This repository contains pure Starlark implementation of CUDA rules. These rules provide some macros and rules that make it easier to build CUDA with Bazel.

Paste the following snippet into your WORKSPACE file and replace the placeholders to actual values.

load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")

http_archive(
    name = "rules_cuda",
    sha256 = "{sha256_to_replace}",
    strip_prefix = "rules_cuda-{git_commit_hash}",
    urls = ["https://github.com/cloudhan/rules_cuda/archive/{git_commit_hash}.tar.gz"],
)

load("@rules_cuda//cuda:deps.bzl", "register_detected_cuda_toolchains", "rules_cuda_deps")

rules_cuda_deps()

register_detected_cuda_toolchains()

NOTE: the use of register_detected_cuda_toolchains depends on the environment variable CUDA_PATH. You must also ensure the host compiler is available. On windows, this means that you will also need to set the environment variable BAZEL_VC properly.

detect_cuda_toolkit and detect_clang determains how the toolchains are detected.

• cuda_library: Can be used to compile and create static library for CUDA kernel code. The resulting targets can be consumed by C/C++ Rules.
• cuda_objects: If you don't understand what device link means, you must never use it. This rule produce incomplete object files that can only be consumed by cuda_library. It is created for relocatable device code and device link time optimization source files.

Some flags are defined in cuda/BUILD.bazel. To use them, for example:

bazel build --@rules_cuda//cuda:archs=compute_61:compute_61,sm_61

In .bazelrc file, you can define shortcut alias for the flag, for example:

# Convenient flag shortcuts.
build --flag_alias=cuda_archs=@rules_cuda//cuda:archs

and then you can use it as following:

bazel build --cuda_archs=compute_61:compute_61,sm_61

• @rules_cuda//cuda:archs

  Select the cuda archs to support. See cuda_archs specification DSL grammar.

• @rules_cuda//cuda:compiler

  Select the cuda compiler, available options are nvcc or clang

• @rules_cuda//cuda:copts

  Add the copts to all cuda compile actions.

• @rules_cuda//cuda:runtime

  Set the default cudart to link, for example, --@rules_cuda//cuda:runtime=@local_cuda//:cuda_runtime_static link the static cuda runtime.

See examples for basic usage.

See this examples repo for extended real world usage.

Sometimes the following error occurs:

cc1plus: fatal error: /tmp/tmpxft_00000002_00000019-2.cpp: No such file or directory

The problem is caused by nvcc use PID to determine temporary file name, and with --spawn_strategy linux-sandbox which is the default strategy on Linux, the PIDs nvcc sees are all very small numbers, say 2~4 due to sanboxing. linux-sandbox is not hermetic because it mount root into the sandbox, thus, /tmp is shared between sandboxes, which is causing name conflict under high parallelism. Similar problem has been reported at nvidia forums.

To avoid it:

• Use --spawn_strategy local should eliminate the case because it will let nvcc sees the true PIDs.
• Use --experimental_use_hermetic_linux_sandbox should eliminate the case because it will avoid the sharing of /tmp.
• Add -objtemp option to the command should reduce the case from happening.