Cranelift code generation comes to Rust
(lwn.net)
from mac@programming.dev to rust@programming.dev on 24 Mar 2024 18:43
https://programming.dev/post/11880226
from mac@programming.dev to rust@programming.dev on 24 Mar 2024 18:43
https://programming.dev/post/11880226
threaded - newest
Didn’t read the rest. But this is clearly inaccurate, as most Rustaceans probably already know.
Cranelift can be used in release builds. The performance is not competitive with LLVM. But some projects are completely useless (too slow) when built with the debug profile. So, some of us use a special release profile where Cranelift backend is used, and debug symbols are not stripped. This way, one can enjoy a quicker edit/compile/debug cycle with usable, if not the best, performance in built binaries.
Please, send an email to lwn@lwn.net to report this issue to them, they usually fix things quickly.
I read the rest of the article, and it appears to have been partially written before support for codegen backends landing in cargo.
That “latest progress report” has the relevant info ;)
So, basically, you would add this to the top of
Cargo.toml
:Then add a custom profile, for example:
Then build with:
Another option is to compile dependencies with LLVM and optimizations, which will likely be done only once in the first clean build, and then compile your main binary with Cranelift, thus getting the juicy fast compile times without having to worry about the slow dependencies.
Yes. And to complete the pro tips, the choice of linker can be very relevant. Using
mold
would come recommended nowadays.So that “special release build” is the build you do debugging with. Shouldn’t you just modify the otherwise useless debug profile and turn on all the optimizations necessary to make it usable?
Well, obviously that will depend on what defaults (and how many?!) a developer is going to change".
doc.rust-lang.org/cargo/reference/profiles.html#d…
And the debug (
dev
) profile has its uses. It’s just not necessarily the best for typical day-to-day development in many projects.I actually use two steps of profile inheritance, with
-cl
(for Cranelift) inheriting from a specialrelease-dev
profile. A developer does not have to be limited in how they modify or construct their profiles.Forgot to mention, and this is tangentially related to my comments from yesterday:
Notes:
But the numbers only show a 20% speed increase!
The unattended reader will be left with the impression that Cranelift compiles 20% faster for a 2x slowdown. Some comments below the article confirms that.
What the article author missed (again) is that the biggest Cranelift wins come when used in release/optimized/multi-pass mode. I mention multi-pass because the author should have noticed that the (relatively old) 2020 research paper he linked to tested Cranelift twice, with one mode having the single-pass tag attached to it.
Any Rust user knows that slow builds (sometimes boringly so) are actually release builds. These are the builds where the slowness of LLVM optimizing passes is felt. And these are the builds where Cranelift shines, and is indeed orders of magnitude faster than LLVM.
The fact that Cranelift manages to build non-optimized binaries 20% faster than LLVM is actually impressively good for Cranelift, or impressively bad for LLVM, however you want to look at it.
And that is the problem with researches/authors with no direct field expertise. They can easily miss some very relevant subtleties, leading the readers to make grossly wrong conclusions.
Yeah, I’m no compiler engineer, but testing both release and debug builds is the minimum I’d do. That doesn’t even get into classes of optimizations, like loop unrolling, binary size, macros, or function inlining, which I also expect to be in a compiler comparison.