Phase 3 - Compilation

[epilys]

Hello scryer community,

I am still familiarizing myself with the codebase and current ongoing developments. What is the story on ahead of time / just in time compilation (phase 3)?

Is the WAM implementation mostly stable enough to start toying around with this idea?

A possible first step into AOT compilation into binaries would be defining a WAM -> custom Intermediate Representation that can then be transformed to some other IR (for example LLVM's) to generate native code.

[UWN]

There are many leaks and the WAM code generation needs some refinement, see #1269

[triska]

In addition, here are two issues with examples where WAM instructions are created completely superfluously #1212, #761.

And here are examples where different WAM instructions can be generated to yield better performance: #715, #723.

In such cases, improving the generated WAM code itself may trump any performance improvements that can be obtained by native code, especially in situations where instructions can be omitted entirely.

[triska]

#1590 is another issue that could be very interesting to analyze if you are interested in improving performance of the engine! The included comparison shows that even a system without JIT-compilation can perform ca. 20 times faster than Scryer Prolog in this example, a factor that is unlikely to be achieved by JIT-compilation of the currently generated code alone! The example is reasonably self-contained, and improving it would automatically improve a large set of programs that use library(clpz).