move enzyme flags from general cargo to rustc-specific cargo

[onur-ozkan]

Resolves #130637.

Tracking:

• Tracking Issue for autodiff #124509

[rustbot]

r? @Kobzol

rustbot has assigned @Kobzol.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[ZuseZ4]

I can confirm that cherry-picking this pr (and a minor fix due to being 2 weeks behind) worked for the full rustc-ad fork, I could run all of our (not yet upstreamed) tests: https://github.com/rust-lang/rust/pull/130648/files, so it would be great if this gets approved, since it will enable running enzyme in the compiler explorer.

[Kobzol]

@bors r+

[bors]

📌 Commit f23d0b9 has been approved by Kobzol

It is now in the queue for this repository.

[ZuseZ4]

We don't (currently) build Enzyme/autodiff by default, so I think this could just be rollup=allways, since it won't affect the CI builds.

[onur-ozkan]

We don't (currently) build Enzyme/autodiff by default, so I think this could just be rollup=allways, since it won't affect the CI builds.

We usually rollup PRs unless they are explicitly tagged with rollup=never/iffy.

[bors]

⌛ Testing commit f23d0b9 with merge 80aa6fa...

[ZuseZ4]

@onur-ozkan this was a good improvement, but I think it still needs to become a bit more precise, see e.g. the cranelift build failure: https://github.com/rust-lang-ci/rust/actions/runs/14826601681/job/41620576833

What would be the best way to solve this? Is cranelift trying to bootstrap itself here without LLVM and hence Enzyme? At least it looks suspicious that exactly the one project not needing llvm fails. At least locally I have Enzyme in a few locations, so generally it shouldn't be too hard to specify a suitable link even in /rustlib, where cranelift mostly seems to live. The strange thing is that I can't find a code path in the source code which would end with us adding -lEnzyme-20, without also adding the corresponding -L flag.

./stage0-sysroot/lib/rustlib/x86_64-unknown-linux-gnu/lib/libEnzyme-20.so
./stage1/lib/libEnzyme-20.so
./stage1/lib/rustlib/x86_64-unknown-linux-gnu/lib/libEnzyme-20.so
./enzyme/lib/libEnzyme-20.so
./enzyme/build/Enzyme/libEnzyme-20.so

[onur-ozkan]

I don't have enough time to dive into this context these days.

The strange thing is that I can't find a code path in the source code which would end with us adding -lEnzyme-20, without also adding the corresponding -L flag.

The version part is handled programmatically; it's not hardcoded. You can search for Enzyme- in the source code to find them.