Add setup instructions and scripts #5
Conversation
charleshofer
left a comment
There was a problem hiding this comment.
Looks pretty good. Mostly minor cleanup stuff.
|
|
||
| # install a rocm | ||
| info "Installing ROCm" | ||
| python build/tools/get_rocm.py --rocm-version 6.4.0 || die "error while installing rocm" |
There was a problem hiding this comment.
Probably fine to leave this hardcoded for now, but eventually we should figure out a way to make the ROCm version and the Python version configurable from the stacks.py script. Maybe we want to stick the supported ROCm versions in a file somewhere and make that declarative, and limit users to one of those options?
There was a problem hiding this comment.
Yeah I expect there will be a lot of changes here but I wanted to keep is simple for now.
I was experimenting with how to run this inside the docker start/run with entrypoint and that kind of works but not really, and I wasn't sure how to pass the version info in yet.
It should be easy to pass all this as parameters when we need to.
| } | ||
|
|
||
| install_clang() { | ||
| [[ -e /usr/lib/llvm-18/bin/clang ]] && return |
There was a problem hiding this comment.
Do we want to check for a clang in the PATH as well?
There was a problem hiding this comment.
Actually this clang doesn't seem to get added to PATH from my testing which made me do this instead.
Currently this is a fast path catch to see if we already installed clang-18 from the block we are entering and skips it. I expect it will change when we change the clang install below, which we probably will because using a packaged version would be faster. I just didn't want to solve that that problem yet.
|
|
||
|
|
||
| XLA_DEV_BRANCH = "rocm-jaxlib-v0.5.0" | ||
| MAKE_TEMPLATE = r""" |
There was a problem hiding this comment.
What's the reason for sticking the template here rather than leaving the file where it was and setting the clang path via environment variables?
There was a problem hiding this comment.
Oh yeah, in my head I was going to be injecting a lot more so I figured it would be good to just template it out for now.
Using env var injection isn't a bad idea either. I'd like to be able to detect the gfxNNN for the current system and inject that too which could easily be done with env var as well. I'll think about it, but for now I just wanted to keep it simple.
|
The build docker script compiles all of LLVM, can we avoid that ? |
Updated README on how to set up environment and also added the scripts it refers to.
Updated README on how to set up environment and also added the scripts it refers to.