LLVM 12.0-RC1 Available For Testing This Latest Open-Source Compiler

[phoronix]

Phoronix: LLVM 12.0-RC1 Available For Testing This Latest Open-Source Compiler

Following the LLVM 12 code branching earlier this week, the first release candidate of the forthcoming LLVM 12.0 is now available for testing...

http://www.phoronix.com/scan.php?page=news_item&px=LLVM-12.0-RC1-Released

[Raka555]

If the trend continues it will be slower and more bloated again as the previous version with little to no improvement in the code it generates ...

[kpedersen]

If the trend continues it will be slower and more bloated again as the previous version with little to no improvement in the code it generates ...

Heh, tell me about it. I am working on an OpenBSD port for Emscripten (using LLVM from git master) and it takes a painful amount of time to build each time I run a whole test.

That said, GCC certainly isn't the smallest thing either (I also prefer LLVM's build system). Something about compilers seems to make them grow! Bring back Quake III's fork of LCC. That was beautiful to work on.

My Emscripten port is currently running into memory issues during the WASM binaryen stage. And that is node.js which is also bloomin fat as hell. Nothing new seems to be small and elegant anymore quite frankly.

[wizard69]

...Nothing new seems to be small and elegant anymore quite frankly.

That is because everybody wants their latest and greatest ideas integrated into these products. Well baring a legitimate regression someplace, there does seem to be people everywhere trying to fatten up that, that should remain thin. I'm seeing this in the Python world too, if there isn't a stop put to it our once light and elegant Python will become bloated with all sorts of crap. All to have a feature for 1% of the users base!!!!!!

Now given that thee is at least some goodness going into LLVM/CLang so hopefully they can address some of these performance issues. I don't want to sound like a Luddite but there is a fine line between bloat (extending a language too far beyond its original vision) and a rational feature set addition. C++ itself and the language standard in my opinion went beyond the bloat stage some time ago which is impacting all compilers. It is one reason why I'm kinda hoping that Swift gets a wider adoption. C++ code these days looks like hell and that is directly related to trying to pack too much into the language in my opinion.

[klapaucius]

I wish clangd was able to accept std::ranges instead of flagging them as errors. As of today, ranges support is still missing in clang... :-(

[Grinch]

If the trend continues it will be slower and more bloated again as the previous version with little to no improvement in the code it generates ...

Not my experience, while the compiling has indeed gotten slower, the performance of the generated code has improved and is now slightly ahead of GCC on the code I benchmark on my two systems (core i5, ryzen 3700x). I use PGO and LTO though, I'm not sure what the situation is if you are only using -O3 -march=native.