cloudhan

The root cause is https://github.com/garymm/rules_cuda/blob/d66063672a7e3ea397f40a841b2411e0e4d1ef11/examples/WORKSPACE.bazel#LL33C11-L33C11 in your code. And this file screwed everything https://github.com/NVIDIA/thrust/blob/main/thrust/limits.h. You owe me a cup of coffee for debugging this. 😈

Because there is `/include/thrust`. The system include path of `/include` is transitively added due to deps chain `@thrust -> @cub -> @rules_cuda//:cuda -> @rules_cuda//:cuda_headers`

Reasonable. Would you like to contribute a PR for this? All you need to change is exclude thrust and cub from the glob and add two seperate libraries for cub...

OK, this will be tricky because the cuda toolkit include will be symlinked by bazel into the sandbox. Then all sub directory will be accessible...

I think this should be split into mulitple step. 1. support composing multiple components (say, local_cccl, local_cublas, local_thrust, local_cub) into a unified local_cuda - this might allow reusing pip and...

local_cuda is a name inherited from tf_runtime impl, this should have been called local_cuda_toolkit, so every components stated in the doc will have a position (maybe overrideable). The last step...

Sorry for the delayed reply. I am not sure what you (or presumably for new users) are confusing about, could you please elaborate a bit on why that is important....

> there are 3 types of code You might misunderstanding it a little bit. There is actually 2 languages involved, cuda and cpp, and cuda is a superset of cpp,...

This is from https://github.com/bazel-contrib/rules_cuda/blob/152bbbc9a8680d18b38ca4fba2f9bdb09ee68862/cuda/BUILD#L91-L103, this repo originally is a subtree of tf_runtime and then a full rewrite is carried out starts (with an independent source tree root).

I don't know the reason behind it actually, so let's keep it as is for now. Maybe @chsigg can explain it a bit?