nordic-dev.net/nixpkgs
nordic-dev.net/nixpkgs
2
0
Fork 0
mirror of https://github.com/NixOS/nixpkgs.git synced 2024-12-04 21:03:15 +00:00
Code Issues Packages Projects Releases Wiki Activity
79aad75234
nixpkgs/pkgs/development/cuda-modules
History
Someone Serge 79a7186f1c cudaPackages: updated convention for gpu/runtime checks 
Runtime tests (derivations asking for a relaxed sandbox) are now
expected at p.gpuCheck, p.gpuChecks.<name>, or at
p.tests.<name>.gpuCheck.
2024-06-26 16:00:29 +00:00
..
cuda cudaPackages: fix an annoying warning caused by alias 2024-06-02 11:58:57 +08:00
cuda-library-samples cuda-modules: use stdenv instead of backendStdenv for *Platform access 2024-04-18 16:31:18 +00:00
cuda-samples cuda-modules: use stdenv instead of backendStdenv for *Platform access 2024-04-18 16:31:18 +00:00
cudatoolkit treewide: Remove the definite article from meta.description 2024-06-09 23:08:46 +02:00
cudnn {pkgs/development/cuda-modules,pkgs/test/cuda,pkgs/top-level/cuda-packages.nix}: reformat all CUDA files with nixfmt-rfc-style 2023-03-01 2024-04-01 01:14:28 +00:00
cutensor cuda-modules: use stdenv instead of backendStdenv for *Platform access 2024-04-18 16:31:18 +00:00
generic-builders cuda-modules: use stdenv instead of backendStdenv for *Platform access 2024-04-18 16:31:18 +00:00
modules treewide: Remove the definite article from meta.description 2024-06-09 23:08:46 +02:00
nccl Merge pull request #301416 from ConnorBaker/feat/cudaPackages-lessons-learned-from-cross-compilation-attempt 2024-04-18 20:10:59 -04:00
nccl-tests {pkgs/development/cuda-modules,pkgs/test/cuda,pkgs/top-level/cuda-packages.nix}: reformat all CUDA files with nixfmt-rfc-style 2023-03-01 2024-04-01 01:14:28 +00:00
saxpy cudaPackages: updated convention for gpu/runtime checks 2024-06-26 16:00:29 +00:00
setup-hooks Merge pull request #301416 from ConnorBaker/feat/cudaPackages-lessons-learned-from-cross-compilation-attempt 2024-04-18 20:10:59 -04:00
tensorrt cuda-modules: use stdenv instead of backendStdenv for *Platform access 2024-04-18 16:31:18 +00:00
aliases.nix cudaPackages: fix an alias typo 2024-06-02 11:56:58 +08:00
backend-stdenv.nix cudaPackages.backendStdenv: prune and sort callPackage args 2024-04-10 14:14:00 +00:00
flags.nix cudaPackages.cudaFlags: drop unused cmakeCudaArchitectures attribute 2024-04-18 16:31:18 +00:00
gpus.nix cudaPackages: move derivations to cuda-modules & support aarch64 2023-12-07 16:45:54 +00:00
nvcc-compatibilities.nix cudaPackages_12_3: init at 12.3.0 2023-12-07 16:45:56 +00:00
README.md Add basic documentation on cuda_compat 2023-12-12 16:02:55 +01:00
write-gpu-python-test.nix cudaPackages: updated convention for gpu/runtime checks 2024-06-26 16:00:29 +00:00

README.md

Cuda modules

Note

This document is meant to help CUDA maintainers understand the structure of the CUDA packages in Nixpkgs. It is not meant to be a user-facing document. For a user-facing document, see the CUDA section of the manual.

The files in this directory are added (in some way) to the cudaPackages package set by cuda-packages.nix.

Top-level files

Top-level nix files are included in the initial creation of the cudaPackages scope. These are typically required for the creation of the finalized cudaPackages scope:

  • backend-stdenv.nix: Standard environment for CUDA packages.
  • flags.nix: Flags set, or consumed by, NVCC in order to build packages.
  • gpus.nix: A list of supported NVIDIA GPUs.
  • nvcc-compatibilities.nix: NVCC releases and the version range of GCC/Clang they support.

Top-level directories

  • cuda: CUDA redistributables! Provides extension to cudaPackages scope.
  • cudatoolkit: monolothic CUDA Toolkit run-file installer. Provides extension to cudaPackages scope.
  • cudnn: NVIDIA cuDNN library.
  • cutensor: NVIDIA cuTENSOR library.
  • generic-builders:
    • Contains a builder manifest.nix which operates on the Manifest type defined in modules/generic/manifests. Most packages are built using this builder.
    • Contains a builder multiplex.nix which leverages the Manifest builder. In short, the Multiplex builder adds multiple versions of a single package to single instance of the CUDA Packages package set. It is used primarily for packages like cudnn and cutensor.
  • modules: Nixpkgs modules to check the shape and content of CUDA redistributable and feature manifests. These modules additionally use shims provided by some CUDA packages to allow them to re-use the genericManifestBuilder, even if they don't have manifest files of their own. cudnn and tensorrt are examples of packages which provide such shims. These modules are further described in the Modules documentation.
  • nccl: NVIDIA NCCL library.
  • nccl-tests: NVIDIA NCCL tests.
  • saxpy: Example CMake project that uses CUDA.
  • setup-hooks: Nixpkgs setup hooks for CUDA.
  • tensorrt: NVIDIA TensorRT library.

Distinguished packages

Cuda compatibility

Cuda Compatibility, available as cudaPackages.cuda_compat, is a component which makes it possible to run applications built against a newer CUDA toolkit (for example CUDA 12) on a machine with an older CUDA driver (for example CUDA 11), which isn't possible out of the box. At the time of writing, Cuda Compatibility is only available on the Nvidia Jetson architecture, but Nvidia might release support for more architectures in the future.

As Cuda Compatibility strictly increases the range of supported applications, we try our best to enable it by default on supported platforms.

Functioning

cuda_compat simply provides a new libcuda.so (and associated variants) that needs to be used in place of the default CUDA driver's libcuda.so. However, the other shared libraries of the default driver must still be accessible: cuda_compat isn't a complete drop-in replacement for the driver (and that's the point, otherwise, it would just be a newer driver).

Nvidia's recommendation is to set LD_LIBRARY_PATH to points to cuda_compat's driver. This is fine for a manual, one-shot usage, but in general setting LD_LIBRARY_PATH is a red flag. This is global state which short-circuits most of other dynamic libraries resolution mechanisms and can break things in non-obvious ways, especially with other Nix-built software.

Cuda compat with Nix

Since cuda_compat is a known derivation, the easy way to do this in Nix would be to add cuda_compat as a dependency of CUDA libraries and applications and let Nix does its magic by filling the DT_RUNPATH fields. However, cuda_compat itself depends on libnvrm_mem and libnvrm_gpu which are loaded dynamically at runtime from /run/opengl-driver. This doesn't please the Nix sandbox when building, which can't find those (a second minor issue is that addOpenGLRunpathHook prepends the /run/opengl-driver path, so that would still take precedence).

The current solution is to do something similar to addOpenGLRunpathHook: the addCudaCompatRunpathHook prepends to the path to cuda_compat's libcuda.so to the DT_RUNPATH of whichever package includes the hook as a dependency, and we include the hook by default for packages in cudaPackages (by adding it as a inputs in genericManifestBuilder). We also make sure it's included after addOpenGLRunpathHook, so that it appears before in the DT_RUNPATH and takes precedence.