CI the release mode tests are faster than debug mode tests - because the time savings by rustc's frontend are cancelled out by the extra time taken by lower2 and r8 to deal with all the extra code generated because it's unoptimised by rustc's frontend (optimise1 is run for both debug/release modes, but this shows that optimisation at a higher level such is key to cutting down build times at lower levels).

There are also optimisations that would be very difficult for r8 or the JVM to perform at the low level of JVM bytecode, that we can more easily see at a higher level of abstraction (though currently the optimise1 tasks are very simple things that r8/JVM could do later, more advanced analysis is planned), but the purpose is mainly decreasing build times.

Is there a better way you can think of for this? Thanks so much for your suggestion & help! :)