-
Notifications
You must be signed in to change notification settings - Fork 8
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Enable cuda #14
base: main
Are you sure you want to change the base?
Enable cuda #14
Conversation
Signed-off-by: Xinyu Wang <xinyu.wang@tier4.jp>
I'm not sure why Josh removed it in ee82c84. At the same time this patch is for the ROS1 version of the package, which to my knowledge is only used in a test non-default config of a single package of Autoware.ai. And no one is complaining, so probably not used.
Who/what is going to compile this package with the CUDA enablement? The CI? The user? |
The user, since this package provides the runtime lib. |
Then I am confused on how it would all come together. Does your comment here autowarefoundation/modelzoo#85 (comment) about not having tvm_vendor in universe still stand? Would the user be expected to check out this repository manually?
I don't see any problem with merging this PR now. Compiling the models for CUDA is independent of the runtime functionality provided by this package. And if the CUDA runtime doesn't work on some models, it will most likely be TVM that has to be fixed, not tvm_vendor. |
Sorry for the confusion. I think the way how it works currently is that
Yes, my comment still stands. After merging this package, we will make a release to rosdistro. Then the package would be available after image built.
No, user does not need to checkout this repo. ROS would do it during setup. But if anyone wants to use it before the rosdistro release, he/she has to build this repo manually. |
Precisely. I am not sure how the release process works, but I assume that it is automated, and that this process uses an environment that doesn't have cuda available.
Because of the tvm runtime shared object:
|
I think the problem occurs here because the runtime was compiled with cuda at the first place. But if you check the |
I am not familiar with this. But if you are confident that it handles my concerns, then great. |
@wep21 Do you have any comment/concern about this pr? Thanks. |
@angry-crab I have no concern about this pr, but will you add this package into autoware repos because ros buildfarm environment does not have cuda? |
I thought that when we call |
Yes, I think so. |
I guess then we don't have to make a rosdistro release at this time. |
Resolves Enable cuda
BTW, I notice that
USE_LLVM
is set to OFF inconfig.cmake.patch
. Is that intended?