Running examples: Dependency problems, Linux target, errors in codegen_nvvm

[welahi]

System: Alma Linux 8.10
CUDA: 12.6.3

I cloned the Rust-CUDA and tried to compile examples/cuda/cpu/add using cargo run, having Rust nightly installed. The rust-toolchain file makes rustup install 1.59.0-nightly. However, the log-crate, which is required in version 0.4.25, needs Rust 1.60.0 or higher.

Changing the toolchain nightly-2022-04-07 leads to a similar error, where openssl-sys v0.9.104 needs 1.63.0 or higher.

Then, I removed the toolchain-file to just use the newest version of the rustc. The problem with the dependencies was solved, but the buildscript of rustc_codegen_nvvm nags about the target. In the buildscript I found this uncommented match arm

// NOTE(RDambrosio016): currently disabled because of weird issues with segfaults and building the C++ shim
// "x86_64-unknown-linux-gnu" => "linux-x86_64",

Is Linux basically not supported?

I removed the comment and tried again, this time getting a bunch of errors while trying to compile codegen_nvvm. Some of them where solved running rustup component add rust-src rustc-dev llvm-tools-preview. However, I'm stuck here now:

1. I still get the error that the extern crate rustc_attr could not be found, and the suggested solution is again to install the llvm tools, which I already did.
2. More importantly, I get 104 other errors like use rustc_codegen_ssa::traits::BaseTypeMethods;: no BaseTypeMethods in traits in crates/rustc_codegen_nvvm/src/abi.rs:9:5, which I have no idea how to solve.

Does anyone know how to resolve the issues?

[jorge-ortega]

Hi @welahi. We've just got done updating the toolchain to night-2025-03-02. Do the latest changes resolve your issues?

[welahi]

Thanks for rebooting the project :) Disclaimer: I needed Rust-CUDA for a student project, which is over now; feel free to de-prioritize my issue.

rust_codegen_nvvm still makes problems. I've setup a new clean environment (new rust/cargo nightly installation, cloned Rust-CUDA repo just right now). I cdd into examples/cuda/cpu/add and run cargo build.

Please note that I'm still working on Linux, so i had to manually uncomment the "x86_64-unknown-linux-gnu" => "linux-x86_64" branch arm in line 57 in build.rs of rust_codegen_nvvm.

It seems that there is some trouble with finding the shared library, which was supposed to be build. Here's the detailed error message:

error: failed to run custom build command for `add v0.1.0 (/my/path/to/Rust-CUDA/examples/cuda/cpu/add)`

Caused by:
  process didn't exit successfully: `/my/path/to/Rust-CUDA/target/debug/build/add-136490b7a26d28dc/build-script-build` (exit status: 101)
  --- stdout
  cargo:rerun-if-changed=../../gpu/add_gpu

  --- stderr
  error: failed to run `rustc` to learn about target-specific information

  Caused by:
    process didn't exit successfully: `/my/path/to/.rustup/toolchains/nightly-2025-03-02-x86_64-unknown-linux-gnu/bin/rustc - --crate-name ___ --print=file-names -Zcodegen-backend=/my/path/to/Rust-CUDA/target/debug/deps/librustc_codegen_nvvm.so '-Zcrate-attr=feature(register_tool)' '-Zcrate-attr=register_tool(nvvm_internal)' -Zcrate-attr=no_std -Zsaturating_float_casts=false '-Cllvm-args=-arch=compute_61 --override-libm' --target nvptx64-nvidia-cuda --crate-type bin --crate-type rlib --crate-type dylib --crate-type cdylib --crate-type staticlib --crate-type proc-macro --print=sysroot --print=split-debuginfo --print=crate-name --print=cfg -Wwarnings` (exit status: 1)
    --- stderr
    error: couldn't load codegen backend /my/path/to/Rust-CUDA/target/debug/deps/librustc_codegen_nvvm.so: libnvvm.so.4: cannot open shared object file: No such file or directory


  thread 'main' panicked at crates/cuda_builder/src/lib.rs:529:10:
  Did not find output file in rustc output
  note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

[jorge-ortega]

I'm only one windows, but have encountered similar issues where the codegen backend failed to load due to my environment not being setup correctly. The docker file in our repo has a step in it to configure the loader to include the nvvm libs (see this line). Could you try that and see if that works?

[jorge-ortega]

I hit this today in our ubuntu24 container. I've tried adding the path to LD_LIBRARY_PATH and to ldconfig, but it's still not picking it up.

# ldconfig -p | grep libnvvm
        libnvvm.so.4 (libc6,x86-64) => /usr/local/cuda/nvvm/lib64/libnvvm.so.4
        libnvvm.so (libc6,x86-64) => /usr/local/cuda/nvvm/lib64/libnvvm.so
# ldd /usr/local/cuda/nvvm/lib64/libnvvm.so
        linux-vdso.so.1 (0x00007ffc6c856000)
        libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f3542005000)
        librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f3542000000)
        libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f3541ffb000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f3541f12000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f3541d00000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f3545e31000)

Can anyone else try running cargo build -p add in the ubuntu24 container? I'm still trying to figure out how to trace why it fails to load, but haven't had any luck. @adamcavendish would be great if you could also repro this.

[adamcavendish]

I hit this today in our ubuntu24 container. I've tried adding the path to LD_LIBRARY_PATH and to ldconfig, but it's still not picking it up.

# ldconfig -p | grep libnvvm
        libnvvm.so.4 (libc6,x86-64) => /usr/local/cuda/nvvm/lib64/libnvvm.so.4
        libnvvm.so (libc6,x86-64) => /usr/local/cuda/nvvm/lib64/libnvvm.so
# ldd /usr/local/cuda/nvvm/lib64/libnvvm.so
        linux-vdso.so.1 (0x00007ffc6c856000)
        libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f3542005000)
        librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f3542000000)
        libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f3541ffb000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f3541f12000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f3541d00000)
        /lib64/ld-linux-x86-64.so.2 (0x00007f3545e31000)

Can anyone else try running cargo build -p add in the ubuntu24 container? I'm still trying to figure out how to trace why it fails to load, but haven't had any luck. @adamcavendish would be great if you could also repro this.

Sure. I'll get add example revived later today.

[welahi]

Just as an update from my side: As I mentioned, my student project has been done for a while and now my access to the system where I set this up has been finally revoked; hence, I won't be able to try anything new and from my perspective this issue can be closed. I didn't close it because other people seem to have encountered similar issues, but feel free to close it when you think it is appropriate.