qbe-rs VS minivm

> I think it uses a different backend than LLVM

harec uses https://c9x.me/compile/

Have you considered QBE?

I really hope other new projects (like QBE) can really grow and become widely used

I really liked targetting QBE (https://c9x.me/compile/) as an IR, as it gave me lots of back-end optimisations for free 😊.

There is: QBE.

You might be underestimating the accuracy of the CPU models LLVM uses.

For x86, the same data the code generator uses drives llvm-mca[1], which given a loop body can tell you the throughput, latency, and microarchitectural bottlenecks (decoding, ports, dependencies, store forwarding, etc.)—if not always precisely, then still not worse then IACA, the tool written at Intel by people who presumably knew how the CPUs work, unlike LLVM contributors and the rest of us who can only guess and measure. This separately for Haswell, Sandy Bridge, Skylake, etc.; not “x86”.

Now, is this the best model you can get? Not exactly[2], but it’s close enough to not matter. Do we often need machine code to be optimized to that level of detail? Perhaps not[3], and with that in mind you can shave at least a factor of ten off LLVM’s considerable bulk at the cost of 20—30% of performance[4,5]. But if you do want those as well, it seems that the complexity of LLVM is a fair price, or has the right order of magnitude at least.

(Frontend not included, C++ frontend required to bootstrap sold separately, at a similar markup compared to a C-only frontend with somewhat worse ergonomics.)

[1] https://llvm.org/docs/CommandGuide/llvm-mca.html

[2] https://www.uops.info/

[3] https://briancallahan.net/blog/20211010.html

[4] https://c9x.me/compile/

[5] https://drewdevault.com/talks/qbe.html

You can try buildling you own backend like llvm. A good example or starting point is probably QBE since it is extremely small but very functional.

How big is the whole backend? I've heard that it is small but I wanted to compare it to QBE which is around 8 KLoC and it is quite interesting too.

> I developed a unique way to write interpreters based on threaded code jumps and basic block versioning when I made MiniVM (https://github.com/FastVM/minivm). It was both larger and more dynamic than WebAssembly.

I'd be very interested to read more about this. It looks like you are using "one big function" with computed goto (https://github.com/FastVM/Web49/blob/main/src/interp/interp....). My experience working on this problem led me to the same conclusion as Mike Pall, which is that compilers do not do well with this pattern (particularly when it comes to register allocation): http://lua-users.org/lists/lua-l/2011-02/msg00742.html

I'm curious how you worked around the problem of poor register allocation in the compiler. I've come to the conclusion that tail calls are the best solution to this problem: https://blog.reverberate.org/2021/04/21/musttail-efficient-i...

Maybe take a look at MiniVM[0]? It was on HN a couple months ago[1].

[0]: https://github.com/fastvm/minivm

MiniVM: A zero-dependency cross-language runtime on par with LuaJIT and C\ (19 comments)