[PATCH] D34197: Added Dockerfiles to build clang from sources.

[Justin Lebar via Phabricator via llvm-commits]

jlebar added inline comments.


================
Comment at: docs/Docker.rst:2
+=========================================
+A guide to Dockerfiles for building clang
+=========================================
----------------
ilya-biryukov wrote:
> jlebar wrote:
> > (I got cc'ed because you used the word "cuda".  :)
> > 
> > One thing that would be helpful to me, as someone coming in with zero context, is: What is the difference between the two docker images?  I have a vague idea from the name, but does this mean that I can't use the build generated via debian8 for cuda?  Or, do I have to run the debian8 build *on* debian8?  It would be helpful to me to have answers to those questions in the document.
> You can use clang inside debian8 image to compile for cuda, but you won't have access to the GPU if you run the image you've built in a container, i.e. you can build the cuda binary, but you can't really run it on the GPU.
> Nvidia has their own wrapper around docker for running containers that do have access to the GPU. And you should also use their base images for that to work.
> 
> Added the docs explaining which image one would want to choose.
The explanation is helpful, thank you!


https://reviews.llvm.org/D34197